1 /* SPDX-License-Identifier: GPL-2.0-only */
6 * Derived from book3s_rmhandlers.S and other files, which are:
8 * Copyright SUSE Linux Products GmbH 2009
13 #include <linux/export.h>
14 #include <linux/linkage.h>
15 #include <linux/objtool.h>
16 #include <asm/ppc_asm.h>
17 #include <asm/code-patching-asm.h>
18 #include <asm/kvm_asm.h>
22 #include <asm/ptrace.h>
23 #include <asm/hvcall.h>
24 #include <asm/asm-offsets.h>
25 #include <asm/exception-64s.h>
26 #include <asm/kvm_book3s_asm.h>
27 #include <asm/book3s/64/mmu-hash.h>
30 #include <asm/thread_info.h>
31 #include <asm/asm-compat.h>
32 #include <asm/feature-fixups.h>
33 #include <asm/cpuidle.h>
35 /* Values in HSTATE_NAPPING(r13) */
36 #define NAPPING_CEDE 1
37 #define NAPPING_NOVCPU 2
38 #define NAPPING_UNSPLIT 3
40 /* Stack frame offsets for kvmppc_hv_entry */
42 #define STACK_SLOT_TRAP (SFS-4)
43 #define STACK_SLOT_TID (SFS-16)
44 #define STACK_SLOT_PSSCR (SFS-24)
45 #define STACK_SLOT_PID (SFS-32)
46 #define STACK_SLOT_IAMR (SFS-40)
47 #define STACK_SLOT_CIABR (SFS-48)
48 #define STACK_SLOT_DAWR0 (SFS-56)
49 #define STACK_SLOT_DAWRX0 (SFS-64)
50 #define STACK_SLOT_HFSCR (SFS-72)
51 #define STACK_SLOT_AMR (SFS-80)
52 #define STACK_SLOT_UAMOR (SFS-88)
53 #define STACK_SLOT_FSCR (SFS-96)
56 * Use the last LPID (all implemented LPID bits = 1) for partition switching.
57 * This is reserved in the LPID allocator. POWER7 only implements 0x3ff, but
58 * we write 0xfff into the LPID SPR anyway, which seems to work and just
59 * ignores the top bits.
61 #define LPID_RSVD 0xfff
64 * Call kvmppc_hv_entry in real mode.
65 * Must be called with interrupts hard-disabled.
69 * LR = return address to continue at after eventually re-enabling MMU
71 _GLOBAL_TOC(kvmppc_hv_entry_trampoline)
73 std r0, PPC_LR_STKOFF(r1)
76 std r10, HSTATE_HOST_MSR(r13)
77 LOAD_REG_ADDR(r5, kvmppc_call_hv_entry)
82 mtmsrd r0,1 /* clear RI in MSR */
88 ld r4, HSTATE_KVM_VCPU(r13)
91 /* Back from guest - restore host state and return to caller */
94 /* Restore host DABR and DABRX */
95 ld r5,HSTATE_DABR(r13)
99 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
102 ld r3,PACA_SPRG_VDSO(r13)
103 mtspr SPRN_SPRG_VDSO_WRITE,r3
105 /* Reload the host's PMU registers */
106 bl kvmhv_load_host_pmu
109 * Reload DEC. HDEC interrupts were disabled when
110 * we reloaded the host's LPCR value.
112 ld r3, HSTATE_DECEXP(r13)
117 /* hwthread_req may have got set by cede or no vcpu, so clear it */
119 stb r0, HSTATE_HWTHREAD_REQ(r13)
122 * For external interrupts we need to call the Linux
123 * handler to process the interrupt. We do that by jumping
124 * to absolute address 0x500 for external interrupts.
125 * The [h]rfid at the end of the handler will return to
126 * the book3s_hv_interrupts.S code. For other interrupts
127 * we do the rfid to get back to the book3s_hv_interrupts.S
130 ld r8, 112+PPC_LR_STKOFF(r1)
132 ld r7, HSTATE_HOST_MSR(r13)
134 /* Return the trap number on this thread as the return value */
137 /* RFI into the highmem handler */
141 mtmsrd r6, 1 /* Clear RI in MSR */
146 kvmppc_primary_no_guest:
147 /* We handle this much like a ceded vcpu */
148 /* put the HDEC into the DEC, since HDEC interrupts don't wake us */
149 /* HDEC may be larger than DEC for arch >= v3.00, but since the */
150 /* HDEC value came from DEC in the first place, it will fit */
154 * Make sure the primary has finished the MMU switch.
155 * We should never get here on a secondary thread, but
156 * check it for robustness' sake.
158 ld r5, HSTATE_KVM_VCORE(r13)
159 65: lbz r0, VCORE_IN_GUEST(r5)
166 /* set our bit in napping_threads */
167 ld r5, HSTATE_KVM_VCORE(r13)
168 lbz r7, HSTATE_PTID(r13)
171 addi r6, r5, VCORE_NAPPING_THREADS
176 /* order napping_threads update vs testing entry_exit_map */
179 lwz r7, VCORE_ENTRY_EXIT(r5)
181 bge kvm_novcpu_exit /* another thread already exiting */
182 li r3, NAPPING_NOVCPU
183 stb r3, HSTATE_NAPPING(r13)
185 li r3, 0 /* Don't wake on privileged (OS) doorbell */
190 * Entered from kvm_start_guest if kvm_hstate.napping is set
196 ld r1, HSTATE_HOST_R1(r13)
197 ld r5, HSTATE_KVM_VCORE(r13)
199 stb r0, HSTATE_NAPPING(r13)
201 /* check the wake reason */
202 bl kvmppc_check_wake_reason
205 * Restore volatile registers since we could have called
206 * a C routine in kvmppc_check_wake_reason.
209 ld r5, HSTATE_KVM_VCORE(r13)
211 /* see if any other thread is already exiting */
212 lwz r0, VCORE_ENTRY_EXIT(r5)
216 /* clear our bit in napping_threads */
217 lbz r7, HSTATE_PTID(r13)
220 addi r6, r5, VCORE_NAPPING_THREADS
226 /* See if the wake reason means we need to exit */
230 /* See if our timeslice has expired (HDEC is negative) */
233 li r12, BOOK3S_INTERRUPT_HV_DECREMENTER
237 /* Got an IPI but other vcpus aren't yet exiting, must be a latecomer */
238 ld r4, HSTATE_KVM_VCPU(r13)
240 beq kvmppc_primary_no_guest
242 #ifdef CONFIG_KVM_BOOK3S_HV_P8_TIMING
243 addi r3, r4, VCPU_TB_RMENTRY
244 bl kvmhv_start_timing
249 #ifdef CONFIG_KVM_BOOK3S_HV_P8_TIMING
250 ld r4, HSTATE_KVM_VCPU(r13)
253 addi r3, r4, VCPU_TB_RMEXIT
254 bl kvmhv_accumulate_time
257 stw r12, STACK_SLOT_TRAP(r1)
258 bl kvmhv_commence_exit
260 b kvmhv_switch_to_host
263 * We come in here when wakened from Linux offline idle code.
265 * r3 contains the SRR1 wakeup value, SRR1 is trashed.
267 _GLOBAL(idle_kvm_start_guest)
270 std r5, 8(r1) // Save CR in caller's frame
271 std r0, 16(r1) // Save LR in caller's frame
272 // Create frame on emergency stack
273 ld r4, PACAEMERGSP(r13)
274 stdu r1, -SWITCH_FRAME_SIZE(r4)
275 // Switch to new frame on emergency stack
277 std r3, 32(r1) // Save SRR1 wakeup value
281 * Could avoid this and pass it through in r3. For now,
282 * code expects it to be in SRR1.
287 stb r0,PACA_FTRACE_ENABLED(r13)
289 li r0,KVM_HWTHREAD_IN_KVM
290 stb r0,HSTATE_HWTHREAD_STATE(r13)
292 /* kvm cede / napping does not come through here */
293 lbz r0,HSTATE_NAPPING(r13)
300 stb r0, HSTATE_NAPPING(r13)
305 * We weren't napping due to cede, so this must be a secondary
306 * thread being woken up to run a guest, or being woken up due
307 * to a stray IPI. (Or due to some machine check or hypervisor
308 * maintenance interrupt while the core is in KVM.)
311 /* Check the wake reason in SRR1 to see why we got here */
312 bl kvmppc_check_wake_reason
314 * kvmppc_check_wake_reason could invoke a C routine, but we
315 * have no volatile registers to restore when we return.
321 /* get vcore pointer, NULL if we have nothing to run */
322 ld r5,HSTATE_KVM_VCORE(r13)
324 /* if we have no vcore to run, go back to sleep */
327 kvm_secondary_got_guest:
329 // About to go to guest, clear saved SRR1
333 /* Set HSTATE_DSCR(r13) to something sensible */
334 ld r6, PACA_DSCR_DEFAULT(r13)
335 std r6, HSTATE_DSCR(r13)
337 /* On thread 0 of a subcore, set HDEC to max */
338 lbz r4, HSTATE_PTID(r13)
341 lis r6,0x7fff /* MAX_INT@h */
343 /* and set per-LPAR registers, if doing dynamic micro-threading */
344 ld r6, HSTATE_SPLIT_MODE(r13)
347 ld r0, KVM_SPLIT_RPR(r6)
349 ld r0, KVM_SPLIT_PMMAR(r6)
351 ld r0, KVM_SPLIT_LDBAR(r6)
355 /* Order load of vcpu after load of vcore */
357 ld r4, HSTATE_KVM_VCPU(r13)
360 /* Back from the guest, go back to nap */
361 /* Clear our vcpu and vcore pointers so we don't come back in early */
363 std r0, HSTATE_KVM_VCPU(r13)
365 * Once we clear HSTATE_KVM_VCORE(r13), the code in
366 * kvmppc_run_core() is going to assume that all our vcpu
367 * state is visible in memory. This lwsync makes sure
371 std r0, HSTATE_KVM_VCORE(r13)
374 * All secondaries exiting guest will fall through this path.
375 * Before proceeding, just check for HMI interrupt and
376 * invoke opal hmi handler. By now we are sure that the
377 * primary thread on this core/subcore has already made partition
378 * switch/TB resync and we are good to call opal hmi handler.
380 cmpwi r12, BOOK3S_INTERRUPT_HMI
383 li r3,0 /* NULL argument */
384 bl CFUNC(hmi_exception_realmode)
386 * At this point we have finished executing in the guest.
387 * We need to wait for hwthread_req to become zero, since
388 * we may not turn on the MMU while hwthread_req is non-zero.
389 * While waiting we also need to check if we get given a vcpu to run.
392 lbz r3, HSTATE_HWTHREAD_REQ(r13)
396 li r0, KVM_HWTHREAD_IN_KERNEL
397 stb r0, HSTATE_HWTHREAD_STATE(r13)
398 /* need to recheck hwthread_req after a barrier, to avoid race */
400 lbz r3, HSTATE_HWTHREAD_REQ(r13)
405 * Jump to idle_return_gpr_loss, which returns to the
406 * idle_kvm_start_guest caller.
410 rlwimi r4, r3, 0, LPCR_PECE0 | LPCR_PECE1
412 // Return SRR1 wakeup value, or 0 if we went into the guest
415 ld r1, 0(r1) // Switch back to caller stack
416 ld r0, 16(r1) // Reload LR
417 ld r5, 8(r1) // Reload CR
424 ld r5, HSTATE_KVM_VCORE(r13)
427 ld r3, HSTATE_SPLIT_MODE(r13)
430 lbz r0, KVM_SPLIT_DO_NAP(r3)
436 b kvm_secondary_got_guest
438 54: li r0, KVM_HWTHREAD_IN_KVM
439 stb r0, HSTATE_HWTHREAD_STATE(r13)
443 * Here the primary thread is trying to return the core to
444 * whole-core mode, so we need to nap.
448 * When secondaries are napping in kvm_unsplit_nap() with
449 * hwthread_req = 1, HMI goes ignored even though subcores are
450 * already exited the guest. Hence HMI keeps waking up secondaries
451 * from nap in a loop and secondaries always go back to nap since
452 * no vcore is assigned to them. This makes impossible for primary
453 * thread to get hold of secondary threads resulting into a soft
454 * lockup in KVM path.
456 * Let us check if HMI is pending and handle it before we go to nap.
458 cmpwi r12, BOOK3S_INTERRUPT_HMI
460 li r3, 0 /* NULL argument */
461 bl CFUNC(hmi_exception_realmode)
464 * Ensure that secondary doesn't nap when it has
465 * its vcore pointer set.
467 sync /* matches smp_mb() before setting split_info.do_nap */
468 ld r0, HSTATE_KVM_VCORE(r13)
471 /* clear any pending message */
473 lis r6, (PPC_DBELL_SERVER << (63-36))@h
475 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
476 /* Set kvm_split_mode.napped[tid] = 1 */
477 ld r3, HSTATE_SPLIT_MODE(r13)
479 lhz r4, PACAPACAINDEX(r13)
480 clrldi r4, r4, 61 /* micro-threading => P8 => 8 threads/core */
481 addi r4, r4, KVM_SPLIT_NAPPED
483 /* Check the do_nap flag again after setting napped[] */
485 lbz r0, KVM_SPLIT_DO_NAP(r3)
488 li r3, NAPPING_UNSPLIT
489 stb r3, HSTATE_NAPPING(r13)
490 li r3, (LPCR_PECEDH | LPCR_PECE0) >> 4
492 rlwimi r5, r3, 4, (LPCR_PECEDP | LPCR_PECEDH | LPCR_PECE0 | LPCR_PECE1)
499 /******************************************************************************
503 *****************************************************************************/
505 SYM_CODE_START_LOCAL(kvmppc_hv_entry)
509 * R4 = vcpu pointer (or NULL)
514 * all other volatile GPRS = free
515 * Does not preserve non-volatile GPRs or CR fields
518 std r0, PPC_LR_STKOFF(r1)
521 /* Save R1 in the PACA */
522 std r1, HSTATE_HOST_R1(r13)
524 li r6, KVM_GUEST_MODE_HOST_HV
525 stb r6, HSTATE_IN_GUEST(r13)
527 #ifdef CONFIG_KVM_BOOK3S_HV_P8_TIMING
528 /* Store initial timestamp */
531 addi r3, r4, VCPU_TB_RMENTRY
532 bl kvmhv_start_timing
536 ld r5, HSTATE_KVM_VCORE(r13)
537 ld r9, VCORE_KVM(r5) /* pointer to struct kvm */
540 * POWER7/POWER8 host -> guest partition switch code.
541 * We don't have to lock against concurrent tlbies,
542 * but we do have to coordinate across hardware threads.
544 /* Set bit in entry map iff exit map is zero. */
546 lbz r6, HSTATE_PTID(r13)
548 addi r8, r5, VCORE_ENTRY_EXIT
550 cmpwi r3, 0x100 /* any threads starting to exit? */
551 bge secondary_too_late /* if so we're too late to the party */
556 /* Primary thread switches to guest partition. */
562 li r0,LPID_RSVD /* switch to reserved LPID */
565 mtspr SPRN_SDR1,r6 /* switch to partition page table */
569 /* See if we need to flush the TLB. */
570 mr r3, r9 /* kvm pointer */
571 lhz r4, PACAPACAINDEX(r13) /* physical cpu number */
572 li r5, 0 /* nested vcpu pointer */
573 bl kvmppc_check_need_tlb_flush
575 ld r5, HSTATE_KVM_VCORE(r13)
577 /* Add timebase offset onto timebase */
578 22: ld r8,VCORE_TB_OFFSET(r5)
581 std r8, VCORE_TB_OFFSET_APPL(r5)
582 mftb r6 /* current host timebase */
584 mtspr SPRN_TBU40,r8 /* update upper 40 bits */
585 mftb r7 /* check if lower 24 bits overflowed */
590 addis r8,r8,0x100 /* if so, increment upper 40 bits */
593 /* Load guest PCR value to select appropriate compat mode */
594 37: ld r7, VCORE_PCR(r5)
595 LOAD_REG_IMMEDIATE(r6, PCR_MASK)
603 /* DPDES and VTB are shared between threads */
604 ld r8, VCORE_DPDES(r5)
608 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
610 /* Mark the subcore state as inside guest */
611 bl kvmppc_subcore_enter_guest
613 ld r5, HSTATE_KVM_VCORE(r13)
614 ld r4, HSTATE_KVM_VCPU(r13)
616 stb r0,VCORE_IN_GUEST(r5) /* signal secondaries to continue */
618 /* Do we have a guest vcpu to run? */
620 beq kvmppc_primary_no_guest
622 /* Increment yield count if they have a VPA */
626 li r6, LPPACA_YIELDCOUNT
631 stb r6, VCPU_VPA_DIRTY(r4)
634 /* Save purr/spurr */
637 std r5,HSTATE_PURR(r13)
638 std r6,HSTATE_SPURR(r13)
644 /* Save host values of some registers */
648 mfspr r7, SPRN_DAWRX0
650 std r5, STACK_SLOT_CIABR(r1)
651 std r6, STACK_SLOT_DAWR0(r1)
652 std r7, STACK_SLOT_DAWRX0(r1)
653 std r8, STACK_SLOT_IAMR(r1)
655 std r5, STACK_SLOT_FSCR(r1)
656 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
659 std r5, STACK_SLOT_AMR(r1)
661 std r6, STACK_SLOT_UAMOR(r1)
664 /* Set partition DABR */
665 /* Do this before re-enabling PMU to avoid P7 DABR corruption bug */
666 lwz r5,VCPU_DABRX(r4)
671 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
673 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
676 END_FTR_SECTION_IFCLR(CPU_FTR_TM)
678 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
682 li r5, 0 /* don't preserve non-vol regs */
683 bl kvmppc_restore_tm_hv
685 ld r4, HSTATE_KVM_VCPU(r13)
689 /* Load guest PMU registers; r4 = vcpu pointer here */
691 bl kvmhv_load_guest_pmu
693 /* Load up FP, VMX and VSX registers */
694 ld r4, HSTATE_KVM_VCPU(r13)
697 ld r14, VCPU_GPR(R14)(r4)
698 ld r15, VCPU_GPR(R15)(r4)
699 ld r16, VCPU_GPR(R16)(r4)
700 ld r17, VCPU_GPR(R17)(r4)
701 ld r18, VCPU_GPR(R18)(r4)
702 ld r19, VCPU_GPR(R19)(r4)
703 ld r20, VCPU_GPR(R20)(r4)
704 ld r21, VCPU_GPR(R21)(r4)
705 ld r22, VCPU_GPR(R22)(r4)
706 ld r23, VCPU_GPR(R23)(r4)
707 ld r24, VCPU_GPR(R24)(r4)
708 ld r25, VCPU_GPR(R25)(r4)
709 ld r26, VCPU_GPR(R26)(r4)
710 ld r27, VCPU_GPR(R27)(r4)
711 ld r28, VCPU_GPR(R28)(r4)
712 ld r29, VCPU_GPR(R29)(r4)
713 ld r30, VCPU_GPR(R30)(r4)
714 ld r31, VCPU_GPR(R31)(r4)
716 /* Switch DSCR to guest value */
721 /* Skip next section on POWER7 */
723 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
724 /* Load up POWER8-specific registers */
726 lwz r6, VCPU_PSPB(r4)
732 * Handle broken DAWR case by not writing it. This means we
733 * can still store the DAWR register for migration.
735 LOAD_REG_ADDR(r5, dawr_force_enable)
739 ld r5, VCPU_DAWR0(r4)
740 ld r6, VCPU_DAWRX0(r4)
742 mtspr SPRN_DAWRX0, r6
744 ld r7, VCPU_CIABR(r4)
749 ld r8, VCPU_EBBHR(r4)
752 ld r5, VCPU_EBBRR(r4)
753 ld r6, VCPU_BESCR(r4)
754 lwz r7, VCPU_GUEST_PID(r4)
760 /* POWER8-only registers */
761 ld r5, VCPU_TCSCR(r4)
763 ld r7, VCPU_CSIGR(r4)
772 ld r5, VCPU_SPRG0(r4)
773 ld r6, VCPU_SPRG1(r4)
774 ld r7, VCPU_SPRG2(r4)
775 ld r8, VCPU_SPRG3(r4)
781 /* Load up DAR and DSISR */
783 lwz r6, VCPU_DSISR(r4)
787 /* Restore AMR and UAMOR, set AMOR to all 1s */
793 /* Restore state of CTRL run bit; the host currently has it set to 1 */
800 /* Secondary threads wait for primary to have done partition switch */
801 ld r5, HSTATE_KVM_VCORE(r13)
802 lbz r6, HSTATE_PTID(r13)
805 lbz r0, VCORE_IN_GUEST(r5)
809 20: lwz r3, VCORE_ENTRY_EXIT(r5)
812 lbz r0, VCORE_IN_GUEST(r5)
823 * Set the decrementer to the guest decrementer.
825 ld r8,VCPU_DEC_EXPIRES(r4)
830 /* Check if HDEC expires soon */
833 cmpdi r3, 512 /* 1 microsecond */
836 /* Clear out and reload the SLB */
842 /* Load up guest SLB entries (N.B. slb_max will be 0 for radix) */
843 lwz r5,VCPU_SLB_MAX(r4)
848 1: ld r8,VCPU_SLB_E(r6)
851 addi r6,r6,VCPU_SLB_SIZE
855 deliver_guest_interrupt: /* r4 = vcpu, r13 = paca */
856 /* Check if we can deliver an external or decrementer interrupt now */
857 ld r0, VCPU_PENDING_EXC(r4)
861 bl CFUNC(kvmppc_guest_entry_inject_int)
862 ld r4, HSTATE_KVM_VCPU(r13)
871 /* r11 = vcpu->arch.msr & ~MSR_HV */
872 rldicl r11, r11, 63 - MSR_HV_LG, 1
873 rotldi r11, r11, 1 + MSR_HV_LG
884 * R10: value for HSRR0
885 * R11: value for HSRR1
890 stb r0,VCPU_CEDED(r4) /* cancel cede */
894 /* Activate guest mode, so faults get handled by KVM */
895 li r9, KVM_GUEST_MODE_GUEST_HV
896 stb r9, HSTATE_IN_GUEST(r13)
898 #ifdef CONFIG_KVM_BOOK3S_HV_P8_TIMING
899 /* Accumulate timing */
900 addi r3, r4, VCPU_TB_GUEST
901 bl kvmhv_accumulate_time
909 END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
912 END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
917 ld r1, VCPU_GPR(R1)(r4)
918 ld r5, VCPU_GPR(R5)(r4)
919 ld r8, VCPU_GPR(R8)(r4)
920 ld r9, VCPU_GPR(R9)(r4)
921 ld r10, VCPU_GPR(R10)(r4)
922 ld r11, VCPU_GPR(R11)(r4)
923 ld r12, VCPU_GPR(R12)(r4)
924 ld r13, VCPU_GPR(R13)(r4)
928 END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
930 ld r6, VCPU_GPR(R6)(r4)
931 ld r7, VCPU_GPR(R7)(r4)
936 ld r0, VCPU_GPR(R0)(r4)
937 ld r2, VCPU_GPR(R2)(r4)
938 ld r3, VCPU_GPR(R3)(r4)
939 ld r4, VCPU_GPR(R4)(r4)
942 SYM_CODE_END(kvmppc_hv_entry)
946 stw r12, STACK_SLOT_TRAP(r1)
949 stw r12, VCPU_TRAP(r4)
950 #ifdef CONFIG_KVM_BOOK3S_HV_P8_TIMING
951 addi r3, r4, VCPU_TB_RMEXIT
952 bl kvmhv_accumulate_time
954 11: b kvmhv_switch_to_host
961 li r12, BOOK3S_INTERRUPT_HV_DECREMENTER
962 12: stw r12, VCPU_TRAP(r4)
964 #ifdef CONFIG_KVM_BOOK3S_HV_P8_TIMING
965 addi r3, r4, VCPU_TB_RMEXIT
966 bl kvmhv_accumulate_time
970 /******************************************************************************
974 *****************************************************************************/
977 * We come here from the first-level interrupt handlers.
979 .globl kvmppc_interrupt_hv
983 * R9 = HSTATE_IN_GUEST
984 * R12 = (guest CR << 32) | interrupt vector
986 * guest R12 saved in shadow VCPU SCRATCH0
987 * guest R13 saved in SPRN_SCRATCH0
988 * guest R9 saved in HSTATE_SCRATCH2
990 /* We're now back in the host but in guest MMU context */
991 cmpwi r9,KVM_GUEST_MODE_HOST_HV
992 beq kvmppc_bad_host_intr
993 li r9, KVM_GUEST_MODE_HOST_HV
994 stb r9, HSTATE_IN_GUEST(r13)
996 ld r9, HSTATE_KVM_VCPU(r13)
1000 std r0, VCPU_GPR(R0)(r9)
1001 std r1, VCPU_GPR(R1)(r9)
1002 std r2, VCPU_GPR(R2)(r9)
1003 std r3, VCPU_GPR(R3)(r9)
1004 std r4, VCPU_GPR(R4)(r9)
1005 std r5, VCPU_GPR(R5)(r9)
1006 std r6, VCPU_GPR(R6)(r9)
1007 std r7, VCPU_GPR(R7)(r9)
1008 std r8, VCPU_GPR(R8)(r9)
1009 ld r0, HSTATE_SCRATCH2(r13)
1010 std r0, VCPU_GPR(R9)(r9)
1011 std r10, VCPU_GPR(R10)(r9)
1012 std r11, VCPU_GPR(R11)(r9)
1013 ld r3, HSTATE_SCRATCH0(r13)
1014 std r3, VCPU_GPR(R12)(r9)
1015 /* CR is in the high half of r12 */
1019 ld r3, HSTATE_CFAR(r13)
1020 std r3, VCPU_CFAR(r9)
1021 END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
1023 ld r4, HSTATE_PPR(r13)
1024 std r4, VCPU_PPR(r9)
1025 END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1027 /* Restore R1/R2 so we can handle faults */
1028 ld r1, HSTATE_HOST_R1(r13)
1031 mfspr r10, SPRN_SRR0
1032 mfspr r11, SPRN_SRR1
1033 std r10, VCPU_SRR0(r9)
1034 std r11, VCPU_SRR1(r9)
1035 /* trap is in the low half of r12, clear CR from the high half */
1037 andi. r0, r12, 2 /* need to read HSRR0/1? */
1039 mfspr r10, SPRN_HSRR0
1040 mfspr r11, SPRN_HSRR1
1042 1: std r10, VCPU_PC(r9)
1043 std r11, VCPU_MSR(r9)
1047 std r3, VCPU_GPR(R13)(r9)
1050 stw r12,VCPU_TRAP(r9)
1053 * Now that we have saved away SRR0/1 and HSRR0/1,
1054 * interrupts are recoverable in principle, so set MSR_RI.
1055 * This becomes important for relocation-on interrupts from
1056 * the guest, which we can get in radix mode on POWER9.
1061 #ifdef CONFIG_KVM_BOOK3S_HV_P8_TIMING
1062 addi r3, r9, VCPU_TB_RMINTR
1064 bl kvmhv_accumulate_time
1065 ld r5, VCPU_GPR(R5)(r9)
1066 ld r6, VCPU_GPR(R6)(r9)
1067 ld r7, VCPU_GPR(R7)(r9)
1068 ld r8, VCPU_GPR(R8)(r9)
1071 /* Save HEIR (HV emulation assist reg) in emul_inst
1072 if this is an HEI (HV emulation interrupt, e40) */
1073 li r3,KVM_INST_FETCH_FAILED
1074 std r3,VCPU_LAST_INST(r9)
1075 cmpwi r12,BOOK3S_INTERRUPT_H_EMUL_ASSIST
1078 11: std r3,VCPU_HEIR(r9)
1080 /* these are volatile across C function calls */
1083 std r3, VCPU_CTR(r9)
1084 std r4, VCPU_XER(r9)
1086 /* Save more register state */
1089 std r3, VCPU_DAR(r9)
1090 stw r4, VCPU_DSISR(r9)
1092 /* If this is a page table miss then see if it's theirs or ours */
1093 cmpwi r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
1095 std r3, VCPU_FAULT_DAR(r9)
1096 stw r4, VCPU_FAULT_DSISR(r9)
1097 cmpwi r12, BOOK3S_INTERRUPT_H_INST_STORAGE
1100 /* See if this is a leftover HDEC interrupt */
1101 cmpwi r12,BOOK3S_INTERRUPT_HV_DECREMENTER
1107 bge fast_guest_return
1109 /* See if this is an hcall we can handle in real mode */
1110 cmpwi r12,BOOK3S_INTERRUPT_SYSCALL
1111 beq hcall_try_real_mode
1113 /* Hypervisor doorbell - exit only if host IPI flag set */
1114 cmpwi r12, BOOK3S_INTERRUPT_H_DOORBELL
1116 lbz r0, HSTATE_HOST_IPI(r13)
1118 beq maybe_reenter_guest
1121 /* If it's a hypervisor facility unavailable interrupt, save HFSCR */
1122 cmpwi r12, BOOK3S_INTERRUPT_H_FAC_UNAVAIL
1124 mfspr r3, SPRN_HFSCR
1125 std r3, VCPU_HFSCR(r9)
1128 /* External interrupt ? */
1129 cmpwi r12, BOOK3S_INTERRUPT_EXTERNAL
1130 beq kvmppc_guest_external
1131 /* See if it is a machine check */
1132 cmpwi r12, BOOK3S_INTERRUPT_MACHINE_CHECK
1133 beq machine_check_realmode
1134 /* Or a hypervisor maintenance interrupt */
1135 cmpwi r12, BOOK3S_INTERRUPT_HMI
1138 guest_exit_cont: /* r9 = vcpu, r12 = trap, r13 = paca */
1140 #ifdef CONFIG_KVM_BOOK3S_HV_P8_TIMING
1141 addi r3, r9, VCPU_TB_RMEXIT
1143 bl kvmhv_accumulate_time
1147 * Possibly flush the link stack here, before we do a blr in
1148 * kvmhv_switch_to_host.
1151 patch_site 1b patch__call_kvm_flush_link_stack
1153 /* For hash guest, read the guest SLB and save it away */
1155 lwz r0,VCPU_SLB_NR(r9) /* number of entries in SLB */
1160 andis. r0,r8,SLB_ESID_V@h
1162 add r8,r8,r6 /* put index in */
1164 std r8,VCPU_SLB_E(r7)
1165 std r3,VCPU_SLB_V(r7)
1166 addi r7,r7,VCPU_SLB_SIZE
1170 /* Finally clear out the SLB */
1175 stw r5,VCPU_SLB_MAX(r9)
1177 /* load host SLB entries */
1178 ld r8,PACA_SLBSHADOWPTR(r13)
1180 .rept SLB_NUM_BOLTED
1181 li r3, SLBSHADOW_SAVEAREA
1185 andis. r7,r5,SLB_ESID_V@h
1192 stw r12, STACK_SLOT_TRAP(r1)
1195 /* Do this before kvmhv_commence_exit so we know TB is guest TB */
1196 ld r3, HSTATE_KVM_VCORE(r13)
1201 std r5,VCPU_DEC_EXPIRES(r9)
1203 /* Increment exit count, poke other threads to exit */
1205 bl kvmhv_commence_exit
1207 ld r9, HSTATE_KVM_VCPU(r13)
1209 /* Stop others sending VCPU interrupts to this physical CPU */
1211 stw r0, VCPU_CPU(r9)
1212 stw r0, VCPU_THREAD_CPU(r9)
1214 /* Save guest CTRL register, set runlatch to 1 if it was clear */
1216 stw r6,VCPU_CTRL(r9)
1223 * Save the guest PURR/SPURR
1228 ld r8,VCPU_SPURR(r9)
1229 std r5,VCPU_PURR(r9)
1230 std r6,VCPU_SPURR(r9)
1235 * Restore host PURR/SPURR and add guest times
1236 * so that the time in the guest gets accounted.
1238 ld r3,HSTATE_PURR(r13)
1239 ld r4,HSTATE_SPURR(r13)
1247 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
1248 /* Save POWER8-specific registers */
1252 std r5, VCPU_IAMR(r9)
1253 stw r6, VCPU_PSPB(r9)
1254 std r7, VCPU_FSCR(r9)
1258 std r7, VCPU_TAR(r9)
1259 mfspr r8, SPRN_EBBHR
1260 std r8, VCPU_EBBHR(r9)
1261 mfspr r5, SPRN_EBBRR
1262 mfspr r6, SPRN_BESCR
1265 std r5, VCPU_EBBRR(r9)
1266 std r6, VCPU_BESCR(r9)
1267 stw r7, VCPU_GUEST_PID(r9)
1268 std r8, VCPU_WORT(r9)
1269 mfspr r5, SPRN_TCSCR
1271 mfspr r7, SPRN_CSIGR
1273 std r5, VCPU_TCSCR(r9)
1274 std r6, VCPU_ACOP(r9)
1275 std r7, VCPU_CSIGR(r9)
1276 std r8, VCPU_TACR(r9)
1278 ld r5, STACK_SLOT_FSCR(r1)
1280 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1282 * Restore various registers to 0, where non-zero values
1283 * set by the guest could disrupt the host.
1288 mtspr SPRN_TCSCR, r0
1289 /* Set MMCRS to 1<<31 to freeze and disable the SPMC counters */
1292 mtspr SPRN_MMCRS, r0
1294 /* Save and restore AMR, IAMR and UAMOR before turning on the MMU */
1295 ld r8, STACK_SLOT_IAMR(r1)
1298 8: /* Power7 jumps back in here */
1302 std r6,VCPU_UAMOR(r9)
1303 ld r5,STACK_SLOT_AMR(r1)
1304 ld r6,STACK_SLOT_UAMOR(r1)
1306 mtspr SPRN_UAMOR, r6
1308 /* Switch DSCR back to host value */
1310 ld r7, HSTATE_DSCR(r13)
1311 std r8, VCPU_DSCR(r9)
1314 /* Save non-volatile GPRs */
1315 std r14, VCPU_GPR(R14)(r9)
1316 std r15, VCPU_GPR(R15)(r9)
1317 std r16, VCPU_GPR(R16)(r9)
1318 std r17, VCPU_GPR(R17)(r9)
1319 std r18, VCPU_GPR(R18)(r9)
1320 std r19, VCPU_GPR(R19)(r9)
1321 std r20, VCPU_GPR(R20)(r9)
1322 std r21, VCPU_GPR(R21)(r9)
1323 std r22, VCPU_GPR(R22)(r9)
1324 std r23, VCPU_GPR(R23)(r9)
1325 std r24, VCPU_GPR(R24)(r9)
1326 std r25, VCPU_GPR(R25)(r9)
1327 std r26, VCPU_GPR(R26)(r9)
1328 std r27, VCPU_GPR(R27)(r9)
1329 std r28, VCPU_GPR(R28)(r9)
1330 std r29, VCPU_GPR(R29)(r9)
1331 std r30, VCPU_GPR(R30)(r9)
1332 std r31, VCPU_GPR(R31)(r9)
1335 mfspr r3, SPRN_SPRG0
1336 mfspr r4, SPRN_SPRG1
1337 mfspr r5, SPRN_SPRG2
1338 mfspr r6, SPRN_SPRG3
1339 std r3, VCPU_SPRG0(r9)
1340 std r4, VCPU_SPRG1(r9)
1341 std r5, VCPU_SPRG2(r9)
1342 std r6, VCPU_SPRG3(r9)
1348 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1351 END_FTR_SECTION_IFCLR(CPU_FTR_TM)
1353 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
1357 li r5, 0 /* don't preserve non-vol regs */
1358 bl kvmppc_save_tm_hv
1360 ld r9, HSTATE_KVM_VCPU(r13)
1364 /* Increment yield count if they have a VPA */
1365 ld r8, VCPU_VPA(r9) /* do they have a VPA? */
1368 li r4, LPPACA_YIELDCOUNT
1373 stb r3, VCPU_VPA_DIRTY(r9)
1375 /* Save PMU registers if requested */
1376 /* r8 and cr0.eq are live here */
1379 beq 21f /* if no VPA, save PMU stuff anyway */
1380 lbz r4, LPPACA_PMCINUSE(r8)
1381 21: bl kvmhv_save_guest_pmu
1382 ld r9, HSTATE_KVM_VCPU(r13)
1384 /* Restore host values of some registers */
1386 ld r5, STACK_SLOT_CIABR(r1)
1387 ld r6, STACK_SLOT_DAWR0(r1)
1388 ld r7, STACK_SLOT_DAWRX0(r1)
1389 mtspr SPRN_CIABR, r5
1391 * If the DAWR doesn't work, it's ok to write these here as
1392 * this value should always be zero
1394 mtspr SPRN_DAWR0, r6
1395 mtspr SPRN_DAWRX0, r7
1396 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1399 * POWER7/POWER8 guest -> host partition switch code.
1400 * We don't have to lock against tlbies but we do
1401 * have to coordinate the hardware threads.
1402 * Here STACK_SLOT_TRAP(r1) contains the trap number.
1404 kvmhv_switch_to_host:
1405 /* Secondary threads wait for primary to do partition switch */
1406 ld r5,HSTATE_KVM_VCORE(r13)
1407 ld r4,VCORE_KVM(r5) /* pointer to struct kvm */
1408 lbz r3,HSTATE_PTID(r13)
1412 13: lbz r3,VCORE_IN_GUEST(r5)
1418 /* Primary thread waits for all the secondaries to exit guest */
1419 15: lwz r3,VCORE_ENTRY_EXIT(r5)
1420 rlwinm r0,r3,32-8,0xff
1426 /* Did we actually switch to the guest at all? */
1427 lbz r6, VCORE_IN_GUEST(r5)
1431 /* Primary thread switches back to host partition */
1432 lwz r7,KVM_HOST_LPID(r4)
1433 ld r6,KVM_HOST_SDR1(r4)
1434 li r8,LPID_RSVD /* switch to reserved LPID */
1437 mtspr SPRN_SDR1,r6 /* switch to host page table */
1442 /* DPDES and VTB are shared between threads */
1443 mfspr r7, SPRN_DPDES
1445 std r7, VCORE_DPDES(r5)
1446 std r8, VCORE_VTB(r5)
1447 /* clear DPDES so we don't get guest doorbells in the host */
1449 mtspr SPRN_DPDES, r8
1450 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1452 /* Subtract timebase offset from timebase */
1453 ld r8, VCORE_TB_OFFSET_APPL(r5)
1457 std r0, VCORE_TB_OFFSET_APPL(r5)
1458 mftb r6 /* current guest timebase */
1460 mtspr SPRN_TBU40,r8 /* update upper 40 bits */
1461 mftb r7 /* check if lower 24 bits overflowed */
1466 addis r8,r8,0x100 /* if so, increment upper 40 bits */
1471 * If this is an HMI, we called kvmppc_realmode_hmi_handler
1472 * above, which may or may not have already called
1473 * kvmppc_subcore_exit_guest. Fortunately, all that
1474 * kvmppc_subcore_exit_guest does is clear a flag, so calling
1475 * it again here is benign even if kvmppc_realmode_hmi_handler
1476 * has already called it.
1478 bl kvmppc_subcore_exit_guest
1480 30: ld r5,HSTATE_KVM_VCORE(r13)
1481 ld r4,VCORE_KVM(r5) /* pointer to struct kvm */
1484 ld r0, VCORE_PCR(r5)
1485 LOAD_REG_IMMEDIATE(r6, PCR_MASK)
1490 /* Signal secondary CPUs to continue */
1492 stb r0,VCORE_IN_GUEST(r5)
1493 19: lis r8,0x7fff /* MAX_INT@h */
1496 16: ld r8,KVM_HOST_LPCR(r4)
1500 #ifdef CONFIG_KVM_BOOK3S_HV_P8_TIMING
1501 /* Finish timing, if we have a vcpu */
1502 ld r4, HSTATE_KVM_VCPU(r13)
1506 bl kvmhv_accumulate_time
1509 /* Unset guest mode */
1510 li r0, KVM_GUEST_MODE_NONE
1511 stb r0, HSTATE_IN_GUEST(r13)
1513 lwz r12, STACK_SLOT_TRAP(r1) /* return trap # in r12 */
1514 ld r0, SFS+PPC_LR_STKOFF(r1)
1520 .global kvm_flush_link_stack
1521 kvm_flush_link_stack:
1522 /* Save LR into r0 */
1525 /* Flush the link stack. On Power8 it's up to 32 entries in size. */
1527 ANNOTATE_INTRA_FUNCTION_CALL
1531 /* And on Power9 it's up to 64. */
1534 ANNOTATE_INTRA_FUNCTION_CALL
1537 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1543 kvmppc_guest_external:
1544 /* External interrupt, first check for host_ipi. If this is
1545 * set, we know the host wants us out so let's do it now
1547 bl CFUNC(kvmppc_read_intr)
1550 * Restore the active volatile registers after returning from
1553 ld r9, HSTATE_KVM_VCPU(r13)
1554 li r12, BOOK3S_INTERRUPT_EXTERNAL
1557 * kvmppc_read_intr return codes:
1559 * Exit to host (r3 > 0)
1560 * 1 An interrupt is pending that needs to be handled by the host
1561 * Exit guest and return to host by branching to guest_exit_cont
1563 * 2 Passthrough that needs completion in the host
1564 * Exit guest and return to host by branching to guest_exit_cont
1565 * However, we also set r12 to BOOK3S_INTERRUPT_HV_RM_HARD
1566 * to indicate to the host to complete handling the interrupt
1568 * Before returning to guest, we check if any CPU is heading out
1569 * to the host and if so, we head out also. If no CPUs are heading
1570 * check return values <= 0.
1572 * Return to guest (r3 <= 0)
1573 * 0 No external interrupt is pending
1574 * -1 A guest wakeup IPI (which has now been cleared)
1575 * In either case, we return to guest to deliver any pending
1578 * -2 A PCI passthrough external interrupt was handled
1579 * (interrupt was delivered directly to guest)
1580 * Return to guest to deliver any pending guest interrupts.
1586 /* Return code = 2 */
1587 li r12, BOOK3S_INTERRUPT_HV_RM_HARD
1588 stw r12, VCPU_TRAP(r9)
1591 1: /* Return code <= 1 */
1595 /* Return code <= 0 */
1596 maybe_reenter_guest:
1597 ld r5, HSTATE_KVM_VCORE(r13)
1598 lwz r0, VCORE_ENTRY_EXIT(r5)
1601 blt deliver_guest_interrupt
1605 * Check whether an HDSI is an HPTE not found fault or something else.
1606 * If it is an HPTE not found fault that is due to the guest accessing
1607 * a page that they have mapped but which we have paged out, then
1608 * we continue on with the guest exit path. In all other cases,
1609 * reflect the HDSI to the guest as a DSI.
1613 mfspr r6, SPRN_HDSISR
1614 /* HPTE not found fault or protection fault? */
1615 andis. r0, r6, (DSISR_NOHPTE | DSISR_PROTFAULT)@h
1616 beq 1f /* if not, send it to the guest */
1617 andi. r0, r11, MSR_DR /* data relocation enabled? */
1620 PPC_SLBFEE_DOT(R5, R0) /* if so, look up SLB */
1621 li r0, BOOK3S_INTERRUPT_DATA_SEGMENT
1622 bne 7f /* if no SLB entry found */
1623 4: std r4, VCPU_FAULT_DAR(r9)
1624 stw r6, VCPU_FAULT_DSISR(r9)
1626 /* Search the hash table. */
1627 mr r3, r9 /* vcpu pointer */
1628 li r7, 1 /* data fault */
1629 bl CFUNC(kvmppc_hpte_hv_fault)
1630 ld r9, HSTATE_KVM_VCPU(r13)
1632 ld r11, VCPU_MSR(r9)
1633 li r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
1634 cmpdi r3, 0 /* retry the instruction */
1636 cmpdi r3, -1 /* handle in kernel mode */
1638 cmpdi r3, -2 /* MMIO emulation; need instr word */
1641 /* Synthesize a DSI (or DSegI) for the guest */
1642 ld r4, VCPU_FAULT_DAR(r9)
1644 1: li r0, BOOK3S_INTERRUPT_DATA_STORAGE
1645 mtspr SPRN_DSISR, r6
1646 7: mtspr SPRN_DAR, r4
1647 mtspr SPRN_SRR0, r10
1648 mtspr SPRN_SRR1, r11
1650 bl kvmppc_msr_interrupt
1651 fast_interrupt_c_return:
1652 6: ld r7, VCPU_CTR(r9)
1659 3: ld r5, VCPU_KVM(r9) /* not relocated, use VRMA */
1660 ld r5, KVM_VRMA_SLB_V(r5)
1663 /* If this is for emulated MMIO, load the instruction word */
1664 2: li r8, KVM_INST_FETCH_FAILED /* In case lwz faults */
1666 /* Set guest mode to 'jump over instruction' so if lwz faults
1667 * we'll just continue at the next IP. */
1668 li r0, KVM_GUEST_MODE_SKIP
1669 stb r0, HSTATE_IN_GUEST(r13)
1671 /* Do the access with MSR:DR enabled */
1673 ori r4, r3, MSR_DR /* Enable paging for data */
1678 /* Store the result */
1679 std r8, VCPU_LAST_INST(r9)
1681 /* Unset guest mode. */
1682 li r0, KVM_GUEST_MODE_HOST_HV
1683 stb r0, HSTATE_IN_GUEST(r13)
1687 * Similarly for an HISI, reflect it to the guest as an ISI unless
1688 * it is an HPTE not found fault for a page that we have paged out.
1691 andis. r0, r11, SRR1_ISI_NOPT@h
1693 andi. r0, r11, MSR_IR /* instruction relocation enabled? */
1696 PPC_SLBFEE_DOT(R5, R0) /* if so, look up SLB */
1697 li r0, BOOK3S_INTERRUPT_INST_SEGMENT
1698 bne 7f /* if no SLB entry found */
1700 /* Search the hash table. */
1701 mr r3, r9 /* vcpu pointer */
1704 li r7, 0 /* instruction fault */
1705 bl CFUNC(kvmppc_hpte_hv_fault)
1706 ld r9, HSTATE_KVM_VCPU(r13)
1708 ld r11, VCPU_MSR(r9)
1709 li r12, BOOK3S_INTERRUPT_H_INST_STORAGE
1710 cmpdi r3, 0 /* retry the instruction */
1711 beq fast_interrupt_c_return
1712 cmpdi r3, -1 /* handle in kernel mode */
1715 /* Synthesize an ISI (or ISegI) for the guest */
1717 1: li r0, BOOK3S_INTERRUPT_INST_STORAGE
1718 7: mtspr SPRN_SRR0, r10
1719 mtspr SPRN_SRR1, r11
1721 bl kvmppc_msr_interrupt
1722 b fast_interrupt_c_return
1724 3: ld r6, VCPU_KVM(r9) /* not relocated, use VRMA */
1725 ld r5, KVM_VRMA_SLB_V(r6)
1729 * Try to handle an hcall in real mode.
1730 * Returns to the guest if we handle it, or continues on up to
1731 * the kernel if we can't (i.e. if we don't have a handler for
1732 * it, or if the handler returns H_TOO_HARD).
1734 * r5 - r8 contain hcall args,
1735 * r9 = vcpu, r10 = pc, r11 = msr, r12 = trap, r13 = paca
1737 hcall_try_real_mode:
1738 ld r3,VCPU_GPR(R3)(r9)
1740 /* sc 1 from userspace - reflect to guest syscall */
1741 bne sc_1_fast_return
1743 cmpldi r3,hcall_real_table_end - hcall_real_table
1745 /* See if this hcall is enabled for in-kernel handling */
1747 srdi r0, r3, 8 /* r0 = (r3 / 4) >> 6 */
1748 sldi r0, r0, 3 /* index into kvm->arch.enabled_hcalls[] */
1750 ld r0, KVM_ENABLED_HCALLS(r4)
1751 rlwinm r4, r3, 32-2, 0x3f /* r4 = (r3 / 4) & 0x3f */
1755 /* Get pointer to handler, if any, and call it */
1756 LOAD_REG_ADDR(r4, hcall_real_table)
1762 mr r3,r9 /* get vcpu pointer */
1763 ld r4,VCPU_GPR(R4)(r9)
1766 beq hcall_real_fallback
1767 ld r4,HSTATE_KVM_VCPU(r13)
1768 std r3,VCPU_GPR(R3)(r4)
1776 li r10, BOOK3S_INTERRUPT_SYSCALL
1777 bl kvmppc_msr_interrupt
1781 /* We've attempted a real mode hcall, but it's punted it back
1782 * to userspace. We need to restore some clobbered volatiles
1783 * before resuming the pass-it-to-qemu path */
1784 hcall_real_fallback:
1785 li r12,BOOK3S_INTERRUPT_SYSCALL
1786 ld r9, HSTATE_KVM_VCPU(r13)
1790 .globl hcall_real_table
1792 .long 0 /* 0 - unused */
1793 .long DOTSYM(kvmppc_h_remove) - hcall_real_table
1794 .long DOTSYM(kvmppc_h_enter) - hcall_real_table
1795 .long DOTSYM(kvmppc_h_read) - hcall_real_table
1796 .long DOTSYM(kvmppc_h_clear_mod) - hcall_real_table
1797 .long DOTSYM(kvmppc_h_clear_ref) - hcall_real_table
1798 .long DOTSYM(kvmppc_h_protect) - hcall_real_table
1801 .long 0 /* 0x24 - H_SET_SPRG0 */
1802 .long DOTSYM(kvmppc_h_set_dabr) - hcall_real_table
1803 .long DOTSYM(kvmppc_rm_h_page_init) - hcall_real_table
1817 #ifdef CONFIG_KVM_XICS
1818 .long DOTSYM(xics_rm_h_eoi) - hcall_real_table
1819 .long DOTSYM(xics_rm_h_cppr) - hcall_real_table
1820 .long DOTSYM(xics_rm_h_ipi) - hcall_real_table
1821 .long 0 /* 0x70 - H_IPOLL */
1822 .long DOTSYM(xics_rm_h_xirr) - hcall_real_table
1824 .long 0 /* 0x64 - H_EOI */
1825 .long 0 /* 0x68 - H_CPPR */
1826 .long 0 /* 0x6c - H_IPI */
1827 .long 0 /* 0x70 - H_IPOLL */
1828 .long 0 /* 0x74 - H_XIRR */
1856 .long DOTSYM(kvmppc_h_cede) - hcall_real_table
1857 .long DOTSYM(kvmppc_rm_h_confer) - hcall_real_table
1873 .long DOTSYM(kvmppc_h_bulk_remove) - hcall_real_table
1877 .long DOTSYM(kvmppc_h_set_xdabr) - hcall_real_table
1991 #ifdef CONFIG_KVM_XICS
1992 .long DOTSYM(xics_rm_h_xirr_x) - hcall_real_table
1994 .long 0 /* 0x2fc - H_XIRR_X*/
1996 .long DOTSYM(kvmppc_rm_h_random) - hcall_real_table
1997 .globl hcall_real_table_end
1998 hcall_real_table_end:
2000 _GLOBAL_TOC(kvmppc_h_set_xdabr)
2001 EXPORT_SYMBOL_GPL(kvmppc_h_set_xdabr)
2002 andi. r0, r5, DABRX_USER | DABRX_KERNEL
2004 li r0, DABRX_USER | DABRX_KERNEL | DABRX_BTI
2007 6: li r3, H_PARAMETER
2010 _GLOBAL_TOC(kvmppc_h_set_dabr)
2011 EXPORT_SYMBOL_GPL(kvmppc_h_set_dabr)
2012 li r5, DABRX_USER | DABRX_KERNEL
2016 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2017 std r4,VCPU_DABR(r3)
2018 stw r5, VCPU_DABRX(r3)
2019 mtspr SPRN_DABRX, r5
2020 /* Work around P7 bug where DABR can get corrupted on mtspr */
2021 1: mtspr SPRN_DABR,r4
2030 LOAD_REG_ADDR(r11, dawr_force_enable)
2037 /* Emulate H_SET_DABR/X on P8 for the sake of compat mode guests */
2038 rlwimi r5, r4, 5, DAWRX_DR | DAWRX_DW
2039 rlwimi r5, r4, 2, DAWRX_WT
2041 std r4, VCPU_DAWR0(r3)
2042 std r5, VCPU_DAWRX0(r3)
2044 * If came in through the real mode hcall handler then it is necessary
2045 * to write the registers since the return path won't. Otherwise it is
2046 * sufficient to store then in the vcpu struct as they will be loaded
2047 * next time the vcpu is run.
2050 andi. r6, r6, MSR_DR /* in real mode? */
2052 mtspr SPRN_DAWR0, r4
2053 mtspr SPRN_DAWRX0, r5
2057 _GLOBAL(kvmppc_h_cede) /* r3 = vcpu pointer, r11 = msr, r13 = paca */
2059 std r11,VCPU_MSR(r3)
2061 stb r0,VCPU_CEDED(r3)
2062 sync /* order setting ceded vs. testing prodded */
2063 lbz r5,VCPU_PRODDED(r3)
2065 bne kvm_cede_prodded
2066 li r12,0 /* set trap to 0 to say hcall is handled */
2067 stw r12,VCPU_TRAP(r3)
2069 std r0,VCPU_GPR(R3)(r3)
2072 * Set our bit in the bitmask of napping threads unless all the
2073 * other threads are already napping, in which case we send this
2076 ld r5,HSTATE_KVM_VCORE(r13)
2077 lbz r6,HSTATE_PTID(r13)
2078 lwz r8,VCORE_ENTRY_EXIT(r5)
2082 addi r6,r5,VCORE_NAPPING_THREADS
2089 /* order napping_threads update vs testing entry_exit_map */
2092 stb r0,HSTATE_NAPPING(r13)
2093 lwz r7,VCORE_ENTRY_EXIT(r5)
2095 bge 33f /* another thread already exiting */
2098 * Although not specifically required by the architecture, POWER7
2099 * preserves the following registers in nap mode, even if an SMT mode
2100 * switch occurs: SLB entries, PURR, SPURR, AMOR, UAMOR, AMR, SPRG0-3,
2101 * DAR, DSISR, DABR, DABRX, DSCR, PMCx, MMCRx, SIAR, SDAR.
2103 /* Save non-volatile GPRs */
2104 std r14, VCPU_GPR(R14)(r3)
2105 std r15, VCPU_GPR(R15)(r3)
2106 std r16, VCPU_GPR(R16)(r3)
2107 std r17, VCPU_GPR(R17)(r3)
2108 std r18, VCPU_GPR(R18)(r3)
2109 std r19, VCPU_GPR(R19)(r3)
2110 std r20, VCPU_GPR(R20)(r3)
2111 std r21, VCPU_GPR(R21)(r3)
2112 std r22, VCPU_GPR(R22)(r3)
2113 std r23, VCPU_GPR(R23)(r3)
2114 std r24, VCPU_GPR(R24)(r3)
2115 std r25, VCPU_GPR(R25)(r3)
2116 std r26, VCPU_GPR(R26)(r3)
2117 std r27, VCPU_GPR(R27)(r3)
2118 std r28, VCPU_GPR(R28)(r3)
2119 std r29, VCPU_GPR(R29)(r3)
2120 std r30, VCPU_GPR(R30)(r3)
2121 std r31, VCPU_GPR(R31)(r3)
2126 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2129 END_FTR_SECTION_IFCLR(CPU_FTR_TM)
2131 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
2133 ld r3, HSTATE_KVM_VCPU(r13)
2135 li r5, 0 /* don't preserve non-vol regs */
2136 bl kvmppc_save_tm_hv
2142 * Set DEC to the smaller of DEC and HDEC, so that we wake
2143 * no later than the end of our timeslice (HDEC interrupts
2144 * don't wake us from nap).
2155 /* save expiry time of guest decrementer */
2157 ld r4, HSTATE_KVM_VCPU(r13)
2158 std r3, VCPU_DEC_EXPIRES(r4)
2160 #ifdef CONFIG_KVM_BOOK3S_HV_P8_TIMING
2161 ld r4, HSTATE_KVM_VCPU(r13)
2162 addi r3, r4, VCPU_TB_CEDE
2163 bl kvmhv_accumulate_time
2166 lis r3, LPCR_PECEDP@h /* Do wake on privileged doorbell */
2168 /* Go back to host stack */
2169 ld r1, HSTATE_HOST_R1(r13)
2172 * Take a nap until a decrementer or external or doobell interrupt
2173 * occurs, with PECE1 and PECE0 set in LPCR.
2174 * On POWER8, set PECEDH, and if we are ceding, also set PECEDP.
2175 * Also clear the runlatch bit before napping.
2179 mtspr SPRN_CTRLT, r0
2182 stb r0,HSTATE_HWTHREAD_REQ(r13)
2184 ori r5,r5,LPCR_PECE0 | LPCR_PECE1
2186 ori r5, r5, LPCR_PECEDH
2187 rlwimi r5, r3, 0, LPCR_PECEDP
2188 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2190 kvm_nap_sequence: /* desired LPCR value in r5 */
2191 li r3, PNV_THREAD_NAP
2195 bl isa206_idle_insn_mayloss
2198 mtspr SPRN_CTRLT, r0
2203 stb r0, PACA_FTRACE_ENABLED(r13)
2205 li r0, KVM_HWTHREAD_IN_KVM
2206 stb r0, HSTATE_HWTHREAD_STATE(r13)
2208 lbz r0, HSTATE_NAPPING(r13)
2209 cmpwi r0, NAPPING_CEDE
2211 cmpwi r0, NAPPING_NOVCPU
2212 beq kvm_novcpu_wakeup
2213 cmpwi r0, NAPPING_UNSPLIT
2214 beq kvm_unsplit_wakeup
2215 twi 31,0,0 /* Nap state must not be zero */
2223 /* Woken by external or decrementer interrupt */
2225 /* get vcpu pointer */
2226 ld r4, HSTATE_KVM_VCPU(r13)
2228 #ifdef CONFIG_KVM_BOOK3S_HV_P8_TIMING
2229 addi r3, r4, VCPU_TB_RMINTR
2230 bl kvmhv_accumulate_time
2233 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2236 END_FTR_SECTION_IFCLR(CPU_FTR_TM)
2238 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
2242 li r5, 0 /* don't preserve non-vol regs */
2243 bl kvmppc_restore_tm_hv
2245 ld r4, HSTATE_KVM_VCPU(r13)
2249 /* load up FP state */
2252 /* Restore guest decrementer */
2253 ld r3, VCPU_DEC_EXPIRES(r4)
2259 ld r14, VCPU_GPR(R14)(r4)
2260 ld r15, VCPU_GPR(R15)(r4)
2261 ld r16, VCPU_GPR(R16)(r4)
2262 ld r17, VCPU_GPR(R17)(r4)
2263 ld r18, VCPU_GPR(R18)(r4)
2264 ld r19, VCPU_GPR(R19)(r4)
2265 ld r20, VCPU_GPR(R20)(r4)
2266 ld r21, VCPU_GPR(R21)(r4)
2267 ld r22, VCPU_GPR(R22)(r4)
2268 ld r23, VCPU_GPR(R23)(r4)
2269 ld r24, VCPU_GPR(R24)(r4)
2270 ld r25, VCPU_GPR(R25)(r4)
2271 ld r26, VCPU_GPR(R26)(r4)
2272 ld r27, VCPU_GPR(R27)(r4)
2273 ld r28, VCPU_GPR(R28)(r4)
2274 ld r29, VCPU_GPR(R29)(r4)
2275 ld r30, VCPU_GPR(R30)(r4)
2276 ld r31, VCPU_GPR(R31)(r4)
2278 /* Check the wake reason in SRR1 to see why we got here */
2279 bl kvmppc_check_wake_reason
2282 * Restore volatile registers since we could have called a
2283 * C routine in kvmppc_check_wake_reason
2285 * r3 tells us whether we need to return to host or not
2286 * WARNING: it gets checked further down:
2287 * should not modify r3 until this check is done.
2289 ld r4, HSTATE_KVM_VCPU(r13)
2291 /* clear our bit in vcore->napping_threads */
2292 34: ld r5,HSTATE_KVM_VCORE(r13)
2293 lbz r7,HSTATE_PTID(r13)
2296 addi r6,r5,VCORE_NAPPING_THREADS
2302 stb r0,HSTATE_NAPPING(r13)
2304 /* See if the wake reason saved in r3 means we need to exit */
2305 stw r12, VCPU_TRAP(r4)
2309 b maybe_reenter_guest
2311 /* cede when already previously prodded case */
2314 stb r0,VCPU_PRODDED(r3)
2315 sync /* order testing prodded vs. clearing ceded */
2316 stb r0,VCPU_CEDED(r3)
2320 /* we've ceded but we want to give control to the host */
2322 ld r9, HSTATE_KVM_VCPU(r13)
2325 /* Try to do machine check recovery in real mode */
2326 machine_check_realmode:
2327 mr r3, r9 /* get vcpu pointer */
2328 bl kvmppc_realmode_machine_check
2330 /* all machine checks go to virtual mode for further handling */
2331 ld r9, HSTATE_KVM_VCPU(r13)
2332 li r12, BOOK3S_INTERRUPT_MACHINE_CHECK
2336 * Call C code to handle a HMI in real mode.
2337 * Only the primary thread does the call, secondary threads are handled
2338 * by calling hmi_exception_realmode() after kvmppc_hv_entry returns.
2339 * r9 points to the vcpu on entry
2342 lbz r0, HSTATE_PTID(r13)
2345 bl CFUNC(kvmppc_realmode_hmi_handler)
2346 ld r9, HSTATE_KVM_VCPU(r13)
2347 li r12, BOOK3S_INTERRUPT_HMI
2351 * Check the reason we woke from nap, and take appropriate action.
2353 * 0 if nothing needs to be done
2354 * 1 if something happened that needs to be handled by the host
2355 * -1 if there was a guest wakeup (IPI or msgsnd)
2356 * -2 if we handled a PCI passthrough interrupt (returned by
2357 * kvmppc_read_intr only)
2359 * Also sets r12 to the interrupt vector for any interrupt that needs
2360 * to be handled now by the host (0x500 for external interrupt), or zero.
2361 * Modifies all volatile registers (since it may call a C function).
2362 * This routine calls kvmppc_read_intr, a C function, if an external
2363 * interrupt is pending.
2365 SYM_FUNC_START_LOCAL(kvmppc_check_wake_reason)
2368 rlwinm r6, r6, 45-31, 0xf /* extract wake reason field (P8) */
2370 rlwinm r6, r6, 45-31, 0xe /* P7 wake reason field is 3 bits */
2371 ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_207S)
2372 cmpwi r6, 8 /* was it an external interrupt? */
2373 beq 7f /* if so, see what it was */
2376 cmpwi r6, 6 /* was it the decrementer? */
2379 cmpwi r6, 5 /* privileged doorbell? */
2381 cmpwi r6, 3 /* hypervisor doorbell? */
2383 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2384 cmpwi r6, 0xa /* Hypervisor maintenance ? */
2386 li r3, 1 /* anything else, return 1 */
2389 /* hypervisor doorbell */
2390 3: li r12, BOOK3S_INTERRUPT_H_DOORBELL
2393 * Clear the doorbell as we will invoke the handler
2394 * explicitly in the guest exit path.
2396 lis r6, (PPC_DBELL_SERVER << (63-36))@h
2398 /* see if it's a host IPI */
2400 lbz r0, HSTATE_HOST_IPI(r13)
2403 /* if not, return -1 */
2407 /* Woken up due to Hypervisor maintenance interrupt */
2408 4: li r12, BOOK3S_INTERRUPT_HMI
2412 /* external interrupt - create a stack frame so we can call C */
2414 std r0, PPC_LR_STKOFF(r1)
2415 stdu r1, -PPC_MIN_STKFRM(r1)
2416 bl CFUNC(kvmppc_read_intr)
2418 li r12, BOOK3S_INTERRUPT_EXTERNAL
2423 * Return code of 2 means PCI passthrough interrupt, but
2424 * we need to return back to host to complete handling the
2425 * interrupt. Trap reason is expected in r12 by guest
2428 li r12, BOOK3S_INTERRUPT_HV_RM_HARD
2430 ld r0, PPC_MIN_STKFRM+PPC_LR_STKOFF(r1)
2431 addi r1, r1, PPC_MIN_STKFRM
2434 SYM_FUNC_END(kvmppc_check_wake_reason)
2437 * Save away FP, VMX and VSX registers.
2439 * N.B. r30 and r31 are volatile across this function,
2440 * thus it is not callable from C.
2442 SYM_FUNC_START_LOCAL(kvmppc_save_fp)
2447 #ifdef CONFIG_ALTIVEC
2449 oris r8,r8,MSR_VEC@h
2450 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
2454 oris r8,r8,MSR_VSX@h
2455 END_FTR_SECTION_IFSET(CPU_FTR_VSX)
2458 addi r3,r3,VCPU_FPRS
2460 #ifdef CONFIG_ALTIVEC
2462 addi r3,r31,VCPU_VRS
2464 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
2466 mfspr r6,SPRN_VRSAVE
2467 stw r6,VCPU_VRSAVE(r31)
2470 SYM_FUNC_END(kvmppc_save_fp)
2473 * Load up FP, VMX and VSX registers
2475 * N.B. r30 and r31 are volatile across this function,
2476 * thus it is not callable from C.
2478 SYM_FUNC_START_LOCAL(kvmppc_load_fp)
2483 #ifdef CONFIG_ALTIVEC
2485 oris r8,r8,MSR_VEC@h
2486 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
2490 oris r8,r8,MSR_VSX@h
2491 END_FTR_SECTION_IFSET(CPU_FTR_VSX)
2494 addi r3,r4,VCPU_FPRS
2496 #ifdef CONFIG_ALTIVEC
2498 addi r3,r31,VCPU_VRS
2500 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
2502 lwz r7,VCPU_VRSAVE(r31)
2503 mtspr SPRN_VRSAVE,r7
2507 SYM_FUNC_END(kvmppc_load_fp)
2509 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2511 * Save transactional state and TM-related registers.
2512 * Called with r3 pointing to the vcpu struct and r4 containing
2513 * the guest MSR value.
2514 * r5 is non-zero iff non-volatile register state needs to be maintained.
2515 * If r5 == 0, this can modify all checkpointed registers, but
2516 * restores r1 and r2 before exit.
2518 _GLOBAL_TOC(kvmppc_save_tm_hv)
2519 EXPORT_SYMBOL_GPL(kvmppc_save_tm_hv)
2520 /* See if we need to handle fake suspend mode */
2523 END_FTR_SECTION_IFCLR(CPU_FTR_P9_TM_HV_ASSIST)
2525 lbz r0, HSTATE_FAKE_SUSPEND(r13) /* Were we fake suspended? */
2527 beq __kvmppc_save_tm
2529 /* The following code handles the fake_suspend = 1 case */
2531 std r0, PPC_LR_STKOFF(r1)
2532 stdu r1, -TM_FRAME_SIZE(r1)
2537 rldimi r8, r0, MSR_TM_LG, 63-MSR_TM_LG
2540 rldicl. r8, r8, 64 - MSR_TS_S_LG, 62 /* Did we actually hrfid? */
2543 bl pnv_power9_force_smt4_catch
2544 END_FTR_SECTION_IFSET(CPU_FTR_P9_TM_XER_SO_BUG)
2548 * It's possible that treclaim. may modify registers, if we have lost
2549 * track of fake-suspend state in the guest due to it using rfscv.
2550 * Save and restore registers in case this occurs.
2555 /* SPRN_TAR would need to be saved here if the kernel ever used it */
2563 std r1, HSTATE_HOST_R1(r13)
2565 /* We have to treclaim here because that's the only way to do S->N */
2566 li r3, TM_CAUSE_KVM_RESCHED
2570 ld r1, HSTATE_HOST_R1(r13)
2584 * We were in fake suspend, so we are not going to save the
2585 * register state as the guest checkpointed state (since
2586 * we already have it), therefore we can now use any volatile GPR.
2587 * In fact treclaim in fake suspend state doesn't modify
2592 bl pnv_power9_force_smt4_release
2593 END_FTR_SECTION_IFSET(CPU_FTR_P9_TM_XER_SO_BUG)
2597 mfspr r3, SPRN_PSSCR
2598 /* PSSCR_FAKE_SUSPEND is a write-only bit, but clear it anyway */
2599 li r0, PSSCR_FAKE_SUSPEND
2601 mtspr SPRN_PSSCR, r3
2603 /* Don't save TEXASR, use value from last exit in real suspend state */
2604 ld r9, HSTATE_KVM_VCPU(r13)
2605 mfspr r5, SPRN_TFHAR
2606 mfspr r6, SPRN_TFIAR
2607 std r5, VCPU_TFHAR(r9)
2608 std r6, VCPU_TFIAR(r9)
2610 addi r1, r1, TM_FRAME_SIZE
2611 ld r0, PPC_LR_STKOFF(r1)
2616 * Restore transactional state and TM-related registers.
2617 * Called with r3 pointing to the vcpu struct
2618 * and r4 containing the guest MSR value.
2619 * r5 is non-zero iff non-volatile register state needs to be maintained.
2620 * This potentially modifies all checkpointed registers.
2621 * It restores r1 and r2 from the PACA.
2623 _GLOBAL_TOC(kvmppc_restore_tm_hv)
2624 EXPORT_SYMBOL_GPL(kvmppc_restore_tm_hv)
2626 * If we are doing TM emulation for the guest on a POWER9 DD2,
2627 * then we don't actually do a trechkpt -- we either set up
2628 * fake-suspend mode, or emulate a TM rollback.
2631 b __kvmppc_restore_tm
2632 END_FTR_SECTION_IFCLR(CPU_FTR_P9_TM_HV_ASSIST)
2634 std r0, PPC_LR_STKOFF(r1)
2637 stb r0, HSTATE_FAKE_SUSPEND(r13)
2639 /* Turn on TM so we can restore TM SPRs */
2642 rldimi r5, r0, MSR_TM_LG, 63-MSR_TM_LG
2646 * The user may change these outside of a transaction, so they must
2647 * always be context switched.
2649 ld r5, VCPU_TFHAR(r3)
2650 ld r6, VCPU_TFIAR(r3)
2651 ld r7, VCPU_TEXASR(r3)
2652 mtspr SPRN_TFHAR, r5
2653 mtspr SPRN_TFIAR, r6
2654 mtspr SPRN_TEXASR, r7
2656 rldicl. r5, r4, 64 - MSR_TS_S_LG, 62
2657 beqlr /* TM not active in guest */
2659 /* Make sure the failure summary is set */
2660 oris r7, r7, (TEXASR_FS)@h
2661 mtspr SPRN_TEXASR, r7
2663 cmpwi r5, 1 /* check for suspended state */
2665 stb r5, HSTATE_FAKE_SUSPEND(r13)
2666 b 9f /* and return */
2667 10: stdu r1, -PPC_MIN_STKFRM(r1)
2668 /* guest is in transactional state, so simulate rollback */
2669 bl kvmhv_emulate_tm_rollback
2671 addi r1, r1, PPC_MIN_STKFRM
2672 9: ld r0, PPC_LR_STKOFF(r1)
2675 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
2678 * We come here if we get any exception or interrupt while we are
2679 * executing host real mode code while in guest MMU context.
2680 * r12 is (CR << 32) | vector
2681 * r13 points to our PACA
2682 * r12 is saved in HSTATE_SCRATCH0(r13)
2683 * r9 is saved in HSTATE_SCRATCH2(r13)
2684 * r13 is saved in HSPRG1
2685 * cfar is saved in HSTATE_CFAR(r13)
2686 * ppr is saved in HSTATE_PPR(r13)
2688 kvmppc_bad_host_intr:
2690 * Switch to the emergency stack, but start half-way down in
2691 * case we were already on it.
2695 ld r1, PACAEMERGSP(r13)
2696 subi r1, r1, THREAD_SIZE/2 + INT_FRAME_SIZE
2708 mfspr r3, SPRN_HSRR0
2709 mfspr r4, SPRN_HSRR1
2711 mfspr r6, SPRN_HDSISR
2713 1: mfspr r3, SPRN_SRR0
2716 mfspr r6, SPRN_DSISR
2721 ld r9, HSTATE_SCRATCH2(r13)
2722 ld r12, HSTATE_SCRATCH0(r13)
2724 SAVE_GPRS(9, 12, r1)
2727 ld r5, HSTATE_CFAR(r13)
2728 std r5, ORIG_GPR3(r1)
2732 lbz r6, PACAIRQSOFTMASK(r13)
2738 LOAD_REG_IMMEDIATE(3, STACK_FRAME_REGS_MARKER)
2739 std r3, STACK_INT_FRAME_MARKER(r1)
2742 * XXX On POWER7 and POWER8, we just spin here since we don't
2743 * know what the other threads are doing (and we don't want to
2744 * coordinate with them) - but at least we now have register state
2745 * in memory that we might be able to look at from another CPU.
2750 * This mimics the MSR transition on IRQ delivery. The new guest MSR is taken
2751 * from VCPU_INTR_MSR and is modified based on the required TM state changes.
2752 * r11 has the guest MSR value (in/out)
2753 * r9 has a vcpu pointer (in)
2754 * r0 is used as a scratch register
2756 SYM_FUNC_START_LOCAL(kvmppc_msr_interrupt)
2757 rldicl r0, r11, 64 - MSR_TS_S_LG, 62
2758 cmpwi r0, 2 /* Check if we are in transactional state.. */
2759 ld r11, VCPU_INTR_MSR(r9)
2761 /* ... if transactional, change to suspended */
2763 1: rldimi r11, r0, MSR_TS_S_LG, 63 - MSR_TS_T_LG
2765 SYM_FUNC_END(kvmppc_msr_interrupt)
2768 * void kvmhv_load_guest_pmu(struct kvm_vcpu *vcpu)
2770 * Load up guest PMU state. R3 points to the vcpu struct.
2772 SYM_FUNC_START_LOCAL(kvmhv_load_guest_pmu)
2776 sldi r3, r3, 31 /* MMCR0_FC (freeze counters) bit */
2777 mtspr SPRN_MMCR0, r3 /* freeze all counters, disable ints */
2780 ld r3, VCPU_MMCR(r4)
2781 andi. r5, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
2782 cmpwi r5, MMCR0_PMAO
2783 beql kvmppc_fix_pmao
2784 END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
2785 lwz r3, VCPU_PMC(r4) /* always load up guest PMU registers */
2786 lwz r5, VCPU_PMC + 4(r4) /* to prevent information leak */
2787 lwz r6, VCPU_PMC + 8(r4)
2788 lwz r7, VCPU_PMC + 12(r4)
2789 lwz r8, VCPU_PMC + 16(r4)
2790 lwz r9, VCPU_PMC + 20(r4)
2797 ld r3, VCPU_MMCR(r4)
2798 ld r5, VCPU_MMCR + 8(r4)
2799 ld r6, VCPU_MMCRA(r4)
2800 ld r7, VCPU_SIAR(r4)
2801 ld r8, VCPU_SDAR(r4)
2802 mtspr SPRN_MMCR1, r5
2803 mtspr SPRN_MMCRA, r6
2807 ld r5, VCPU_MMCR + 16(r4)
2808 ld r6, VCPU_SIER(r4)
2809 mtspr SPRN_MMCR2, r5
2811 lwz r7, VCPU_PMC + 24(r4)
2812 lwz r8, VCPU_PMC + 28(r4)
2813 ld r9, VCPU_MMCRS(r4)
2814 mtspr SPRN_SPMC1, r7
2815 mtspr SPRN_SPMC2, r8
2816 mtspr SPRN_MMCRS, r9
2817 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2818 mtspr SPRN_MMCR0, r3
2822 SYM_FUNC_END(kvmhv_load_guest_pmu)
2825 * void kvmhv_load_host_pmu(void)
2827 * Reload host PMU state saved in the PACA by kvmhv_save_host_pmu.
2829 SYM_FUNC_START_LOCAL(kvmhv_load_host_pmu)
2831 lbz r4, PACA_PMCINUSE(r13) /* is the host using the PMU? */
2833 beq 23f /* skip if not */
2835 ld r3, HSTATE_MMCR0(r13)
2836 andi. r4, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
2837 cmpwi r4, MMCR0_PMAO
2838 beql kvmppc_fix_pmao
2839 END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
2840 lwz r3, HSTATE_PMC1(r13)
2841 lwz r4, HSTATE_PMC2(r13)
2842 lwz r5, HSTATE_PMC3(r13)
2843 lwz r6, HSTATE_PMC4(r13)
2844 lwz r8, HSTATE_PMC5(r13)
2845 lwz r9, HSTATE_PMC6(r13)
2852 ld r3, HSTATE_MMCR0(r13)
2853 ld r4, HSTATE_MMCR1(r13)
2854 ld r5, HSTATE_MMCRA(r13)
2855 ld r6, HSTATE_SIAR(r13)
2856 ld r7, HSTATE_SDAR(r13)
2857 mtspr SPRN_MMCR1, r4
2858 mtspr SPRN_MMCRA, r5
2862 ld r8, HSTATE_MMCR2(r13)
2863 ld r9, HSTATE_SIER(r13)
2864 mtspr SPRN_MMCR2, r8
2866 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2867 mtspr SPRN_MMCR0, r3
2871 SYM_FUNC_END(kvmhv_load_host_pmu)
2874 * void kvmhv_save_guest_pmu(struct kvm_vcpu *vcpu, bool pmu_in_use)
2876 * Save guest PMU state into the vcpu struct.
2877 * r3 = vcpu, r4 = full save flag (PMU in use flag set in VPA)
2879 SYM_FUNC_START_LOCAL(kvmhv_save_guest_pmu)
2884 * POWER8 seems to have a hardware bug where setting
2885 * MMCR0[PMAE] along with MMCR0[PMC1CE] and/or MMCR0[PMCjCE]
2886 * when some counters are already negative doesn't seem
2887 * to cause a performance monitor alert (and hence interrupt).
2888 * The effect of this is that when saving the PMU state,
2889 * if there is no PMU alert pending when we read MMCR0
2890 * before freezing the counters, but one becomes pending
2891 * before we read the counters, we lose it.
2892 * To work around this, we need a way to freeze the counters
2893 * before reading MMCR0. Normally, freezing the counters
2894 * is done by writing MMCR0 (to set MMCR0[FC]) which
2895 * unavoidably writes MMCR0[PMA0] as well. On POWER8,
2896 * we can also freeze the counters using MMCR2, by writing
2897 * 1s to all the counter freeze condition bits (there are
2898 * 9 bits each for 6 counters).
2900 li r3, -1 /* set all freeze bits */
2902 mfspr r10, SPRN_MMCR2
2903 mtspr SPRN_MMCR2, r3
2905 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2907 sldi r3, r3, 31 /* MMCR0_FC (freeze counters) bit */
2908 mfspr r4, SPRN_MMCR0 /* save MMCR0 */
2909 mtspr SPRN_MMCR0, r3 /* freeze all counters, disable ints */
2910 mfspr r6, SPRN_MMCRA
2911 /* Clear MMCRA in order to disable SDAR updates */
2913 mtspr SPRN_MMCRA, r7
2915 cmpwi r8, 0 /* did they ask for PMU stuff to be saved? */
2917 std r3, VCPU_MMCR(r9) /* if not, set saved MMCR0 to FC */
2919 21: mfspr r5, SPRN_MMCR1
2922 std r4, VCPU_MMCR(r9)
2923 std r5, VCPU_MMCR + 8(r9)
2924 std r6, VCPU_MMCRA(r9)
2926 std r10, VCPU_MMCR + 16(r9)
2927 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2928 std r7, VCPU_SIAR(r9)
2929 std r8, VCPU_SDAR(r9)
2936 stw r3, VCPU_PMC(r9)
2937 stw r4, VCPU_PMC + 4(r9)
2938 stw r5, VCPU_PMC + 8(r9)
2939 stw r6, VCPU_PMC + 12(r9)
2940 stw r7, VCPU_PMC + 16(r9)
2941 stw r8, VCPU_PMC + 20(r9)
2944 std r5, VCPU_SIER(r9)
2945 mfspr r6, SPRN_SPMC1
2946 mfspr r7, SPRN_SPMC2
2947 mfspr r8, SPRN_MMCRS
2948 stw r6, VCPU_PMC + 24(r9)
2949 stw r7, VCPU_PMC + 28(r9)
2950 std r8, VCPU_MMCRS(r9)
2952 mtspr SPRN_MMCRS, r4
2953 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2955 SYM_FUNC_END(kvmhv_save_guest_pmu)
2958 * This works around a hardware bug on POWER8E processors, where
2959 * writing a 1 to the MMCR0[PMAO] bit doesn't generate a
2960 * performance monitor interrupt. Instead, when we need to have
2961 * an interrupt pending, we have to arrange for a counter to overflow.
2965 mtspr SPRN_MMCR2, r3
2966 lis r3, (MMCR0_PMXE | MMCR0_FCECE)@h
2967 ori r3, r3, MMCR0_PMCjCE | MMCR0_C56RUN
2968 mtspr SPRN_MMCR0, r3
2975 #ifdef CONFIG_KVM_BOOK3S_HV_P8_TIMING
2977 * Start timing an activity
2978 * r3 = pointer to time accumulation struct, r4 = vcpu
2981 ld r5, HSTATE_KVM_VCORE(r13)
2982 ld r6, VCORE_TB_OFFSET_APPL(r5)
2984 subf r5, r6, r5 /* subtract current timebase offset */
2985 std r3, VCPU_CUR_ACTIVITY(r4)
2986 std r5, VCPU_ACTIVITY_START(r4)
2990 * Accumulate time to one activity and start another.
2991 * r3 = pointer to new time accumulation struct, r4 = vcpu
2993 kvmhv_accumulate_time:
2994 ld r5, HSTATE_KVM_VCORE(r13)
2995 ld r8, VCORE_TB_OFFSET_APPL(r5)
2996 ld r5, VCPU_CUR_ACTIVITY(r4)
2997 ld r6, VCPU_ACTIVITY_START(r4)
2998 std r3, VCPU_CUR_ACTIVITY(r4)
3000 subf r7, r8, r7 /* subtract current timebase offset */
3001 std r7, VCPU_ACTIVITY_START(r4)
3005 ld r8, TAS_SEQCOUNT(r5)
3008 std r8, TAS_SEQCOUNT(r5)
3010 ld r7, TAS_TOTAL(r5)
3012 std r7, TAS_TOTAL(r5)
3018 3: std r3, TAS_MIN(r5)
3024 std r8, TAS_SEQCOUNT(r5)