1 /* SPDX-License-Identifier: GPL-2.0-only */
4 * Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
6 * Derived from book3s_rmhandlers.S and other files, which are:
8 * Copyright SUSE Linux Products GmbH 2009
10 * Authors: Alexander Graf <agraf@suse.de>
13 #include <asm/ppc_asm.h>
14 #include <asm/kvm_asm.h>
18 #include <asm/ptrace.h>
19 #include <asm/hvcall.h>
20 #include <asm/asm-offsets.h>
21 #include <asm/exception-64s.h>
22 #include <asm/kvm_book3s_asm.h>
23 #include <asm/book3s/64/mmu-hash.h>
24 #include <asm/export.h>
27 #include <asm/xive-regs.h>
28 #include <asm/thread_info.h>
29 #include <asm/asm-compat.h>
30 #include <asm/feature-fixups.h>
31 #include <asm/cpuidle.h>
32 #include <asm/ultravisor-api.h>
34 /* Sign-extend HDEC if not on POWER9 */
35 #define EXTEND_HDEC(reg) \
38 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
40 /* Values in HSTATE_NAPPING(r13) */
41 #define NAPPING_CEDE 1
42 #define NAPPING_NOVCPU 2
43 #define NAPPING_UNSPLIT 3
45 /* Stack frame offsets for kvmppc_hv_entry */
47 #define STACK_SLOT_TRAP (SFS-4)
48 #define STACK_SLOT_SHORT_PATH (SFS-8)
49 #define STACK_SLOT_TID (SFS-16)
50 #define STACK_SLOT_PSSCR (SFS-24)
51 #define STACK_SLOT_PID (SFS-32)
52 #define STACK_SLOT_IAMR (SFS-40)
53 #define STACK_SLOT_CIABR (SFS-48)
54 #define STACK_SLOT_DAWR (SFS-56)
55 #define STACK_SLOT_DAWRX (SFS-64)
56 #define STACK_SLOT_HFSCR (SFS-72)
57 #define STACK_SLOT_AMR (SFS-80)
58 #define STACK_SLOT_UAMOR (SFS-88)
59 /* the following is used by the P9 short path */
60 #define STACK_SLOT_NVGPRS (SFS-152) /* 18 gprs */
63 * Call kvmppc_hv_entry in real mode.
64 * Must be called with interrupts hard-disabled.
68 * LR = return address to continue at after eventually re-enabling MMU
70 _GLOBAL_TOC(kvmppc_hv_entry_trampoline)
72 std r0, PPC_LR_STKOFF(r1)
75 std r10, HSTATE_HOST_MSR(r13)
76 LOAD_REG_ADDR(r5, kvmppc_call_hv_entry)
81 mtmsrd r0,1 /* clear RI in MSR */
88 /* On P9, do LPCR setting, if necessary */
89 ld r3, HSTATE_SPLIT_MODE(r13)
92 lwz r4, KVM_SPLIT_DO_SET(r3)
98 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
100 ld r4, HSTATE_KVM_VCPU(r13)
103 /* Back from guest - restore host state and return to caller */
106 /* Restore host DABR and DABRX */
107 ld r5,HSTATE_DABR(r13)
111 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
114 ld r3,PACA_SPRG_VDSO(r13)
115 mtspr SPRN_SPRG_VDSO_WRITE,r3
117 /* Reload the host's PMU registers */
118 bl kvmhv_load_host_pmu
121 * Reload DEC. HDEC interrupts were disabled when
122 * we reloaded the host's LPCR value.
124 ld r3, HSTATE_DECEXP(r13)
129 /* hwthread_req may have got set by cede or no vcpu, so clear it */
131 stb r0, HSTATE_HWTHREAD_REQ(r13)
134 * For external interrupts we need to call the Linux
135 * handler to process the interrupt. We do that by jumping
136 * to absolute address 0x500 for external interrupts.
137 * The [h]rfid at the end of the handler will return to
138 * the book3s_hv_interrupts.S code. For other interrupts
139 * we do the rfid to get back to the book3s_hv_interrupts.S
142 ld r8, 112+PPC_LR_STKOFF(r1)
144 ld r7, HSTATE_HOST_MSR(r13)
146 /* Return the trap number on this thread as the return value */
150 * If we came back from the guest via a relocation-on interrupt,
151 * we will be in virtual mode at this point, which makes it a
152 * little easier to get back to the caller.
155 andi. r0, r0, MSR_IR /* in real mode? */
158 /* RFI into the highmem handler */
162 mtmsrd r6, 1 /* Clear RI in MSR */
167 /* Virtual-mode return */
172 kvmppc_primary_no_guest:
173 /* We handle this much like a ceded vcpu */
174 /* put the HDEC into the DEC, since HDEC interrupts don't wake us */
175 /* HDEC may be larger than DEC for arch >= v3.00, but since the */
176 /* HDEC value came from DEC in the first place, it will fit */
180 * Make sure the primary has finished the MMU switch.
181 * We should never get here on a secondary thread, but
182 * check it for robustness' sake.
184 ld r5, HSTATE_KVM_VCORE(r13)
185 65: lbz r0, VCORE_IN_GUEST(r5)
192 /* set our bit in napping_threads */
193 ld r5, HSTATE_KVM_VCORE(r13)
194 lbz r7, HSTATE_PTID(r13)
197 addi r6, r5, VCORE_NAPPING_THREADS
202 /* order napping_threads update vs testing entry_exit_map */
205 lwz r7, VCORE_ENTRY_EXIT(r5)
207 bge kvm_novcpu_exit /* another thread already exiting */
208 li r3, NAPPING_NOVCPU
209 stb r3, HSTATE_NAPPING(r13)
211 li r3, 0 /* Don't wake on privileged (OS) doorbell */
216 * Entered from kvm_start_guest if kvm_hstate.napping is set
222 ld r1, HSTATE_HOST_R1(r13)
223 ld r5, HSTATE_KVM_VCORE(r13)
225 stb r0, HSTATE_NAPPING(r13)
227 /* check the wake reason */
228 bl kvmppc_check_wake_reason
231 * Restore volatile registers since we could have called
232 * a C routine in kvmppc_check_wake_reason.
235 ld r5, HSTATE_KVM_VCORE(r13)
237 /* see if any other thread is already exiting */
238 lwz r0, VCORE_ENTRY_EXIT(r5)
242 /* clear our bit in napping_threads */
243 lbz r7, HSTATE_PTID(r13)
246 addi r6, r5, VCORE_NAPPING_THREADS
252 /* See if the wake reason means we need to exit */
256 /* See if our timeslice has expired (HDEC is negative) */
259 li r12, BOOK3S_INTERRUPT_HV_DECREMENTER
263 /* Got an IPI but other vcpus aren't yet exiting, must be a latecomer */
264 ld r4, HSTATE_KVM_VCPU(r13)
266 beq kvmppc_primary_no_guest
268 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
269 addi r3, r4, VCPU_TB_RMENTRY
270 bl kvmhv_start_timing
275 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
276 ld r4, HSTATE_KVM_VCPU(r13)
279 addi r3, r4, VCPU_TB_RMEXIT
280 bl kvmhv_accumulate_time
283 stw r12, STACK_SLOT_TRAP(r1)
284 bl kvmhv_commence_exit
286 b kvmhv_switch_to_host
289 * We come in here when wakened from Linux offline idle code.
291 * r3 contains the SRR1 wakeup value, SRR1 is trashed.
293 _GLOBAL(idle_kvm_start_guest)
294 ld r4,PACAEMERGSP(r13)
300 subi r1,r4,STACK_FRAME_OVERHEAD
304 * Could avoid this and pass it through in r3. For now,
305 * code expects it to be in SRR1.
310 stb r0,PACA_FTRACE_ENABLED(r13)
312 li r0,KVM_HWTHREAD_IN_KVM
313 stb r0,HSTATE_HWTHREAD_STATE(r13)
315 /* kvm cede / napping does not come through here */
316 lbz r0,HSTATE_NAPPING(r13)
323 stb r0, HSTATE_NAPPING(r13)
328 * We weren't napping due to cede, so this must be a secondary
329 * thread being woken up to run a guest, or being woken up due
330 * to a stray IPI. (Or due to some machine check or hypervisor
331 * maintenance interrupt while the core is in KVM.)
334 /* Check the wake reason in SRR1 to see why we got here */
335 bl kvmppc_check_wake_reason
337 * kvmppc_check_wake_reason could invoke a C routine, but we
338 * have no volatile registers to restore when we return.
344 /* get vcore pointer, NULL if we have nothing to run */
345 ld r5,HSTATE_KVM_VCORE(r13)
347 /* if we have no vcore to run, go back to sleep */
350 kvm_secondary_got_guest:
352 /* Set HSTATE_DSCR(r13) to something sensible */
353 ld r6, PACA_DSCR_DEFAULT(r13)
354 std r6, HSTATE_DSCR(r13)
356 /* On thread 0 of a subcore, set HDEC to max */
357 lbz r4, HSTATE_PTID(r13)
360 LOAD_REG_ADDR(r6, decrementer_max)
363 /* and set per-LPAR registers, if doing dynamic micro-threading */
364 ld r6, HSTATE_SPLIT_MODE(r13)
368 ld r0, KVM_SPLIT_RPR(r6)
370 ld r0, KVM_SPLIT_PMMAR(r6)
372 ld r0, KVM_SPLIT_LDBAR(r6)
376 /* On P9 we use the split_info for coordinating LPCR changes */
377 lwz r4, KVM_SPLIT_DO_SET(r6)
384 ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
386 /* Order load of vcpu after load of vcore */
388 ld r4, HSTATE_KVM_VCPU(r13)
391 /* Back from the guest, go back to nap */
392 /* Clear our vcpu and vcore pointers so we don't come back in early */
394 std r0, HSTATE_KVM_VCPU(r13)
396 * Once we clear HSTATE_KVM_VCORE(r13), the code in
397 * kvmppc_run_core() is going to assume that all our vcpu
398 * state is visible in memory. This lwsync makes sure
402 std r0, HSTATE_KVM_VCORE(r13)
405 * All secondaries exiting guest will fall through this path.
406 * Before proceeding, just check for HMI interrupt and
407 * invoke opal hmi handler. By now we are sure that the
408 * primary thread on this core/subcore has already made partition
409 * switch/TB resync and we are good to call opal hmi handler.
411 cmpwi r12, BOOK3S_INTERRUPT_HMI
414 li r3,0 /* NULL argument */
415 bl hmi_exception_realmode
417 * At this point we have finished executing in the guest.
418 * We need to wait for hwthread_req to become zero, since
419 * we may not turn on the MMU while hwthread_req is non-zero.
420 * While waiting we also need to check if we get given a vcpu to run.
423 lbz r3, HSTATE_HWTHREAD_REQ(r13)
427 li r0, KVM_HWTHREAD_IN_KERNEL
428 stb r0, HSTATE_HWTHREAD_STATE(r13)
429 /* need to recheck hwthread_req after a barrier, to avoid race */
431 lbz r3, HSTATE_HWTHREAD_REQ(r13)
436 * Jump to idle_return_gpr_loss, which returns to the
437 * idle_kvm_start_guest caller.
441 rlwimi r4, r3, 0, LPCR_PECE0 | LPCR_PECE1
443 /* set up r3 for return */
446 addi r1, r1, STACK_FRAME_OVERHEAD
455 ld r5, HSTATE_KVM_VCORE(r13)
458 ld r3, HSTATE_SPLIT_MODE(r13)
461 lwz r0, KVM_SPLIT_DO_SET(r3)
464 lwz r0, KVM_SPLIT_DO_RESTORE(r3)
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 /* Set LPCR, LPIDR etc. on P9 */
488 bl kvmhv_p9_restore_lpcr
493 * Here the primary thread is trying to return the core to
494 * whole-core mode, so we need to nap.
498 * When secondaries are napping in kvm_unsplit_nap() with
499 * hwthread_req = 1, HMI goes ignored even though subcores are
500 * already exited the guest. Hence HMI keeps waking up secondaries
501 * from nap in a loop and secondaries always go back to nap since
502 * no vcore is assigned to them. This makes impossible for primary
503 * thread to get hold of secondary threads resulting into a soft
504 * lockup in KVM path.
506 * Let us check if HMI is pending and handle it before we go to nap.
508 cmpwi r12, BOOK3S_INTERRUPT_HMI
510 li r3, 0 /* NULL argument */
511 bl hmi_exception_realmode
514 * Ensure that secondary doesn't nap when it has
515 * its vcore pointer set.
517 sync /* matches smp_mb() before setting split_info.do_nap */
518 ld r0, HSTATE_KVM_VCORE(r13)
521 /* clear any pending message */
523 lis r6, (PPC_DBELL_SERVER << (63-36))@h
525 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
526 /* Set kvm_split_mode.napped[tid] = 1 */
527 ld r3, HSTATE_SPLIT_MODE(r13)
529 lbz r4, HSTATE_TID(r13)
530 addi r4, r4, KVM_SPLIT_NAPPED
532 /* Check the do_nap flag again after setting napped[] */
534 lbz r0, KVM_SPLIT_DO_NAP(r3)
537 li r3, NAPPING_UNSPLIT
538 stb r3, HSTATE_NAPPING(r13)
539 li r3, (LPCR_PECEDH | LPCR_PECE0) >> 4
541 rlwimi r5, r3, 4, (LPCR_PECEDP | LPCR_PECEDH | LPCR_PECE0 | LPCR_PECE1)
548 /******************************************************************************
552 *****************************************************************************/
554 .global kvmppc_hv_entry
559 * R4 = vcpu pointer (or NULL)
564 * all other volatile GPRS = free
565 * Does not preserve non-volatile GPRs or CR fields
568 std r0, PPC_LR_STKOFF(r1)
571 /* Save R1 in the PACA */
572 std r1, HSTATE_HOST_R1(r13)
574 li r6, KVM_GUEST_MODE_HOST_HV
575 stb r6, HSTATE_IN_GUEST(r13)
577 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
578 /* Store initial timestamp */
581 addi r3, r4, VCPU_TB_RMENTRY
582 bl kvmhv_start_timing
586 ld r5, HSTATE_KVM_VCORE(r13)
587 ld r9, VCORE_KVM(r5) /* pointer to struct kvm */
590 * POWER7/POWER8 host -> guest partition switch code.
591 * We don't have to lock against concurrent tlbies,
592 * but we do have to coordinate across hardware threads.
594 /* Set bit in entry map iff exit map is zero. */
596 lbz r6, HSTATE_PTID(r13)
598 addi r8, r5, VCORE_ENTRY_EXIT
600 cmpwi r3, 0x100 /* any threads starting to exit? */
601 bge secondary_too_late /* if so we're too late to the party */
606 /* Primary thread switches to guest partition. */
613 li r0,LPID_RSVD /* switch to reserved LPID */
616 mtspr SPRN_SDR1,r6 /* switch to partition page table */
617 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
621 /* See if we need to flush the TLB. */
622 mr r3, r9 /* kvm pointer */
623 lhz r4, PACAPACAINDEX(r13) /* physical cpu number */
624 li r5, 0 /* nested vcpu pointer */
625 bl kvmppc_check_need_tlb_flush
627 ld r5, HSTATE_KVM_VCORE(r13)
629 /* Add timebase offset onto timebase */
630 22: ld r8,VCORE_TB_OFFSET(r5)
633 std r8, VCORE_TB_OFFSET_APPL(r5)
634 mftb r6 /* current host timebase */
636 mtspr SPRN_TBU40,r8 /* update upper 40 bits */
637 mftb r7 /* check if lower 24 bits overflowed */
642 addis r8,r8,0x100 /* if so, increment upper 40 bits */
645 /* Load guest PCR value to select appropriate compat mode */
646 37: ld r7, VCORE_PCR(r5)
653 /* DPDES and VTB are shared between threads */
654 ld r8, VCORE_DPDES(r5)
658 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
660 /* Mark the subcore state as inside guest */
661 bl kvmppc_subcore_enter_guest
663 ld r5, HSTATE_KVM_VCORE(r13)
664 ld r4, HSTATE_KVM_VCPU(r13)
666 stb r0,VCORE_IN_GUEST(r5) /* signal secondaries to continue */
668 /* Do we have a guest vcpu to run? */
670 beq kvmppc_primary_no_guest
672 /* Increment yield count if they have a VPA */
676 li r6, LPPACA_YIELDCOUNT
681 stb r6, VCPU_VPA_DIRTY(r4)
684 /* Save purr/spurr */
687 std r5,HSTATE_PURR(r13)
688 std r6,HSTATE_SPURR(r13)
694 /* Save host values of some registers */
699 std r5, STACK_SLOT_TID(r1)
700 std r6, STACK_SLOT_PSSCR(r1)
701 std r7, STACK_SLOT_PID(r1)
703 std r5, STACK_SLOT_HFSCR(r1)
704 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
710 std r5, STACK_SLOT_CIABR(r1)
711 std r6, STACK_SLOT_DAWR(r1)
712 std r7, STACK_SLOT_DAWRX(r1)
713 std r8, STACK_SLOT_IAMR(r1)
714 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
717 std r5, STACK_SLOT_AMR(r1)
719 std r6, STACK_SLOT_UAMOR(r1)
722 /* Set partition DABR */
723 /* Do this before re-enabling PMU to avoid P7 DABR corruption bug */
724 lwz r5,VCPU_DABRX(r4)
729 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
731 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
733 * Branch around the call if both CPU_FTR_TM and
734 * CPU_FTR_P9_TM_HV_ASSIST are off.
738 END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
740 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
744 li r5, 0 /* don't preserve non-vol regs */
745 bl kvmppc_restore_tm_hv
747 ld r4, HSTATE_KVM_VCPU(r13)
751 /* Load guest PMU registers; r4 = vcpu pointer here */
753 bl kvmhv_load_guest_pmu
755 /* Load up FP, VMX and VSX registers */
756 ld r4, HSTATE_KVM_VCPU(r13)
759 ld r14, VCPU_GPR(R14)(r4)
760 ld r15, VCPU_GPR(R15)(r4)
761 ld r16, VCPU_GPR(R16)(r4)
762 ld r17, VCPU_GPR(R17)(r4)
763 ld r18, VCPU_GPR(R18)(r4)
764 ld r19, VCPU_GPR(R19)(r4)
765 ld r20, VCPU_GPR(R20)(r4)
766 ld r21, VCPU_GPR(R21)(r4)
767 ld r22, VCPU_GPR(R22)(r4)
768 ld r23, VCPU_GPR(R23)(r4)
769 ld r24, VCPU_GPR(R24)(r4)
770 ld r25, VCPU_GPR(R25)(r4)
771 ld r26, VCPU_GPR(R26)(r4)
772 ld r27, VCPU_GPR(R27)(r4)
773 ld r28, VCPU_GPR(R28)(r4)
774 ld r29, VCPU_GPR(R29)(r4)
775 ld r30, VCPU_GPR(R30)(r4)
776 ld r31, VCPU_GPR(R31)(r4)
778 /* Switch DSCR to guest value */
783 /* Skip next section on POWER7 */
785 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
786 /* Load up POWER8-specific registers */
788 lwz r6, VCPU_PSPB(r4)
794 * Handle broken DAWR case by not writing it. This means we
795 * can still store the DAWR register for migration.
797 LOAD_REG_ADDR(r5, dawr_force_enable)
802 ld r6, VCPU_DAWRX(r4)
806 ld r7, VCPU_CIABR(r4)
811 ld r8, VCPU_EBBHR(r4)
814 ld r5, VCPU_EBBRR(r4)
815 ld r6, VCPU_BESCR(r4)
816 lwz r7, VCPU_GUEST_PID(r4)
823 /* POWER8-only registers */
824 ld r5, VCPU_TCSCR(r4)
826 ld r7, VCPU_CSIGR(r4)
834 /* POWER9-only registers */
836 ld r6, VCPU_PSSCR(r4)
837 lbz r8, HSTATE_FAKE_SUSPEND(r13)
838 oris r6, r6, PSSCR_EC@h /* This makes stop trap to HV */
839 rldimi r6, r8, PSSCR_FAKE_SUSPEND_LG, 63 - PSSCR_FAKE_SUSPEND_LG
840 ld r7, VCPU_HFSCR(r4)
844 ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
847 ld r5, VCPU_SPRG0(r4)
848 ld r6, VCPU_SPRG1(r4)
849 ld r7, VCPU_SPRG2(r4)
850 ld r8, VCPU_SPRG3(r4)
856 /* Load up DAR and DSISR */
858 lwz r6, VCPU_DSISR(r4)
862 /* Restore AMR and UAMOR, set AMOR to all 1s */
870 /* Restore state of CTRL run bit; assume 1 on entry */
878 /* Secondary threads wait for primary to have done partition switch */
879 ld r5, HSTATE_KVM_VCORE(r13)
880 lbz r6, HSTATE_PTID(r13)
883 lbz r0, VCORE_IN_GUEST(r5)
887 20: lwz r3, VCORE_ENTRY_EXIT(r5)
890 lbz r0, VCORE_IN_GUEST(r5)
901 * Set the decrementer to the guest decrementer.
903 ld r8,VCPU_DEC_EXPIRES(r4)
904 /* r8 is a host timebase value here, convert to guest TB */
905 ld r5,HSTATE_KVM_VCORE(r13)
906 ld r6,VCORE_TB_OFFSET_APPL(r5)
912 /* Check if HDEC expires soon */
915 cmpdi r3, 512 /* 1 microsecond */
918 /* For hash guest, clear out and reload the SLB */
920 lbz r0, KVM_RADIX(r6)
928 /* Load up guest SLB entries (N.B. slb_max will be 0 for radix) */
929 lwz r5,VCPU_SLB_MAX(r4)
934 1: ld r8,VCPU_SLB_E(r6)
937 addi r6,r6,VCPU_SLB_SIZE
941 #ifdef CONFIG_KVM_XICS
942 /* We are entering the guest on that thread, push VCPU to XIVE */
943 ld r11, VCPU_XIVE_SAVED_STATE(r4)
945 lwz r8, VCPU_XIVE_CAM_WORD(r4)
948 li r7, TM_QW1_OS + TM_WORD2
950 andi. r0, r0, MSR_DR /* in real mode? */
952 ld r10, HSTATE_XIVE_TIMA_VIRT(r13)
959 2: ld r10, HSTATE_XIVE_TIMA_PHYS(r13)
966 stb r9, VCPU_XIVE_PUSHED(r4)
970 * We clear the irq_pending flag. There is a small chance of a
971 * race vs. the escalation interrupt happening on another
972 * processor setting it again, but the only consequence is to
973 * cause a spurrious wakeup on the next H_CEDE which is not an
977 stb r0, VCPU_IRQ_PENDING(r4)
980 * In single escalation mode, if the escalation interrupt is
983 lbz r0, VCPU_XIVE_ESC_ON(r4)
986 li r9, XIVE_ESB_SET_PQ_01
987 beq 4f /* in real mode? */
988 ld r10, VCPU_XIVE_ESC_VADDR(r4)
991 4: ld r10, VCPU_XIVE_ESC_RADDR(r4)
995 /* We have a possible subtle race here: The escalation interrupt might
996 * have fired and be on its way to the host queue while we mask it,
997 * and if we unmask it early enough (re-cede right away), there is
998 * a theorical possibility that it fires again, thus landing in the
999 * target queue more than once which is a big no-no.
1001 * Fortunately, solving this is rather easy. If the above load setting
1002 * PQ to 01 returns a previous value where P is set, then we know the
1003 * escalation interrupt is somewhere on its way to the host. In that
1004 * case we simply don't clear the xive_esc_on flag below. It will be
1005 * eventually cleared by the handler for the escalation interrupt.
1007 * Then, when doing a cede, we check that flag again before re-enabling
1008 * the escalation interrupt, and if set, we abort the cede.
1010 andi. r0, r0, XIVE_ESB_VAL_P
1013 /* Now P is 0, we can clear the flag */
1015 stb r0, VCPU_XIVE_ESC_ON(r4)
1018 #endif /* CONFIG_KVM_XICS */
1021 stw r0, STACK_SLOT_SHORT_PATH(r1)
1023 deliver_guest_interrupt: /* r4 = vcpu, r13 = paca */
1024 /* Check if we can deliver an external or decrementer interrupt now */
1025 ld r0, VCPU_PENDING_EXC(r4)
1027 /* On POWER9, also check for emulated doorbell interrupt */
1028 lbz r3, VCPU_DBELL_REQ(r4)
1030 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1034 bl kvmppc_guest_entry_inject_int
1035 ld r4, HSTATE_KVM_VCPU(r13)
1037 ld r6, VCPU_SRR0(r4)
1038 ld r7, VCPU_SRR1(r4)
1044 ld r11, VCPU_MSR(r4)
1045 /* r11 = vcpu->arch.msr & ~MSR_HV */
1046 rldicl r11, r11, 63 - MSR_HV_LG, 1
1047 rotldi r11, r11, 1 + MSR_HV_LG
1048 ori r11, r11, MSR_ME
1058 * R10: value for HSRR0
1059 * R11: value for HSRR1
1064 stb r0,VCPU_CEDED(r4) /* cancel cede */
1065 mtspr SPRN_HSRR0,r10
1066 mtspr SPRN_HSRR1,r11
1068 /* Activate guest mode, so faults get handled by KVM */
1069 li r9, KVM_GUEST_MODE_GUEST_HV
1070 stb r9, HSTATE_IN_GUEST(r13)
1072 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1073 /* Accumulate timing */
1074 addi r3, r4, VCPU_TB_GUEST
1075 bl kvmhv_accumulate_time
1081 ld r5, VCPU_CFAR(r4)
1083 END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
1086 END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1091 ld r1, VCPU_GPR(R1)(r4)
1092 ld r5, VCPU_GPR(R5)(r4)
1093 ld r8, VCPU_GPR(R8)(r4)
1094 ld r9, VCPU_GPR(R9)(r4)
1095 ld r10, VCPU_GPR(R10)(r4)
1096 ld r11, VCPU_GPR(R11)(r4)
1097 ld r12, VCPU_GPR(R12)(r4)
1098 ld r13, VCPU_GPR(R13)(r4)
1102 END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1104 /* Move canary into DSISR to check for later */
1107 mtspr SPRN_HDSISR, r0
1108 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1111 lbz r7, KVM_SECURE_GUEST(r6)
1113 ld r6, VCPU_GPR(R6)(r4)
1114 ld r7, VCPU_GPR(R7)(r4)
1120 ld r0, VCPU_GPR(R0)(r4)
1121 ld r2, VCPU_GPR(R2)(r4)
1122 ld r3, VCPU_GPR(R3)(r4)
1123 ld r4, VCPU_GPR(R4)(r4)
1127 * Use UV_RETURN ultracall to return control back to the Ultravisor after
1128 * processing an hypercall or interrupt that was forwarded (a.k.a. reflected)
1129 * to the Hypervisor.
1131 * All registers have already been loaded, except:
1133 * R2 = SRR1, so UV can detect a synthesized interrupt (if any)
1140 ld r0, VCPU_GPR(R3)(r4)
1143 ori r3, r3, UV_RETURN
1144 ld r4, VCPU_GPR(R4)(r4)
1148 * Enter the guest on a P9 or later system where we have exactly
1149 * one vcpu per vcore and we don't need to go to real mode
1150 * (which implies that host and guest are both using radix MMU mode).
1152 * Most SPRs and all the VSRs have been loaded already.
1154 _GLOBAL(__kvmhv_vcpu_entry_p9)
1155 EXPORT_SYMBOL_GPL(__kvmhv_vcpu_entry_p9)
1157 std r0, PPC_LR_STKOFF(r1)
1161 stw r0, STACK_SLOT_SHORT_PATH(r1)
1163 std r3, HSTATE_KVM_VCPU(r13)
1167 std r1, HSTATE_HOST_R1(r13)
1171 std reg, STACK_SLOT_NVGPRS + ((reg - 14) * 8)(r1)
1177 ld reg, __VCPU_GPR(reg)(r3)
1182 std r10, HSTATE_HOST_MSR(r13)
1185 b fast_guest_entry_c
1186 guest_exit_short_path:
1188 li r0, KVM_GUEST_MODE_NONE
1189 stb r0, HSTATE_IN_GUEST(r13)
1193 std reg, __VCPU_GPR(reg)(r9)
1199 ld reg, STACK_SLOT_NVGPRS + ((reg - 14) * 8)(r1)
1206 mr r3, r12 /* trap number */
1209 ld r0, PPC_LR_STKOFF(r1)
1212 /* If we are in real mode, do a rfid to get back to the caller */
1214 andi. r5, r4, MSR_IR
1216 rldicl r5, r4, 64 - MSR_TS_S_LG, 62 /* extract TS field */
1218 ld r10, HSTATE_HOST_MSR(r13)
1219 rldimi r10, r5, MSR_TS_S_LG, 63 - MSR_TS_T_LG
1220 mtspr SPRN_SRR1, r10
1226 stw r12, STACK_SLOT_TRAP(r1)
1229 stw r12, VCPU_TRAP(r4)
1230 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1231 addi r3, r4, VCPU_TB_RMEXIT
1232 bl kvmhv_accumulate_time
1234 11: b kvmhv_switch_to_host
1241 li r12, BOOK3S_INTERRUPT_HV_DECREMENTER
1242 12: stw r12, VCPU_TRAP(r4)
1244 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1245 addi r3, r4, VCPU_TB_RMEXIT
1246 bl kvmhv_accumulate_time
1250 /******************************************************************************
1254 *****************************************************************************/
1257 * We come here from the first-level interrupt handlers.
1259 .globl kvmppc_interrupt_hv
1260 kvmppc_interrupt_hv:
1262 * Register contents:
1263 * R12 = (guest CR << 32) | interrupt vector
1265 * guest R12 saved in shadow VCPU SCRATCH0
1266 * guest CTR saved in shadow VCPU SCRATCH1 if RELOCATABLE
1267 * guest R13 saved in SPRN_SCRATCH0
1269 std r9, HSTATE_SCRATCH2(r13)
1270 lbz r9, HSTATE_IN_GUEST(r13)
1271 cmpwi r9, KVM_GUEST_MODE_HOST_HV
1272 beq kvmppc_bad_host_intr
1273 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1274 cmpwi r9, KVM_GUEST_MODE_GUEST
1275 ld r9, HSTATE_SCRATCH2(r13)
1276 beq kvmppc_interrupt_pr
1278 /* We're now back in the host but in guest MMU context */
1279 li r9, KVM_GUEST_MODE_HOST_HV
1280 stb r9, HSTATE_IN_GUEST(r13)
1282 ld r9, HSTATE_KVM_VCPU(r13)
1284 /* Save registers */
1286 std r0, VCPU_GPR(R0)(r9)
1287 std r1, VCPU_GPR(R1)(r9)
1288 std r2, VCPU_GPR(R2)(r9)
1289 std r3, VCPU_GPR(R3)(r9)
1290 std r4, VCPU_GPR(R4)(r9)
1291 std r5, VCPU_GPR(R5)(r9)
1292 std r6, VCPU_GPR(R6)(r9)
1293 std r7, VCPU_GPR(R7)(r9)
1294 std r8, VCPU_GPR(R8)(r9)
1295 ld r0, HSTATE_SCRATCH2(r13)
1296 std r0, VCPU_GPR(R9)(r9)
1297 std r10, VCPU_GPR(R10)(r9)
1298 std r11, VCPU_GPR(R11)(r9)
1299 ld r3, HSTATE_SCRATCH0(r13)
1300 std r3, VCPU_GPR(R12)(r9)
1301 /* CR is in the high half of r12 */
1305 ld r3, HSTATE_CFAR(r13)
1306 std r3, VCPU_CFAR(r9)
1307 END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
1309 ld r4, HSTATE_PPR(r13)
1310 std r4, VCPU_PPR(r9)
1311 END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1313 /* Restore R1/R2 so we can handle faults */
1314 ld r1, HSTATE_HOST_R1(r13)
1317 mfspr r10, SPRN_SRR0
1318 mfspr r11, SPRN_SRR1
1319 std r10, VCPU_SRR0(r9)
1320 std r11, VCPU_SRR1(r9)
1321 /* trap is in the low half of r12, clear CR from the high half */
1323 andi. r0, r12, 2 /* need to read HSRR0/1? */
1325 mfspr r10, SPRN_HSRR0
1326 mfspr r11, SPRN_HSRR1
1328 1: std r10, VCPU_PC(r9)
1329 std r11, VCPU_MSR(r9)
1333 std r3, VCPU_GPR(R13)(r9)
1336 stw r12,VCPU_TRAP(r9)
1339 * Now that we have saved away SRR0/1 and HSRR0/1,
1340 * interrupts are recoverable in principle, so set MSR_RI.
1341 * This becomes important for relocation-on interrupts from
1342 * the guest, which we can get in radix mode on POWER9.
1347 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1348 addi r3, r9, VCPU_TB_RMINTR
1350 bl kvmhv_accumulate_time
1351 ld r5, VCPU_GPR(R5)(r9)
1352 ld r6, VCPU_GPR(R6)(r9)
1353 ld r7, VCPU_GPR(R7)(r9)
1354 ld r8, VCPU_GPR(R8)(r9)
1357 /* Save HEIR (HV emulation assist reg) in emul_inst
1358 if this is an HEI (HV emulation interrupt, e40) */
1359 li r3,KVM_INST_FETCH_FAILED
1360 stw r3,VCPU_LAST_INST(r9)
1361 cmpwi r12,BOOK3S_INTERRUPT_H_EMUL_ASSIST
1364 11: stw r3,VCPU_HEIR(r9)
1366 /* these are volatile across C function calls */
1367 #ifdef CONFIG_RELOCATABLE
1368 ld r3, HSTATE_SCRATCH1(r13)
1374 std r3, VCPU_CTR(r9)
1375 std r4, VCPU_XER(r9)
1377 /* Save more register state */
1380 std r3, VCPU_DAR(r9)
1381 stw r4, VCPU_DSISR(r9)
1383 /* If this is a page table miss then see if it's theirs or ours */
1384 cmpwi r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
1386 std r3, VCPU_FAULT_DAR(r9)
1387 stw r4, VCPU_FAULT_DSISR(r9)
1388 cmpwi r12, BOOK3S_INTERRUPT_H_INST_STORAGE
1391 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1392 /* For softpatch interrupt, go off and do TM instruction emulation */
1393 cmpwi r12, BOOK3S_INTERRUPT_HV_SOFTPATCH
1397 /* See if this is a leftover HDEC interrupt */
1398 cmpwi r12,BOOK3S_INTERRUPT_HV_DECREMENTER
1404 bge fast_guest_return
1406 /* See if this is an hcall we can handle in real mode */
1407 cmpwi r12,BOOK3S_INTERRUPT_SYSCALL
1408 beq hcall_try_real_mode
1410 /* Hypervisor doorbell - exit only if host IPI flag set */
1411 cmpwi r12, BOOK3S_INTERRUPT_H_DOORBELL
1416 /* always exit if we're running a nested guest */
1417 ld r0, VCPU_NESTED(r9)
1420 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1421 lbz r0, HSTATE_HOST_IPI(r13)
1423 beq maybe_reenter_guest
1426 /* If it's a hypervisor facility unavailable interrupt, save HFSCR */
1427 cmpwi r12, BOOK3S_INTERRUPT_H_FAC_UNAVAIL
1429 mfspr r3, SPRN_HFSCR
1430 std r3, VCPU_HFSCR(r9)
1433 /* External interrupt ? */
1434 cmpwi r12, BOOK3S_INTERRUPT_EXTERNAL
1435 beq kvmppc_guest_external
1436 /* See if it is a machine check */
1437 cmpwi r12, BOOK3S_INTERRUPT_MACHINE_CHECK
1438 beq machine_check_realmode
1439 /* Or a hypervisor maintenance interrupt */
1440 cmpwi r12, BOOK3S_INTERRUPT_HMI
1443 guest_exit_cont: /* r9 = vcpu, r12 = trap, r13 = paca */
1445 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1446 addi r3, r9, VCPU_TB_RMEXIT
1448 bl kvmhv_accumulate_time
1450 #ifdef CONFIG_KVM_XICS
1451 /* We are exiting, pull the VP from the XIVE */
1452 lbz r0, VCPU_XIVE_PUSHED(r9)
1455 li r7, TM_SPC_PULL_OS_CTX
1458 andi. r0, r0, MSR_DR /* in real mode? */
1460 ld r10, HSTATE_XIVE_TIMA_VIRT(r13)
1463 /* First load to pull the context, we ignore the value */
1466 /* Second load to recover the context state (Words 0 and 1) */
1469 2: ld r10, HSTATE_XIVE_TIMA_PHYS(r13)
1472 /* First load to pull the context, we ignore the value */
1475 /* Second load to recover the context state (Words 0 and 1) */
1477 3: std r11, VCPU_XIVE_SAVED_STATE(r9)
1478 /* Fixup some of the state for the next load */
1481 stb r10, VCPU_XIVE_PUSHED(r9)
1482 stb r10, (VCPU_XIVE_SAVED_STATE+3)(r9)
1483 stb r0, (VCPU_XIVE_SAVED_STATE+4)(r9)
1486 #endif /* CONFIG_KVM_XICS */
1488 /* If we came in through the P9 short path, go back out to C now */
1489 lwz r0, STACK_SLOT_SHORT_PATH(r1)
1491 bne guest_exit_short_path
1493 /* For hash guest, read the guest SLB and save it away */
1495 lbz r0, KVM_RADIX(r5)
1498 bne 3f /* for radix, save 0 entries */
1499 lwz r0,VCPU_SLB_NR(r9) /* number of entries in SLB */
1504 andis. r0,r8,SLB_ESID_V@h
1506 add r8,r8,r6 /* put index in */
1508 std r8,VCPU_SLB_E(r7)
1509 std r3,VCPU_SLB_V(r7)
1510 addi r7,r7,VCPU_SLB_SIZE
1514 /* Finally clear out the SLB */
1519 3: stw r5,VCPU_SLB_MAX(r9)
1521 /* load host SLB entries */
1522 BEGIN_MMU_FTR_SECTION
1524 END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
1525 ld r8,PACA_SLBSHADOWPTR(r13)
1527 .rept SLB_NUM_BOLTED
1528 li r3, SLBSHADOW_SAVEAREA
1532 andis. r7,r5,SLB_ESID_V@h
1540 stw r12, STACK_SLOT_TRAP(r1)
1543 /* Do this before kvmhv_commence_exit so we know TB is guest TB */
1544 ld r3, HSTATE_KVM_VCORE(r13)
1547 /* On P9, if the guest has large decr enabled, don't sign extend */
1549 ld r4, VCORE_LPCR(r3)
1550 andis. r4, r4, LPCR_LD@h
1552 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1555 /* r5 is a guest timebase value here, convert to host TB */
1556 ld r4,VCORE_TB_OFFSET_APPL(r3)
1558 std r5,VCPU_DEC_EXPIRES(r9)
1560 /* Increment exit count, poke other threads to exit */
1562 bl kvmhv_commence_exit
1564 ld r9, HSTATE_KVM_VCPU(r13)
1566 /* Stop others sending VCPU interrupts to this physical CPU */
1568 stw r0, VCPU_CPU(r9)
1569 stw r0, VCPU_THREAD_CPU(r9)
1571 /* Save guest CTRL register, set runlatch to 1 */
1573 stw r6,VCPU_CTRL(r9)
1580 * Save the guest PURR/SPURR
1585 ld r8,VCPU_SPURR(r9)
1586 std r5,VCPU_PURR(r9)
1587 std r6,VCPU_SPURR(r9)
1592 * Restore host PURR/SPURR and add guest times
1593 * so that the time in the guest gets accounted.
1595 ld r3,HSTATE_PURR(r13)
1596 ld r4,HSTATE_SPURR(r13)
1604 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
1605 /* Save POWER8-specific registers */
1609 std r5, VCPU_IAMR(r9)
1610 stw r6, VCPU_PSPB(r9)
1611 std r7, VCPU_FSCR(r9)
1615 std r7, VCPU_TAR(r9)
1616 mfspr r8, SPRN_EBBHR
1617 std r8, VCPU_EBBHR(r9)
1618 mfspr r5, SPRN_EBBRR
1619 mfspr r6, SPRN_BESCR
1622 std r5, VCPU_EBBRR(r9)
1623 std r6, VCPU_BESCR(r9)
1624 stw r7, VCPU_GUEST_PID(r9)
1625 std r8, VCPU_WORT(r9)
1627 mfspr r5, SPRN_TCSCR
1629 mfspr r7, SPRN_CSIGR
1631 std r5, VCPU_TCSCR(r9)
1632 std r6, VCPU_ACOP(r9)
1633 std r7, VCPU_CSIGR(r9)
1634 std r8, VCPU_TACR(r9)
1637 mfspr r6, SPRN_PSSCR
1638 std r5, VCPU_TID(r9)
1639 rldicl r6, r6, 4, 50 /* r6 &= PSSCR_GUEST_VIS */
1641 std r6, VCPU_PSSCR(r9)
1642 /* Restore host HFSCR value */
1643 ld r7, STACK_SLOT_HFSCR(r1)
1644 mtspr SPRN_HFSCR, r7
1645 ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
1647 * Restore various registers to 0, where non-zero values
1648 * set by the guest could disrupt the host.
1654 mtspr SPRN_TCSCR, r0
1655 /* Set MMCRS to 1<<31 to freeze and disable the SPMC counters */
1658 mtspr SPRN_MMCRS, r0
1659 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
1661 /* Save and restore AMR, IAMR and UAMOR before turning on the MMU */
1662 ld r8, STACK_SLOT_IAMR(r1)
1665 8: /* Power7 jumps back in here */
1669 std r6,VCPU_UAMOR(r9)
1670 ld r5,STACK_SLOT_AMR(r1)
1671 ld r6,STACK_SLOT_UAMOR(r1)
1673 mtspr SPRN_UAMOR, r6
1675 /* Switch DSCR back to host value */
1677 ld r7, HSTATE_DSCR(r13)
1678 std r8, VCPU_DSCR(r9)
1681 /* Save non-volatile GPRs */
1682 std r14, VCPU_GPR(R14)(r9)
1683 std r15, VCPU_GPR(R15)(r9)
1684 std r16, VCPU_GPR(R16)(r9)
1685 std r17, VCPU_GPR(R17)(r9)
1686 std r18, VCPU_GPR(R18)(r9)
1687 std r19, VCPU_GPR(R19)(r9)
1688 std r20, VCPU_GPR(R20)(r9)
1689 std r21, VCPU_GPR(R21)(r9)
1690 std r22, VCPU_GPR(R22)(r9)
1691 std r23, VCPU_GPR(R23)(r9)
1692 std r24, VCPU_GPR(R24)(r9)
1693 std r25, VCPU_GPR(R25)(r9)
1694 std r26, VCPU_GPR(R26)(r9)
1695 std r27, VCPU_GPR(R27)(r9)
1696 std r28, VCPU_GPR(R28)(r9)
1697 std r29, VCPU_GPR(R29)(r9)
1698 std r30, VCPU_GPR(R30)(r9)
1699 std r31, VCPU_GPR(R31)(r9)
1702 mfspr r3, SPRN_SPRG0
1703 mfspr r4, SPRN_SPRG1
1704 mfspr r5, SPRN_SPRG2
1705 mfspr r6, SPRN_SPRG3
1706 std r3, VCPU_SPRG0(r9)
1707 std r4, VCPU_SPRG1(r9)
1708 std r5, VCPU_SPRG2(r9)
1709 std r6, VCPU_SPRG3(r9)
1715 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1717 * Branch around the call if both CPU_FTR_TM and
1718 * CPU_FTR_P9_TM_HV_ASSIST are off.
1722 END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
1724 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
1728 li r5, 0 /* don't preserve non-vol regs */
1729 bl kvmppc_save_tm_hv
1731 ld r9, HSTATE_KVM_VCPU(r13)
1735 /* Increment yield count if they have a VPA */
1736 ld r8, VCPU_VPA(r9) /* do they have a VPA? */
1739 li r4, LPPACA_YIELDCOUNT
1744 stb r3, VCPU_VPA_DIRTY(r9)
1746 /* Save PMU registers if requested */
1747 /* r8 and cr0.eq are live here */
1750 beq 21f /* if no VPA, save PMU stuff anyway */
1751 lbz r4, LPPACA_PMCINUSE(r8)
1752 21: bl kvmhv_save_guest_pmu
1753 ld r9, HSTATE_KVM_VCPU(r13)
1755 /* Restore host values of some registers */
1757 ld r5, STACK_SLOT_CIABR(r1)
1758 ld r6, STACK_SLOT_DAWR(r1)
1759 ld r7, STACK_SLOT_DAWRX(r1)
1760 mtspr SPRN_CIABR, r5
1762 * If the DAWR doesn't work, it's ok to write these here as
1763 * this value should always be zero
1766 mtspr SPRN_DAWRX, r7
1767 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1769 ld r5, STACK_SLOT_TID(r1)
1770 ld r6, STACK_SLOT_PSSCR(r1)
1771 ld r7, STACK_SLOT_PID(r1)
1773 mtspr SPRN_PSSCR, r6
1775 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1777 #ifdef CONFIG_PPC_RADIX_MMU
1779 * Are we running hash or radix ?
1782 lbz r0, KVM_RADIX(r5)
1787 * Radix: do eieio; tlbsync; ptesync sequence in case we
1788 * interrupted the guest between a tlbie and a ptesync.
1794 /* Radix: Handle the case where the guest used an illegal PID */
1795 LOAD_REG_ADDR(r4, mmu_base_pid)
1796 lwz r3, VCPU_GUEST_PID(r9)
1802 * Illegal PID, the HW might have prefetched and cached in the TLB
1803 * some translations for the LPID 0 / guest PID combination which
1804 * Linux doesn't know about, so we need to flush that PID out of
1805 * the TLB. First we need to set LPIDR to 0 so tlbiel applies to
1806 * the right context.
1812 /* Then do a congruence class local flush */
1814 lwz r0,KVM_TLB_SETS(r6)
1816 li r7,0x400 /* IS field = 0b01 */
1818 sldi r0,r3,32 /* RS has PID */
1819 1: PPC_TLBIEL(7,0,2,1,1) /* RIC=2, PRS=1, R=1 */
1825 #endif /* CONFIG_PPC_RADIX_MMU */
1828 * POWER7/POWER8 guest -> host partition switch code.
1829 * We don't have to lock against tlbies but we do
1830 * have to coordinate the hardware threads.
1831 * Here STACK_SLOT_TRAP(r1) contains the trap number.
1833 kvmhv_switch_to_host:
1834 /* Secondary threads wait for primary to do partition switch */
1835 ld r5,HSTATE_KVM_VCORE(r13)
1836 ld r4,VCORE_KVM(r5) /* pointer to struct kvm */
1837 lbz r3,HSTATE_PTID(r13)
1841 13: lbz r3,VCORE_IN_GUEST(r5)
1847 /* Primary thread waits for all the secondaries to exit guest */
1848 15: lwz r3,VCORE_ENTRY_EXIT(r5)
1849 rlwinm r0,r3,32-8,0xff
1855 /* Did we actually switch to the guest at all? */
1856 lbz r6, VCORE_IN_GUEST(r5)
1860 /* Primary thread switches back to host partition */
1861 lwz r7,KVM_HOST_LPID(r4)
1863 ld r6,KVM_HOST_SDR1(r4)
1864 li r8,LPID_RSVD /* switch to reserved LPID */
1867 mtspr SPRN_SDR1,r6 /* switch to host page table */
1868 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
1873 /* DPDES and VTB are shared between threads */
1874 mfspr r7, SPRN_DPDES
1876 std r7, VCORE_DPDES(r5)
1877 std r8, VCORE_VTB(r5)
1878 /* clear DPDES so we don't get guest doorbells in the host */
1880 mtspr SPRN_DPDES, r8
1881 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1883 /* Subtract timebase offset from timebase */
1884 ld r8, VCORE_TB_OFFSET_APPL(r5)
1888 std r0, VCORE_TB_OFFSET_APPL(r5)
1889 mftb r6 /* current guest timebase */
1891 mtspr SPRN_TBU40,r8 /* update upper 40 bits */
1892 mftb r7 /* check if lower 24 bits overflowed */
1897 addis r8,r8,0x100 /* if so, increment upper 40 bits */
1902 * If this is an HMI, we called kvmppc_realmode_hmi_handler
1903 * above, which may or may not have already called
1904 * kvmppc_subcore_exit_guest. Fortunately, all that
1905 * kvmppc_subcore_exit_guest does is clear a flag, so calling
1906 * it again here is benign even if kvmppc_realmode_hmi_handler
1907 * has already called it.
1909 bl kvmppc_subcore_exit_guest
1911 30: ld r5,HSTATE_KVM_VCORE(r13)
1912 ld r4,VCORE_KVM(r5) /* pointer to struct kvm */
1915 ld r0, VCORE_PCR(r5)
1921 /* Signal secondary CPUs to continue */
1922 stb r0,VCORE_IN_GUEST(r5)
1923 19: lis r8,0x7fff /* MAX_INT@h */
1928 /* On POWER9 with HPT-on-radix we need to wait for all other threads */
1929 ld r3, HSTATE_SPLIT_MODE(r13)
1932 lwz r8, KVM_SPLIT_DO_RESTORE(r3)
1935 bl kvmhv_p9_restore_lpcr
1939 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1940 ld r8,KVM_HOST_LPCR(r4)
1944 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1945 /* Finish timing, if we have a vcpu */
1946 ld r4, HSTATE_KVM_VCPU(r13)
1950 bl kvmhv_accumulate_time
1953 /* Unset guest mode */
1954 li r0, KVM_GUEST_MODE_NONE
1955 stb r0, HSTATE_IN_GUEST(r13)
1957 lwz r12, STACK_SLOT_TRAP(r1) /* return trap # in r12 */
1958 ld r0, SFS+PPC_LR_STKOFF(r1)
1963 kvmppc_guest_external:
1964 /* External interrupt, first check for host_ipi. If this is
1965 * set, we know the host wants us out so let's do it now
1970 * Restore the active volatile registers after returning from
1973 ld r9, HSTATE_KVM_VCPU(r13)
1974 li r12, BOOK3S_INTERRUPT_EXTERNAL
1977 * kvmppc_read_intr return codes:
1979 * Exit to host (r3 > 0)
1980 * 1 An interrupt is pending that needs to be handled by the host
1981 * Exit guest and return to host by branching to guest_exit_cont
1983 * 2 Passthrough that needs completion in the host
1984 * Exit guest and return to host by branching to guest_exit_cont
1985 * However, we also set r12 to BOOK3S_INTERRUPT_HV_RM_HARD
1986 * to indicate to the host to complete handling the interrupt
1988 * Before returning to guest, we check if any CPU is heading out
1989 * to the host and if so, we head out also. If no CPUs are heading
1990 * check return values <= 0.
1992 * Return to guest (r3 <= 0)
1993 * 0 No external interrupt is pending
1994 * -1 A guest wakeup IPI (which has now been cleared)
1995 * In either case, we return to guest to deliver any pending
1998 * -2 A PCI passthrough external interrupt was handled
1999 * (interrupt was delivered directly to guest)
2000 * Return to guest to deliver any pending guest interrupts.
2006 /* Return code = 2 */
2007 li r12, BOOK3S_INTERRUPT_HV_RM_HARD
2008 stw r12, VCPU_TRAP(r9)
2011 1: /* Return code <= 1 */
2015 /* Return code <= 0 */
2016 maybe_reenter_guest:
2017 ld r5, HSTATE_KVM_VCORE(r13)
2018 lwz r0, VCORE_ENTRY_EXIT(r5)
2021 blt deliver_guest_interrupt
2024 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2026 * Softpatch interrupt for transactional memory emulation cases
2027 * on POWER9 DD2.2. This is early in the guest exit path - we
2028 * haven't saved registers or done a treclaim yet.
2031 /* Save instruction image in HEIR */
2033 stw r3, VCPU_HEIR(r9)
2036 * The cases we want to handle here are those where the guest
2037 * is in real suspend mode and is trying to transition to
2038 * transactional mode.
2040 lbz r0, HSTATE_FAKE_SUSPEND(r13)
2041 cmpwi r0, 0 /* keep exiting guest if in fake suspend */
2043 rldicl r3, r11, 64 - MSR_TS_S_LG, 62
2044 cmpwi r3, 1 /* or if not in suspend state */
2047 /* Call C code to do the emulation */
2049 bl kvmhv_p9_tm_emulation_early
2051 ld r9, HSTATE_KVM_VCPU(r13)
2052 li r12, BOOK3S_INTERRUPT_HV_SOFTPATCH
2054 beq guest_exit_cont /* continue exiting if not handled */
2056 ld r11, VCPU_MSR(r9)
2057 b fast_interrupt_c_return /* go back to guest if handled */
2058 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
2061 * Check whether an HDSI is an HPTE not found fault or something else.
2062 * If it is an HPTE not found fault that is due to the guest accessing
2063 * a page that they have mapped but which we have paged out, then
2064 * we continue on with the guest exit path. In all other cases,
2065 * reflect the HDSI to the guest as a DSI.
2069 lbz r0, KVM_RADIX(r3)
2071 mfspr r6, SPRN_HDSISR
2073 /* Look for DSISR canary. If we find it, retry instruction */
2076 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2078 bne .Lradix_hdsi /* on radix, just save DAR/DSISR/ASDR */
2079 /* HPTE not found fault or protection fault? */
2080 andis. r0, r6, (DSISR_NOHPTE | DSISR_PROTFAULT)@h
2081 beq 1f /* if not, send it to the guest */
2082 andi. r0, r11, MSR_DR /* data relocation enabled? */
2085 mfspr r5, SPRN_ASDR /* on POWER9, use ASDR to get VSID */
2087 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2089 PPC_SLBFEE_DOT(R5, R0) /* if so, look up SLB */
2090 li r0, BOOK3S_INTERRUPT_DATA_SEGMENT
2091 bne 7f /* if no SLB entry found */
2092 4: std r4, VCPU_FAULT_DAR(r9)
2093 stw r6, VCPU_FAULT_DSISR(r9)
2095 /* Search the hash table. */
2096 mr r3, r9 /* vcpu pointer */
2097 li r7, 1 /* data fault */
2098 bl kvmppc_hpte_hv_fault
2099 ld r9, HSTATE_KVM_VCPU(r13)
2101 ld r11, VCPU_MSR(r9)
2102 li r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
2103 cmpdi r3, 0 /* retry the instruction */
2105 cmpdi r3, -1 /* handle in kernel mode */
2107 cmpdi r3, -2 /* MMIO emulation; need instr word */
2110 /* Synthesize a DSI (or DSegI) for the guest */
2111 ld r4, VCPU_FAULT_DAR(r9)
2113 1: li r0, BOOK3S_INTERRUPT_DATA_STORAGE
2114 mtspr SPRN_DSISR, r6
2115 7: mtspr SPRN_DAR, r4
2116 mtspr SPRN_SRR0, r10
2117 mtspr SPRN_SRR1, r11
2119 bl kvmppc_msr_interrupt
2120 fast_interrupt_c_return:
2121 6: ld r7, VCPU_CTR(r9)
2128 3: ld r5, VCPU_KVM(r9) /* not relocated, use VRMA */
2129 ld r5, KVM_VRMA_SLB_V(r5)
2132 /* If this is for emulated MMIO, load the instruction word */
2133 2: li r8, KVM_INST_FETCH_FAILED /* In case lwz faults */
2135 /* Set guest mode to 'jump over instruction' so if lwz faults
2136 * we'll just continue at the next IP. */
2137 li r0, KVM_GUEST_MODE_SKIP
2138 stb r0, HSTATE_IN_GUEST(r13)
2140 /* Do the access with MSR:DR enabled */
2142 ori r4, r3, MSR_DR /* Enable paging for data */
2147 /* Store the result */
2148 stw r8, VCPU_LAST_INST(r9)
2150 /* Unset guest mode. */
2151 li r0, KVM_GUEST_MODE_HOST_HV
2152 stb r0, HSTATE_IN_GUEST(r13)
2156 std r4, VCPU_FAULT_DAR(r9)
2157 stw r6, VCPU_FAULT_DSISR(r9)
2160 std r5, VCPU_FAULT_GPA(r9)
2164 * Similarly for an HISI, reflect it to the guest as an ISI unless
2165 * it is an HPTE not found fault for a page that we have paged out.
2169 lbz r0, KVM_RADIX(r3)
2171 bne .Lradix_hisi /* for radix, just save ASDR */
2172 andis. r0, r11, SRR1_ISI_NOPT@h
2174 andi. r0, r11, MSR_IR /* instruction relocation enabled? */
2177 mfspr r5, SPRN_ASDR /* on POWER9, use ASDR to get VSID */
2179 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2181 PPC_SLBFEE_DOT(R5, R0) /* if so, look up SLB */
2182 li r0, BOOK3S_INTERRUPT_INST_SEGMENT
2183 bne 7f /* if no SLB entry found */
2185 /* Search the hash table. */
2186 mr r3, r9 /* vcpu pointer */
2189 li r7, 0 /* instruction fault */
2190 bl kvmppc_hpte_hv_fault
2191 ld r9, HSTATE_KVM_VCPU(r13)
2193 ld r11, VCPU_MSR(r9)
2194 li r12, BOOK3S_INTERRUPT_H_INST_STORAGE
2195 cmpdi r3, 0 /* retry the instruction */
2196 beq fast_interrupt_c_return
2197 cmpdi r3, -1 /* handle in kernel mode */
2200 /* Synthesize an ISI (or ISegI) for the guest */
2202 1: li r0, BOOK3S_INTERRUPT_INST_STORAGE
2203 7: mtspr SPRN_SRR0, r10
2204 mtspr SPRN_SRR1, r11
2206 bl kvmppc_msr_interrupt
2207 b fast_interrupt_c_return
2209 3: ld r6, VCPU_KVM(r9) /* not relocated, use VRMA */
2210 ld r5, KVM_VRMA_SLB_V(r6)
2214 * Try to handle an hcall in real mode.
2215 * Returns to the guest if we handle it, or continues on up to
2216 * the kernel if we can't (i.e. if we don't have a handler for
2217 * it, or if the handler returns H_TOO_HARD).
2219 * r5 - r8 contain hcall args,
2220 * r9 = vcpu, r10 = pc, r11 = msr, r12 = trap, r13 = paca
2222 hcall_try_real_mode:
2223 ld r3,VCPU_GPR(R3)(r9)
2225 /* sc 1 from userspace - reflect to guest syscall */
2226 bne sc_1_fast_return
2227 /* sc 1 from nested guest - give it to L1 to handle */
2228 ld r0, VCPU_NESTED(r9)
2232 cmpldi r3,hcall_real_table_end - hcall_real_table
2234 /* See if this hcall is enabled for in-kernel handling */
2236 srdi r0, r3, 8 /* r0 = (r3 / 4) >> 6 */
2237 sldi r0, r0, 3 /* index into kvm->arch.enabled_hcalls[] */
2239 ld r0, KVM_ENABLED_HCALLS(r4)
2240 rlwinm r4, r3, 32-2, 0x3f /* r4 = (r3 / 4) & 0x3f */
2244 /* Get pointer to handler, if any, and call it */
2245 LOAD_REG_ADDR(r4, hcall_real_table)
2251 mr r3,r9 /* get vcpu pointer */
2252 ld r4,VCPU_GPR(R4)(r9)
2255 beq hcall_real_fallback
2256 ld r4,HSTATE_KVM_VCPU(r13)
2257 std r3,VCPU_GPR(R3)(r4)
2265 li r10, BOOK3S_INTERRUPT_SYSCALL
2266 bl kvmppc_msr_interrupt
2270 /* We've attempted a real mode hcall, but it's punted it back
2271 * to userspace. We need to restore some clobbered volatiles
2272 * before resuming the pass-it-to-qemu path */
2273 hcall_real_fallback:
2274 li r12,BOOK3S_INTERRUPT_SYSCALL
2275 ld r9, HSTATE_KVM_VCPU(r13)
2279 .globl hcall_real_table
2281 .long 0 /* 0 - unused */
2282 .long DOTSYM(kvmppc_h_remove) - hcall_real_table
2283 .long DOTSYM(kvmppc_h_enter) - hcall_real_table
2284 .long DOTSYM(kvmppc_h_read) - hcall_real_table
2285 .long DOTSYM(kvmppc_h_clear_mod) - hcall_real_table
2286 .long DOTSYM(kvmppc_h_clear_ref) - hcall_real_table
2287 .long DOTSYM(kvmppc_h_protect) - hcall_real_table
2288 #ifdef CONFIG_SPAPR_TCE_IOMMU
2289 .long DOTSYM(kvmppc_h_get_tce) - hcall_real_table
2290 .long DOTSYM(kvmppc_rm_h_put_tce) - hcall_real_table
2295 .long 0 /* 0x24 - H_SET_SPRG0 */
2296 .long DOTSYM(kvmppc_h_set_dabr) - hcall_real_table
2297 .long DOTSYM(kvmppc_rm_h_page_init) - hcall_real_table
2311 #ifdef CONFIG_KVM_XICS
2312 .long DOTSYM(kvmppc_rm_h_eoi) - hcall_real_table
2313 .long DOTSYM(kvmppc_rm_h_cppr) - hcall_real_table
2314 .long DOTSYM(kvmppc_rm_h_ipi) - hcall_real_table
2315 .long DOTSYM(kvmppc_rm_h_ipoll) - hcall_real_table
2316 .long DOTSYM(kvmppc_rm_h_xirr) - hcall_real_table
2318 .long 0 /* 0x64 - H_EOI */
2319 .long 0 /* 0x68 - H_CPPR */
2320 .long 0 /* 0x6c - H_IPI */
2321 .long 0 /* 0x70 - H_IPOLL */
2322 .long 0 /* 0x74 - H_XIRR */
2350 .long DOTSYM(kvmppc_h_cede) - hcall_real_table
2351 .long DOTSYM(kvmppc_rm_h_confer) - hcall_real_table
2367 .long DOTSYM(kvmppc_h_bulk_remove) - hcall_real_table
2371 .long DOTSYM(kvmppc_h_set_xdabr) - hcall_real_table
2372 #ifdef CONFIG_SPAPR_TCE_IOMMU
2373 .long DOTSYM(kvmppc_rm_h_stuff_tce) - hcall_real_table
2374 .long DOTSYM(kvmppc_rm_h_put_tce_indirect) - hcall_real_table
2490 #ifdef CONFIG_KVM_XICS
2491 .long DOTSYM(kvmppc_rm_h_xirr_x) - hcall_real_table
2493 .long 0 /* 0x2fc - H_XIRR_X*/
2495 .long DOTSYM(kvmppc_h_random) - hcall_real_table
2496 .globl hcall_real_table_end
2497 hcall_real_table_end:
2499 _GLOBAL(kvmppc_h_set_xdabr)
2500 EXPORT_SYMBOL_GPL(kvmppc_h_set_xdabr)
2501 andi. r0, r5, DABRX_USER | DABRX_KERNEL
2503 li r0, DABRX_USER | DABRX_KERNEL | DABRX_BTI
2506 6: li r3, H_PARAMETER
2509 _GLOBAL(kvmppc_h_set_dabr)
2510 EXPORT_SYMBOL_GPL(kvmppc_h_set_dabr)
2511 li r5, DABRX_USER | DABRX_KERNEL
2515 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2516 std r4,VCPU_DABR(r3)
2517 stw r5, VCPU_DABRX(r3)
2518 mtspr SPRN_DABRX, r5
2519 /* Work around P7 bug where DABR can get corrupted on mtspr */
2520 1: mtspr SPRN_DABR,r4
2529 LOAD_REG_ADDR(r11, dawr_force_enable)
2536 /* Emulate H_SET_DABR/X on P8 for the sake of compat mode guests */
2537 rlwimi r5, r4, 5, DAWRX_DR | DAWRX_DW
2538 rlwimi r5, r4, 2, DAWRX_WT
2540 std r4, VCPU_DAWR(r3)
2541 std r5, VCPU_DAWRX(r3)
2543 * If came in through the real mode hcall handler then it is necessary
2544 * to write the registers since the return path won't. Otherwise it is
2545 * sufficient to store then in the vcpu struct as they will be loaded
2546 * next time the vcpu is run.
2549 andi. r6, r6, MSR_DR /* in real mode? */
2552 mtspr SPRN_DAWRX, r5
2556 _GLOBAL(kvmppc_h_cede) /* r3 = vcpu pointer, r11 = msr, r13 = paca */
2558 std r11,VCPU_MSR(r3)
2560 stb r0,VCPU_CEDED(r3)
2561 sync /* order setting ceded vs. testing prodded */
2562 lbz r5,VCPU_PRODDED(r3)
2564 bne kvm_cede_prodded
2565 li r12,0 /* set trap to 0 to say hcall is handled */
2566 stw r12,VCPU_TRAP(r3)
2568 std r0,VCPU_GPR(R3)(r3)
2571 * Set our bit in the bitmask of napping threads unless all the
2572 * other threads are already napping, in which case we send this
2575 ld r5,HSTATE_KVM_VCORE(r13)
2576 lbz r6,HSTATE_PTID(r13)
2577 lwz r8,VCORE_ENTRY_EXIT(r5)
2581 addi r6,r5,VCORE_NAPPING_THREADS
2588 /* order napping_threads update vs testing entry_exit_map */
2591 stb r0,HSTATE_NAPPING(r13)
2592 lwz r7,VCORE_ENTRY_EXIT(r5)
2594 bge 33f /* another thread already exiting */
2597 * Although not specifically required by the architecture, POWER7
2598 * preserves the following registers in nap mode, even if an SMT mode
2599 * switch occurs: SLB entries, PURR, SPURR, AMOR, UAMOR, AMR, SPRG0-3,
2600 * DAR, DSISR, DABR, DABRX, DSCR, PMCx, MMCRx, SIAR, SDAR.
2602 /* Save non-volatile GPRs */
2603 std r14, VCPU_GPR(R14)(r3)
2604 std r15, VCPU_GPR(R15)(r3)
2605 std r16, VCPU_GPR(R16)(r3)
2606 std r17, VCPU_GPR(R17)(r3)
2607 std r18, VCPU_GPR(R18)(r3)
2608 std r19, VCPU_GPR(R19)(r3)
2609 std r20, VCPU_GPR(R20)(r3)
2610 std r21, VCPU_GPR(R21)(r3)
2611 std r22, VCPU_GPR(R22)(r3)
2612 std r23, VCPU_GPR(R23)(r3)
2613 std r24, VCPU_GPR(R24)(r3)
2614 std r25, VCPU_GPR(R25)(r3)
2615 std r26, VCPU_GPR(R26)(r3)
2616 std r27, VCPU_GPR(R27)(r3)
2617 std r28, VCPU_GPR(R28)(r3)
2618 std r29, VCPU_GPR(R29)(r3)
2619 std r30, VCPU_GPR(R30)(r3)
2620 std r31, VCPU_GPR(R31)(r3)
2625 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2627 * Branch around the call if both CPU_FTR_TM and
2628 * CPU_FTR_P9_TM_HV_ASSIST are off.
2632 END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
2634 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
2636 ld r3, HSTATE_KVM_VCPU(r13)
2638 li r5, 0 /* don't preserve non-vol regs */
2639 bl kvmppc_save_tm_hv
2645 * Set DEC to the smaller of DEC and HDEC, so that we wake
2646 * no later than the end of our timeslice (HDEC interrupts
2647 * don't wake us from nap).
2653 /* On P9 check whether the guest has large decrementer mode enabled */
2654 ld r6, HSTATE_KVM_VCORE(r13)
2655 ld r6, VCORE_LPCR(r6)
2656 andis. r6, r6, LPCR_LD@h
2658 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2665 /* save expiry time of guest decrementer */
2667 ld r4, HSTATE_KVM_VCPU(r13)
2668 ld r5, HSTATE_KVM_VCORE(r13)
2669 ld r6, VCORE_TB_OFFSET_APPL(r5)
2670 subf r3, r6, r3 /* convert to host TB value */
2671 std r3, VCPU_DEC_EXPIRES(r4)
2673 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2674 ld r4, HSTATE_KVM_VCPU(r13)
2675 addi r3, r4, VCPU_TB_CEDE
2676 bl kvmhv_accumulate_time
2679 lis r3, LPCR_PECEDP@h /* Do wake on privileged doorbell */
2681 /* Go back to host stack */
2682 ld r1, HSTATE_HOST_R1(r13)
2685 * Take a nap until a decrementer or external or doobell interrupt
2686 * occurs, with PECE1 and PECE0 set in LPCR.
2687 * On POWER8, set PECEDH, and if we are ceding, also set PECEDP.
2688 * Also clear the runlatch bit before napping.
2691 mfspr r0, SPRN_CTRLF
2693 mtspr SPRN_CTRLT, r0
2696 stb r0,HSTATE_HWTHREAD_REQ(r13)
2698 ori r5,r5,LPCR_PECE0 | LPCR_PECE1
2700 ori r5, r5, LPCR_PECEDH
2701 rlwimi r5, r3, 0, LPCR_PECEDP
2702 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2704 kvm_nap_sequence: /* desired LPCR value in r5 */
2707 * PSSCR bits: exit criterion = 1 (wakeup based on LPCR at sreset)
2708 * enable state loss = 1 (allow SMT mode switch)
2709 * requested level = 0 (just stop dispatching)
2711 lis r3, (PSSCR_EC | PSSCR_ESL)@h
2712 /* Set LPCR_PECE_HVEE bit to enable wakeup by HV interrupts */
2713 li r4, LPCR_PECE_HVEE@higher
2717 li r3, PNV_THREAD_NAP
2718 ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_300)
2723 bl isa300_idle_stop_mayloss
2725 bl isa206_idle_insn_mayloss
2726 ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_300)
2728 mfspr r0, SPRN_CTRLF
2730 mtspr SPRN_CTRLT, r0
2735 stb r0, PACA_FTRACE_ENABLED(r13)
2737 li r0, KVM_HWTHREAD_IN_KVM
2738 stb r0, HSTATE_HWTHREAD_STATE(r13)
2740 lbz r0, HSTATE_NAPPING(r13)
2741 cmpwi r0, NAPPING_CEDE
2743 cmpwi r0, NAPPING_NOVCPU
2744 beq kvm_novcpu_wakeup
2745 cmpwi r0, NAPPING_UNSPLIT
2746 beq kvm_unsplit_wakeup
2747 twi 31,0,0 /* Nap state must not be zero */
2755 /* Woken by external or decrementer interrupt */
2757 /* get vcpu pointer */
2758 ld r4, HSTATE_KVM_VCPU(r13)
2760 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2761 addi r3, r4, VCPU_TB_RMINTR
2762 bl kvmhv_accumulate_time
2765 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2767 * Branch around the call if both CPU_FTR_TM and
2768 * CPU_FTR_P9_TM_HV_ASSIST are off.
2772 END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
2774 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
2778 li r5, 0 /* don't preserve non-vol regs */
2779 bl kvmppc_restore_tm_hv
2781 ld r4, HSTATE_KVM_VCPU(r13)
2785 /* load up FP state */
2788 /* Restore guest decrementer */
2789 ld r3, VCPU_DEC_EXPIRES(r4)
2790 ld r5, HSTATE_KVM_VCORE(r13)
2791 ld r6, VCORE_TB_OFFSET_APPL(r5)
2792 add r3, r3, r6 /* convert host TB to guest TB value */
2798 ld r14, VCPU_GPR(R14)(r4)
2799 ld r15, VCPU_GPR(R15)(r4)
2800 ld r16, VCPU_GPR(R16)(r4)
2801 ld r17, VCPU_GPR(R17)(r4)
2802 ld r18, VCPU_GPR(R18)(r4)
2803 ld r19, VCPU_GPR(R19)(r4)
2804 ld r20, VCPU_GPR(R20)(r4)
2805 ld r21, VCPU_GPR(R21)(r4)
2806 ld r22, VCPU_GPR(R22)(r4)
2807 ld r23, VCPU_GPR(R23)(r4)
2808 ld r24, VCPU_GPR(R24)(r4)
2809 ld r25, VCPU_GPR(R25)(r4)
2810 ld r26, VCPU_GPR(R26)(r4)
2811 ld r27, VCPU_GPR(R27)(r4)
2812 ld r28, VCPU_GPR(R28)(r4)
2813 ld r29, VCPU_GPR(R29)(r4)
2814 ld r30, VCPU_GPR(R30)(r4)
2815 ld r31, VCPU_GPR(R31)(r4)
2817 /* Check the wake reason in SRR1 to see why we got here */
2818 bl kvmppc_check_wake_reason
2821 * Restore volatile registers since we could have called a
2822 * C routine in kvmppc_check_wake_reason
2824 * r3 tells us whether we need to return to host or not
2825 * WARNING: it gets checked further down:
2826 * should not modify r3 until this check is done.
2828 ld r4, HSTATE_KVM_VCPU(r13)
2830 /* clear our bit in vcore->napping_threads */
2831 34: ld r5,HSTATE_KVM_VCORE(r13)
2832 lbz r7,HSTATE_PTID(r13)
2835 addi r6,r5,VCORE_NAPPING_THREADS
2841 stb r0,HSTATE_NAPPING(r13)
2843 /* See if the wake reason saved in r3 means we need to exit */
2844 stw r12, VCPU_TRAP(r4)
2848 b maybe_reenter_guest
2850 /* cede when already previously prodded case */
2853 stb r0,VCPU_PRODDED(r3)
2854 sync /* order testing prodded vs. clearing ceded */
2855 stb r0,VCPU_CEDED(r3)
2859 /* we've ceded but we want to give control to the host */
2861 ld r9, HSTATE_KVM_VCPU(r13)
2862 #ifdef CONFIG_KVM_XICS
2863 /* are we using XIVE with single escalation? */
2864 ld r10, VCPU_XIVE_ESC_VADDR(r9)
2867 li r6, XIVE_ESB_SET_PQ_00
2869 * If we still have a pending escalation, abort the cede,
2870 * and we must set PQ to 10 rather than 00 so that we don't
2871 * potentially end up with two entries for the escalation
2872 * interrupt in the XIVE interrupt queue. In that case
2873 * we also don't want to set xive_esc_on to 1 here in
2874 * case we race with xive_esc_irq().
2876 lbz r5, VCPU_XIVE_ESC_ON(r9)
2880 stb r0, VCPU_CEDED(r9)
2881 li r6, XIVE_ESB_SET_PQ_10
2884 stb r0, VCPU_XIVE_ESC_ON(r9)
2885 /* make sure store to xive_esc_on is seen before xive_esc_irq runs */
2887 5: /* Enable XIVE escalation */
2889 andi. r0, r0, MSR_DR /* in real mode? */
2893 1: ld r10, VCPU_XIVE_ESC_RADDR(r9)
2896 #endif /* CONFIG_KVM_XICS */
2897 3: b guest_exit_cont
2899 /* Try to do machine check recovery in real mode */
2900 machine_check_realmode:
2901 mr r3, r9 /* get vcpu pointer */
2902 bl kvmppc_realmode_machine_check
2904 /* all machine checks go to virtual mode for further handling */
2905 ld r9, HSTATE_KVM_VCPU(r13)
2906 li r12, BOOK3S_INTERRUPT_MACHINE_CHECK
2910 * Call C code to handle a HMI in real mode.
2911 * Only the primary thread does the call, secondary threads are handled
2912 * by calling hmi_exception_realmode() after kvmppc_hv_entry returns.
2913 * r9 points to the vcpu on entry
2916 lbz r0, HSTATE_PTID(r13)
2919 bl kvmppc_realmode_hmi_handler
2920 ld r9, HSTATE_KVM_VCPU(r13)
2921 li r12, BOOK3S_INTERRUPT_HMI
2925 * Check the reason we woke from nap, and take appropriate action.
2927 * 0 if nothing needs to be done
2928 * 1 if something happened that needs to be handled by the host
2929 * -1 if there was a guest wakeup (IPI or msgsnd)
2930 * -2 if we handled a PCI passthrough interrupt (returned by
2931 * kvmppc_read_intr only)
2933 * Also sets r12 to the interrupt vector for any interrupt that needs
2934 * to be handled now by the host (0x500 for external interrupt), or zero.
2935 * Modifies all volatile registers (since it may call a C function).
2936 * This routine calls kvmppc_read_intr, a C function, if an external
2937 * interrupt is pending.
2939 kvmppc_check_wake_reason:
2942 rlwinm r6, r6, 45-31, 0xf /* extract wake reason field (P8) */
2944 rlwinm r6, r6, 45-31, 0xe /* P7 wake reason field is 3 bits */
2945 ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_207S)
2946 cmpwi r6, 8 /* was it an external interrupt? */
2947 beq 7f /* if so, see what it was */
2950 cmpwi r6, 6 /* was it the decrementer? */
2953 cmpwi r6, 5 /* privileged doorbell? */
2955 cmpwi r6, 3 /* hypervisor doorbell? */
2957 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2958 cmpwi r6, 0xa /* Hypervisor maintenance ? */
2960 li r3, 1 /* anything else, return 1 */
2963 /* hypervisor doorbell */
2964 3: li r12, BOOK3S_INTERRUPT_H_DOORBELL
2967 * Clear the doorbell as we will invoke the handler
2968 * explicitly in the guest exit path.
2970 lis r6, (PPC_DBELL_SERVER << (63-36))@h
2972 /* see if it's a host IPI */
2977 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2978 lbz r0, HSTATE_HOST_IPI(r13)
2981 /* if not, return -1 */
2985 /* Woken up due to Hypervisor maintenance interrupt */
2986 4: li r12, BOOK3S_INTERRUPT_HMI
2990 /* external interrupt - create a stack frame so we can call C */
2992 std r0, PPC_LR_STKOFF(r1)
2993 stdu r1, -PPC_MIN_STKFRM(r1)
2996 li r12, BOOK3S_INTERRUPT_EXTERNAL
3001 * Return code of 2 means PCI passthrough interrupt, but
3002 * we need to return back to host to complete handling the
3003 * interrupt. Trap reason is expected in r12 by guest
3006 li r12, BOOK3S_INTERRUPT_HV_RM_HARD
3008 ld r0, PPC_MIN_STKFRM+PPC_LR_STKOFF(r1)
3009 addi r1, r1, PPC_MIN_STKFRM
3014 * Save away FP, VMX and VSX registers.
3016 * N.B. r30 and r31 are volatile across this function,
3017 * thus it is not callable from C.
3024 #ifdef CONFIG_ALTIVEC
3026 oris r8,r8,MSR_VEC@h
3027 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
3031 oris r8,r8,MSR_VSX@h
3032 END_FTR_SECTION_IFSET(CPU_FTR_VSX)
3035 addi r3,r3,VCPU_FPRS
3037 #ifdef CONFIG_ALTIVEC
3039 addi r3,r31,VCPU_VRS
3041 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
3043 mfspr r6,SPRN_VRSAVE
3044 stw r6,VCPU_VRSAVE(r31)
3049 * Load up FP, VMX and VSX registers
3051 * N.B. r30 and r31 are volatile across this function,
3052 * thus it is not callable from C.
3059 #ifdef CONFIG_ALTIVEC
3061 oris r8,r8,MSR_VEC@h
3062 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
3066 oris r8,r8,MSR_VSX@h
3067 END_FTR_SECTION_IFSET(CPU_FTR_VSX)
3070 addi r3,r4,VCPU_FPRS
3072 #ifdef CONFIG_ALTIVEC
3074 addi r3,r31,VCPU_VRS
3076 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
3078 lwz r7,VCPU_VRSAVE(r31)
3079 mtspr SPRN_VRSAVE,r7
3084 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3086 * Save transactional state and TM-related registers.
3087 * Called with r3 pointing to the vcpu struct and r4 containing
3088 * the guest MSR value.
3089 * r5 is non-zero iff non-volatile register state needs to be maintained.
3090 * If r5 == 0, this can modify all checkpointed registers, but
3091 * restores r1 and r2 before exit.
3093 _GLOBAL_TOC(kvmppc_save_tm_hv)
3094 EXPORT_SYMBOL_GPL(kvmppc_save_tm_hv)
3095 /* See if we need to handle fake suspend mode */
3098 END_FTR_SECTION_IFCLR(CPU_FTR_P9_TM_HV_ASSIST)
3100 lbz r0, HSTATE_FAKE_SUSPEND(r13) /* Were we fake suspended? */
3102 beq __kvmppc_save_tm
3104 /* The following code handles the fake_suspend = 1 case */
3106 std r0, PPC_LR_STKOFF(r1)
3107 stdu r1, -PPC_MIN_STKFRM(r1)
3112 rldimi r8, r0, MSR_TM_LG, 63-MSR_TM_LG
3115 rldicl. r8, r8, 64 - MSR_TS_S_LG, 62 /* Did we actually hrfid? */
3118 bl pnv_power9_force_smt4_catch
3119 END_FTR_SECTION_IFSET(CPU_FTR_P9_TM_XER_SO_BUG)
3122 /* We have to treclaim here because that's the only way to do S->N */
3123 li r3, TM_CAUSE_KVM_RESCHED
3127 * We were in fake suspend, so we are not going to save the
3128 * register state as the guest checkpointed state (since
3129 * we already have it), therefore we can now use any volatile GPR.
3130 * In fact treclaim in fake suspend state doesn't modify
3135 bl pnv_power9_force_smt4_release
3136 END_FTR_SECTION_IFSET(CPU_FTR_P9_TM_XER_SO_BUG)
3140 mfspr r3, SPRN_PSSCR
3141 /* PSSCR_FAKE_SUSPEND is a write-only bit, but clear it anyway */
3142 li r0, PSSCR_FAKE_SUSPEND
3144 mtspr SPRN_PSSCR, r3
3146 /* Don't save TEXASR, use value from last exit in real suspend state */
3147 ld r9, HSTATE_KVM_VCPU(r13)
3148 mfspr r5, SPRN_TFHAR
3149 mfspr r6, SPRN_TFIAR
3150 std r5, VCPU_TFHAR(r9)
3151 std r6, VCPU_TFIAR(r9)
3153 addi r1, r1, PPC_MIN_STKFRM
3154 ld r0, PPC_LR_STKOFF(r1)
3159 * Restore transactional state and TM-related registers.
3160 * Called with r3 pointing to the vcpu struct
3161 * and r4 containing the guest MSR value.
3162 * r5 is non-zero iff non-volatile register state needs to be maintained.
3163 * This potentially modifies all checkpointed registers.
3164 * It restores r1 and r2 from the PACA.
3166 _GLOBAL_TOC(kvmppc_restore_tm_hv)
3167 EXPORT_SYMBOL_GPL(kvmppc_restore_tm_hv)
3169 * If we are doing TM emulation for the guest on a POWER9 DD2,
3170 * then we don't actually do a trechkpt -- we either set up
3171 * fake-suspend mode, or emulate a TM rollback.
3174 b __kvmppc_restore_tm
3175 END_FTR_SECTION_IFCLR(CPU_FTR_P9_TM_HV_ASSIST)
3177 std r0, PPC_LR_STKOFF(r1)
3180 stb r0, HSTATE_FAKE_SUSPEND(r13)
3182 /* Turn on TM so we can restore TM SPRs */
3185 rldimi r5, r0, MSR_TM_LG, 63-MSR_TM_LG
3189 * The user may change these outside of a transaction, so they must
3190 * always be context switched.
3192 ld r5, VCPU_TFHAR(r3)
3193 ld r6, VCPU_TFIAR(r3)
3194 ld r7, VCPU_TEXASR(r3)
3195 mtspr SPRN_TFHAR, r5
3196 mtspr SPRN_TFIAR, r6
3197 mtspr SPRN_TEXASR, r7
3199 rldicl. r5, r4, 64 - MSR_TS_S_LG, 62
3200 beqlr /* TM not active in guest */
3202 /* Make sure the failure summary is set */
3203 oris r7, r7, (TEXASR_FS)@h
3204 mtspr SPRN_TEXASR, r7
3206 cmpwi r5, 1 /* check for suspended state */
3208 stb r5, HSTATE_FAKE_SUSPEND(r13)
3209 b 9f /* and return */
3210 10: stdu r1, -PPC_MIN_STKFRM(r1)
3211 /* guest is in transactional state, so simulate rollback */
3212 bl kvmhv_emulate_tm_rollback
3214 addi r1, r1, PPC_MIN_STKFRM
3215 9: ld r0, PPC_LR_STKOFF(r1)
3218 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
3221 * We come here if we get any exception or interrupt while we are
3222 * executing host real mode code while in guest MMU context.
3223 * r12 is (CR << 32) | vector
3224 * r13 points to our PACA
3225 * r12 is saved in HSTATE_SCRATCH0(r13)
3226 * ctr is saved in HSTATE_SCRATCH1(r13) if RELOCATABLE
3227 * r9 is saved in HSTATE_SCRATCH2(r13)
3228 * r13 is saved in HSPRG1
3229 * cfar is saved in HSTATE_CFAR(r13)
3230 * ppr is saved in HSTATE_PPR(r13)
3232 kvmppc_bad_host_intr:
3234 * Switch to the emergency stack, but start half-way down in
3235 * case we were already on it.
3239 ld r1, PACAEMERGSP(r13)
3240 subi r1, r1, THREAD_SIZE/2 + INT_FRAME_SIZE
3253 mfspr r3, SPRN_HSRR0
3254 mfspr r4, SPRN_HSRR1
3256 mfspr r6, SPRN_HDSISR
3258 1: mfspr r3, SPRN_SRR0
3261 mfspr r6, SPRN_DSISR
3266 ld r9, HSTATE_SCRATCH2(r13)
3267 ld r12, HSTATE_SCRATCH0(r13)
3272 ld r5, HSTATE_CFAR(r13)
3273 std r5, ORIG_GPR3(r1)
3275 #ifdef CONFIG_RELOCATABLE
3276 ld r4, HSTATE_SCRATCH1(r13)
3281 lbz r6, PACAIRQSOFTMASK(r13)
3287 LOAD_REG_IMMEDIATE(3, 0x7265677368657265)
3288 std r3, STACK_FRAME_OVERHEAD-16(r1)
3291 * On POWER9 do a minimal restore of the MMU and call C code,
3292 * which will print a message and panic.
3293 * XXX On POWER7 and POWER8, we just spin here since we don't
3294 * know what the other threads are doing (and we don't want to
3295 * coordinate with them) - but at least we now have register state
3296 * in memory that we might be able to look at from another CPU.
3300 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
3301 ld r9, HSTATE_KVM_VCPU(r13)
3302 ld r10, VCPU_KVM(r9)
3307 mtspr SPRN_CIABR, r0
3308 mtspr SPRN_DAWRX, r0
3310 BEGIN_MMU_FTR_SECTION
3312 END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
3317 ld r8, PACA_SLBSHADOWPTR(r13)
3318 .rept SLB_NUM_BOLTED
3319 li r3, SLBSHADOW_SAVEAREA
3323 andis. r7, r5, SLB_ESID_V@h
3329 4: lwz r7, KVM_HOST_LPID(r10)
3332 ld r8, KVM_HOST_LPCR(r10)
3335 li r0, KVM_GUEST_MODE_NONE
3336 stb r0, HSTATE_IN_GUEST(r13)
3339 * Turn on the MMU and jump to C code
3343 addi r3, r3, 9f - 5b
3345 rldimi r3, r4, 62, 0 /* ensure 0xc000000000000000 bits are set */
3346 ld r4, PACAKMSR(r13)
3350 9: addi r3, r1, STACK_FRAME_OVERHEAD
3351 bl kvmppc_bad_interrupt
3355 * This mimics the MSR transition on IRQ delivery. The new guest MSR is taken
3356 * from VCPU_INTR_MSR and is modified based on the required TM state changes.
3357 * r11 has the guest MSR value (in/out)
3358 * r9 has a vcpu pointer (in)
3359 * r0 is used as a scratch register
3361 kvmppc_msr_interrupt:
3362 rldicl r0, r11, 64 - MSR_TS_S_LG, 62
3363 cmpwi r0, 2 /* Check if we are in transactional state.. */
3364 ld r11, VCPU_INTR_MSR(r9)
3366 /* ... if transactional, change to suspended */
3368 1: rldimi r11, r0, MSR_TS_S_LG, 63 - MSR_TS_T_LG
3372 * Load up guest PMU state. R3 points to the vcpu struct.
3374 _GLOBAL(kvmhv_load_guest_pmu)
3375 EXPORT_SYMBOL_GPL(kvmhv_load_guest_pmu)
3379 sldi r3, r3, 31 /* MMCR0_FC (freeze counters) bit */
3380 mtspr SPRN_MMCR0, r3 /* freeze all counters, disable ints */
3383 ld r3, VCPU_MMCR(r4)
3384 andi. r5, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
3385 cmpwi r5, MMCR0_PMAO
3386 beql kvmppc_fix_pmao
3387 END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
3388 lwz r3, VCPU_PMC(r4) /* always load up guest PMU registers */
3389 lwz r5, VCPU_PMC + 4(r4) /* to prevent information leak */
3390 lwz r6, VCPU_PMC + 8(r4)
3391 lwz r7, VCPU_PMC + 12(r4)
3392 lwz r8, VCPU_PMC + 16(r4)
3393 lwz r9, VCPU_PMC + 20(r4)
3400 ld r3, VCPU_MMCR(r4)
3401 ld r5, VCPU_MMCR + 8(r4)
3402 ld r6, VCPU_MMCR + 16(r4)
3403 ld r7, VCPU_SIAR(r4)
3404 ld r8, VCPU_SDAR(r4)
3405 mtspr SPRN_MMCR1, r5
3406 mtspr SPRN_MMCRA, r6
3410 ld r5, VCPU_MMCR + 24(r4)
3411 ld r6, VCPU_SIER(r4)
3412 mtspr SPRN_MMCR2, r5
3414 BEGIN_FTR_SECTION_NESTED(96)
3415 lwz r7, VCPU_PMC + 24(r4)
3416 lwz r8, VCPU_PMC + 28(r4)
3417 ld r9, VCPU_MMCR + 32(r4)
3418 mtspr SPRN_SPMC1, r7
3419 mtspr SPRN_SPMC2, r8
3420 mtspr SPRN_MMCRS, r9
3421 END_FTR_SECTION_NESTED(CPU_FTR_ARCH_300, 0, 96)
3422 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3423 mtspr SPRN_MMCR0, r3
3429 * Reload host PMU state saved in the PACA by kvmhv_save_host_pmu.
3431 _GLOBAL(kvmhv_load_host_pmu)
3432 EXPORT_SYMBOL_GPL(kvmhv_load_host_pmu)
3434 lbz r4, PACA_PMCINUSE(r13) /* is the host using the PMU? */
3436 beq 23f /* skip if not */
3438 ld r3, HSTATE_MMCR0(r13)
3439 andi. r4, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
3440 cmpwi r4, MMCR0_PMAO
3441 beql kvmppc_fix_pmao
3442 END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
3443 lwz r3, HSTATE_PMC1(r13)
3444 lwz r4, HSTATE_PMC2(r13)
3445 lwz r5, HSTATE_PMC3(r13)
3446 lwz r6, HSTATE_PMC4(r13)
3447 lwz r8, HSTATE_PMC5(r13)
3448 lwz r9, HSTATE_PMC6(r13)
3455 ld r3, HSTATE_MMCR0(r13)
3456 ld r4, HSTATE_MMCR1(r13)
3457 ld r5, HSTATE_MMCRA(r13)
3458 ld r6, HSTATE_SIAR(r13)
3459 ld r7, HSTATE_SDAR(r13)
3460 mtspr SPRN_MMCR1, r4
3461 mtspr SPRN_MMCRA, r5
3465 ld r8, HSTATE_MMCR2(r13)
3466 ld r9, HSTATE_SIER(r13)
3467 mtspr SPRN_MMCR2, r8
3469 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3470 mtspr SPRN_MMCR0, r3
3476 * Save guest PMU state into the vcpu struct.
3477 * r3 = vcpu, r4 = full save flag (PMU in use flag set in VPA)
3479 _GLOBAL(kvmhv_save_guest_pmu)
3480 EXPORT_SYMBOL_GPL(kvmhv_save_guest_pmu)
3485 * POWER8 seems to have a hardware bug where setting
3486 * MMCR0[PMAE] along with MMCR0[PMC1CE] and/or MMCR0[PMCjCE]
3487 * when some counters are already negative doesn't seem
3488 * to cause a performance monitor alert (and hence interrupt).
3489 * The effect of this is that when saving the PMU state,
3490 * if there is no PMU alert pending when we read MMCR0
3491 * before freezing the counters, but one becomes pending
3492 * before we read the counters, we lose it.
3493 * To work around this, we need a way to freeze the counters
3494 * before reading MMCR0. Normally, freezing the counters
3495 * is done by writing MMCR0 (to set MMCR0[FC]) which
3496 * unavoidably writes MMCR0[PMA0] as well. On POWER8,
3497 * we can also freeze the counters using MMCR2, by writing
3498 * 1s to all the counter freeze condition bits (there are
3499 * 9 bits each for 6 counters).
3501 li r3, -1 /* set all freeze bits */
3503 mfspr r10, SPRN_MMCR2
3504 mtspr SPRN_MMCR2, r3
3506 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3508 sldi r3, r3, 31 /* MMCR0_FC (freeze counters) bit */
3509 mfspr r4, SPRN_MMCR0 /* save MMCR0 */
3510 mtspr SPRN_MMCR0, r3 /* freeze all counters, disable ints */
3511 mfspr r6, SPRN_MMCRA
3512 /* Clear MMCRA in order to disable SDAR updates */
3514 mtspr SPRN_MMCRA, r7
3516 cmpwi r8, 0 /* did they ask for PMU stuff to be saved? */
3518 std r3, VCPU_MMCR(r9) /* if not, set saved MMCR0 to FC */
3520 21: mfspr r5, SPRN_MMCR1
3523 std r4, VCPU_MMCR(r9)
3524 std r5, VCPU_MMCR + 8(r9)
3525 std r6, VCPU_MMCR + 16(r9)
3527 std r10, VCPU_MMCR + 24(r9)
3528 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3529 std r7, VCPU_SIAR(r9)
3530 std r8, VCPU_SDAR(r9)
3537 stw r3, VCPU_PMC(r9)
3538 stw r4, VCPU_PMC + 4(r9)
3539 stw r5, VCPU_PMC + 8(r9)
3540 stw r6, VCPU_PMC + 12(r9)
3541 stw r7, VCPU_PMC + 16(r9)
3542 stw r8, VCPU_PMC + 20(r9)
3545 std r5, VCPU_SIER(r9)
3546 BEGIN_FTR_SECTION_NESTED(96)
3547 mfspr r6, SPRN_SPMC1
3548 mfspr r7, SPRN_SPMC2
3549 mfspr r8, SPRN_MMCRS
3550 stw r6, VCPU_PMC + 24(r9)
3551 stw r7, VCPU_PMC + 28(r9)
3552 std r8, VCPU_MMCR + 32(r9)
3554 mtspr SPRN_MMCRS, r4
3555 END_FTR_SECTION_NESTED(CPU_FTR_ARCH_300, 0, 96)
3556 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3560 * This works around a hardware bug on POWER8E processors, where
3561 * writing a 1 to the MMCR0[PMAO] bit doesn't generate a
3562 * performance monitor interrupt. Instead, when we need to have
3563 * an interrupt pending, we have to arrange for a counter to overflow.
3567 mtspr SPRN_MMCR2, r3
3568 lis r3, (MMCR0_PMXE | MMCR0_FCECE)@h
3569 ori r3, r3, MMCR0_PMCjCE | MMCR0_C56RUN
3570 mtspr SPRN_MMCR0, r3
3577 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
3579 * Start timing an activity
3580 * r3 = pointer to time accumulation struct, r4 = vcpu
3583 ld r5, HSTATE_KVM_VCORE(r13)
3584 ld r6, VCORE_TB_OFFSET_APPL(r5)
3586 subf r5, r6, r5 /* subtract current timebase offset */
3587 std r3, VCPU_CUR_ACTIVITY(r4)
3588 std r5, VCPU_ACTIVITY_START(r4)
3592 * Accumulate time to one activity and start another.
3593 * r3 = pointer to new time accumulation struct, r4 = vcpu
3595 kvmhv_accumulate_time:
3596 ld r5, HSTATE_KVM_VCORE(r13)
3597 ld r8, VCORE_TB_OFFSET_APPL(r5)
3598 ld r5, VCPU_CUR_ACTIVITY(r4)
3599 ld r6, VCPU_ACTIVITY_START(r4)
3600 std r3, VCPU_CUR_ACTIVITY(r4)
3602 subf r7, r8, r7 /* subtract current timebase offset */
3603 std r7, VCPU_ACTIVITY_START(r4)
3607 ld r8, TAS_SEQCOUNT(r5)
3610 std r8, TAS_SEQCOUNT(r5)
3612 ld r7, TAS_TOTAL(r5)
3614 std r7, TAS_TOTAL(r5)
3620 3: std r3, TAS_MIN(r5)
3626 std r8, TAS_SEQCOUNT(r5)