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/code-patching-asm.h>
15 #include <asm/kvm_asm.h>
19 #include <asm/ptrace.h>
20 #include <asm/hvcall.h>
21 #include <asm/asm-offsets.h>
22 #include <asm/exception-64s.h>
23 #include <asm/kvm_book3s_asm.h>
24 #include <asm/book3s/64/mmu-hash.h>
25 #include <asm/export.h>
28 #include <asm/xive-regs.h>
29 #include <asm/thread_info.h>
30 #include <asm/asm-compat.h>
31 #include <asm/feature-fixups.h>
32 #include <asm/cpuidle.h>
33 #include <asm/ultravisor-api.h>
35 /* Sign-extend HDEC if not on POWER9 */
36 #define EXTEND_HDEC(reg) \
39 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
41 /* Values in HSTATE_NAPPING(r13) */
42 #define NAPPING_CEDE 1
43 #define NAPPING_NOVCPU 2
44 #define NAPPING_UNSPLIT 3
46 /* Stack frame offsets for kvmppc_hv_entry */
48 #define STACK_SLOT_TRAP (SFS-4)
49 #define STACK_SLOT_SHORT_PATH (SFS-8)
50 #define STACK_SLOT_TID (SFS-16)
51 #define STACK_SLOT_PSSCR (SFS-24)
52 #define STACK_SLOT_PID (SFS-32)
53 #define STACK_SLOT_IAMR (SFS-40)
54 #define STACK_SLOT_CIABR (SFS-48)
55 #define STACK_SLOT_DAWR (SFS-56)
56 #define STACK_SLOT_DAWRX (SFS-64)
57 #define STACK_SLOT_HFSCR (SFS-72)
58 #define STACK_SLOT_AMR (SFS-80)
59 #define STACK_SLOT_UAMOR (SFS-88)
60 /* the following is used by the P9 short path */
61 #define STACK_SLOT_NVGPRS (SFS-152) /* 18 gprs */
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 */
89 /* On P9, do LPCR setting, if necessary */
90 ld r3, HSTATE_SPLIT_MODE(r13)
93 lwz r4, KVM_SPLIT_DO_SET(r3)
99 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
101 ld r4, HSTATE_KVM_VCPU(r13)
104 /* Back from guest - restore host state and return to caller */
107 /* Restore host DABR and DABRX */
108 ld r5,HSTATE_DABR(r13)
112 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
115 ld r3,PACA_SPRG_VDSO(r13)
116 mtspr SPRN_SPRG_VDSO_WRITE,r3
118 /* Reload the host's PMU registers */
119 bl kvmhv_load_host_pmu
122 * Reload DEC. HDEC interrupts were disabled when
123 * we reloaded the host's LPCR value.
125 ld r3, HSTATE_DECEXP(r13)
130 /* hwthread_req may have got set by cede or no vcpu, so clear it */
132 stb r0, HSTATE_HWTHREAD_REQ(r13)
135 * For external interrupts we need to call the Linux
136 * handler to process the interrupt. We do that by jumping
137 * to absolute address 0x500 for external interrupts.
138 * The [h]rfid at the end of the handler will return to
139 * the book3s_hv_interrupts.S code. For other interrupts
140 * we do the rfid to get back to the book3s_hv_interrupts.S
143 ld r8, 112+PPC_LR_STKOFF(r1)
145 ld r7, HSTATE_HOST_MSR(r13)
147 /* Return the trap number on this thread as the return value */
151 * If we came back from the guest via a relocation-on interrupt,
152 * we will be in virtual mode at this point, which makes it a
153 * little easier to get back to the caller.
156 andi. r0, r0, MSR_IR /* in real mode? */
159 /* RFI into the highmem handler */
163 mtmsrd r6, 1 /* Clear RI in MSR */
168 /* Virtual-mode return */
173 kvmppc_primary_no_guest:
174 /* We handle this much like a ceded vcpu */
175 /* put the HDEC into the DEC, since HDEC interrupts don't wake us */
176 /* HDEC may be larger than DEC for arch >= v3.00, but since the */
177 /* HDEC value came from DEC in the first place, it will fit */
181 * Make sure the primary has finished the MMU switch.
182 * We should never get here on a secondary thread, but
183 * check it for robustness' sake.
185 ld r5, HSTATE_KVM_VCORE(r13)
186 65: lbz r0, VCORE_IN_GUEST(r5)
193 /* set our bit in napping_threads */
194 ld r5, HSTATE_KVM_VCORE(r13)
195 lbz r7, HSTATE_PTID(r13)
198 addi r6, r5, VCORE_NAPPING_THREADS
203 /* order napping_threads update vs testing entry_exit_map */
206 lwz r7, VCORE_ENTRY_EXIT(r5)
208 bge kvm_novcpu_exit /* another thread already exiting */
209 li r3, NAPPING_NOVCPU
210 stb r3, HSTATE_NAPPING(r13)
212 li r3, 0 /* Don't wake on privileged (OS) doorbell */
217 * Entered from kvm_start_guest if kvm_hstate.napping is set
223 ld r1, HSTATE_HOST_R1(r13)
224 ld r5, HSTATE_KVM_VCORE(r13)
226 stb r0, HSTATE_NAPPING(r13)
228 /* check the wake reason */
229 bl kvmppc_check_wake_reason
232 * Restore volatile registers since we could have called
233 * a C routine in kvmppc_check_wake_reason.
236 ld r5, HSTATE_KVM_VCORE(r13)
238 /* see if any other thread is already exiting */
239 lwz r0, VCORE_ENTRY_EXIT(r5)
243 /* clear our bit in napping_threads */
244 lbz r7, HSTATE_PTID(r13)
247 addi r6, r5, VCORE_NAPPING_THREADS
253 /* See if the wake reason means we need to exit */
257 /* See if our timeslice has expired (HDEC is negative) */
260 li r12, BOOK3S_INTERRUPT_HV_DECREMENTER
264 /* Got an IPI but other vcpus aren't yet exiting, must be a latecomer */
265 ld r4, HSTATE_KVM_VCPU(r13)
267 beq kvmppc_primary_no_guest
269 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
270 addi r3, r4, VCPU_TB_RMENTRY
271 bl kvmhv_start_timing
276 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
277 ld r4, HSTATE_KVM_VCPU(r13)
280 addi r3, r4, VCPU_TB_RMEXIT
281 bl kvmhv_accumulate_time
284 stw r12, STACK_SLOT_TRAP(r1)
285 bl kvmhv_commence_exit
287 b kvmhv_switch_to_host
290 * We come in here when wakened from Linux offline idle code.
292 * r3 contains the SRR1 wakeup value, SRR1 is trashed.
294 _GLOBAL(idle_kvm_start_guest)
295 ld r4,PACAEMERGSP(r13)
301 subi r1,r4,STACK_FRAME_OVERHEAD
305 * Could avoid this and pass it through in r3. For now,
306 * code expects it to be in SRR1.
311 stb r0,PACA_FTRACE_ENABLED(r13)
313 li r0,KVM_HWTHREAD_IN_KVM
314 stb r0,HSTATE_HWTHREAD_STATE(r13)
316 /* kvm cede / napping does not come through here */
317 lbz r0,HSTATE_NAPPING(r13)
324 stb r0, HSTATE_NAPPING(r13)
329 * We weren't napping due to cede, so this must be a secondary
330 * thread being woken up to run a guest, or being woken up due
331 * to a stray IPI. (Or due to some machine check or hypervisor
332 * maintenance interrupt while the core is in KVM.)
335 /* Check the wake reason in SRR1 to see why we got here */
336 bl kvmppc_check_wake_reason
338 * kvmppc_check_wake_reason could invoke a C routine, but we
339 * have no volatile registers to restore when we return.
345 /* get vcore pointer, NULL if we have nothing to run */
346 ld r5,HSTATE_KVM_VCORE(r13)
348 /* if we have no vcore to run, go back to sleep */
351 kvm_secondary_got_guest:
353 /* Set HSTATE_DSCR(r13) to something sensible */
354 ld r6, PACA_DSCR_DEFAULT(r13)
355 std r6, HSTATE_DSCR(r13)
357 /* On thread 0 of a subcore, set HDEC to max */
358 lbz r4, HSTATE_PTID(r13)
361 LOAD_REG_ADDR(r6, decrementer_max)
364 /* and set per-LPAR registers, if doing dynamic micro-threading */
365 ld r6, HSTATE_SPLIT_MODE(r13)
369 ld r0, KVM_SPLIT_RPR(r6)
371 ld r0, KVM_SPLIT_PMMAR(r6)
373 ld r0, KVM_SPLIT_LDBAR(r6)
377 /* On P9 we use the split_info for coordinating LPCR changes */
378 lwz r4, KVM_SPLIT_DO_SET(r6)
385 ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
387 /* Order load of vcpu after load of vcore */
389 ld r4, HSTATE_KVM_VCPU(r13)
392 /* Back from the guest, go back to nap */
393 /* Clear our vcpu and vcore pointers so we don't come back in early */
395 std r0, HSTATE_KVM_VCPU(r13)
397 * Once we clear HSTATE_KVM_VCORE(r13), the code in
398 * kvmppc_run_core() is going to assume that all our vcpu
399 * state is visible in memory. This lwsync makes sure
403 std r0, HSTATE_KVM_VCORE(r13)
406 * All secondaries exiting guest will fall through this path.
407 * Before proceeding, just check for HMI interrupt and
408 * invoke opal hmi handler. By now we are sure that the
409 * primary thread on this core/subcore has already made partition
410 * switch/TB resync and we are good to call opal hmi handler.
412 cmpwi r12, BOOK3S_INTERRUPT_HMI
415 li r3,0 /* NULL argument */
416 bl hmi_exception_realmode
418 * At this point we have finished executing in the guest.
419 * We need to wait for hwthread_req to become zero, since
420 * we may not turn on the MMU while hwthread_req is non-zero.
421 * While waiting we also need to check if we get given a vcpu to run.
424 lbz r3, HSTATE_HWTHREAD_REQ(r13)
428 li r0, KVM_HWTHREAD_IN_KERNEL
429 stb r0, HSTATE_HWTHREAD_STATE(r13)
430 /* need to recheck hwthread_req after a barrier, to avoid race */
432 lbz r3, HSTATE_HWTHREAD_REQ(r13)
437 * Jump to idle_return_gpr_loss, which returns to the
438 * idle_kvm_start_guest caller.
442 rlwimi r4, r3, 0, LPCR_PECE0 | LPCR_PECE1
444 /* set up r3 for return */
447 addi r1, r1, STACK_FRAME_OVERHEAD
456 ld r5, HSTATE_KVM_VCORE(r13)
459 ld r3, HSTATE_SPLIT_MODE(r13)
462 lwz r0, KVM_SPLIT_DO_SET(r3)
465 lwz r0, KVM_SPLIT_DO_RESTORE(r3)
468 lbz r0, KVM_SPLIT_DO_NAP(r3)
474 b kvm_secondary_got_guest
476 54: li r0, KVM_HWTHREAD_IN_KVM
477 stb r0, HSTATE_HWTHREAD_STATE(r13)
481 /* Set LPCR, LPIDR etc. on P9 */
489 bl kvmhv_p9_restore_lpcr
494 * Here the primary thread is trying to return the core to
495 * whole-core mode, so we need to nap.
499 * When secondaries are napping in kvm_unsplit_nap() with
500 * hwthread_req = 1, HMI goes ignored even though subcores are
501 * already exited the guest. Hence HMI keeps waking up secondaries
502 * from nap in a loop and secondaries always go back to nap since
503 * no vcore is assigned to them. This makes impossible for primary
504 * thread to get hold of secondary threads resulting into a soft
505 * lockup in KVM path.
507 * Let us check if HMI is pending and handle it before we go to nap.
509 cmpwi r12, BOOK3S_INTERRUPT_HMI
511 li r3, 0 /* NULL argument */
512 bl hmi_exception_realmode
515 * Ensure that secondary doesn't nap when it has
516 * its vcore pointer set.
518 sync /* matches smp_mb() before setting split_info.do_nap */
519 ld r0, HSTATE_KVM_VCORE(r13)
522 /* clear any pending message */
524 lis r6, (PPC_DBELL_SERVER << (63-36))@h
526 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
527 /* Set kvm_split_mode.napped[tid] = 1 */
528 ld r3, HSTATE_SPLIT_MODE(r13)
530 lbz r4, HSTATE_TID(r13)
531 addi r4, r4, KVM_SPLIT_NAPPED
533 /* Check the do_nap flag again after setting napped[] */
535 lbz r0, KVM_SPLIT_DO_NAP(r3)
538 li r3, NAPPING_UNSPLIT
539 stb r3, HSTATE_NAPPING(r13)
540 li r3, (LPCR_PECEDH | LPCR_PECE0) >> 4
542 rlwimi r5, r3, 4, (LPCR_PECEDP | LPCR_PECEDH | LPCR_PECE0 | LPCR_PECE1)
549 /******************************************************************************
553 *****************************************************************************/
555 .global kvmppc_hv_entry
560 * R4 = vcpu pointer (or NULL)
565 * all other volatile GPRS = free
566 * Does not preserve non-volatile GPRs or CR fields
569 std r0, PPC_LR_STKOFF(r1)
572 /* Save R1 in the PACA */
573 std r1, HSTATE_HOST_R1(r13)
575 li r6, KVM_GUEST_MODE_HOST_HV
576 stb r6, HSTATE_IN_GUEST(r13)
578 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
579 /* Store initial timestamp */
582 addi r3, r4, VCPU_TB_RMENTRY
583 bl kvmhv_start_timing
587 ld r5, HSTATE_KVM_VCORE(r13)
588 ld r9, VCORE_KVM(r5) /* pointer to struct kvm */
591 * POWER7/POWER8 host -> guest partition switch code.
592 * We don't have to lock against concurrent tlbies,
593 * but we do have to coordinate across hardware threads.
595 /* Set bit in entry map iff exit map is zero. */
597 lbz r6, HSTATE_PTID(r13)
599 addi r8, r5, VCORE_ENTRY_EXIT
601 cmpwi r3, 0x100 /* any threads starting to exit? */
602 bge secondary_too_late /* if so we're too late to the party */
607 /* Primary thread switches to guest partition. */
614 li r0,LPID_RSVD /* switch to reserved LPID */
617 mtspr SPRN_SDR1,r6 /* switch to partition page table */
618 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
622 /* See if we need to flush the TLB. */
623 mr r3, r9 /* kvm pointer */
624 lhz r4, PACAPACAINDEX(r13) /* physical cpu number */
625 li r5, 0 /* nested vcpu pointer */
626 bl kvmppc_check_need_tlb_flush
628 ld r5, HSTATE_KVM_VCORE(r13)
630 /* Add timebase offset onto timebase */
631 22: ld r8,VCORE_TB_OFFSET(r5)
634 std r8, VCORE_TB_OFFSET_APPL(r5)
635 mftb r6 /* current host timebase */
637 mtspr SPRN_TBU40,r8 /* update upper 40 bits */
638 mftb r7 /* check if lower 24 bits overflowed */
643 addis r8,r8,0x100 /* if so, increment upper 40 bits */
646 /* Load guest PCR value to select appropriate compat mode */
647 37: ld r7, VCORE_PCR(r5)
648 LOAD_REG_IMMEDIATE(r6, PCR_MASK)
656 /* DPDES and VTB are shared between threads */
657 ld r8, VCORE_DPDES(r5)
661 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
663 /* Mark the subcore state as inside guest */
664 bl kvmppc_subcore_enter_guest
666 ld r5, HSTATE_KVM_VCORE(r13)
667 ld r4, HSTATE_KVM_VCPU(r13)
669 stb r0,VCORE_IN_GUEST(r5) /* signal secondaries to continue */
671 /* Do we have a guest vcpu to run? */
673 beq kvmppc_primary_no_guest
675 /* Increment yield count if they have a VPA */
679 li r6, LPPACA_YIELDCOUNT
684 stb r6, VCPU_VPA_DIRTY(r4)
687 /* Save purr/spurr */
690 std r5,HSTATE_PURR(r13)
691 std r6,HSTATE_SPURR(r13)
697 /* Save host values of some registers */
702 std r5, STACK_SLOT_TID(r1)
703 std r6, STACK_SLOT_PSSCR(r1)
704 std r7, STACK_SLOT_PID(r1)
706 std r5, STACK_SLOT_HFSCR(r1)
707 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
713 std r5, STACK_SLOT_CIABR(r1)
714 std r6, STACK_SLOT_DAWR(r1)
715 std r7, STACK_SLOT_DAWRX(r1)
716 std r8, STACK_SLOT_IAMR(r1)
717 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
720 std r5, STACK_SLOT_AMR(r1)
722 std r6, STACK_SLOT_UAMOR(r1)
725 /* Set partition DABR */
726 /* Do this before re-enabling PMU to avoid P7 DABR corruption bug */
727 lwz r5,VCPU_DABRX(r4)
732 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
734 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
736 * Branch around the call if both CPU_FTR_TM and
737 * CPU_FTR_P9_TM_HV_ASSIST are off.
741 END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
743 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
747 li r5, 0 /* don't preserve non-vol regs */
748 bl kvmppc_restore_tm_hv
750 ld r4, HSTATE_KVM_VCPU(r13)
754 /* Load guest PMU registers; r4 = vcpu pointer here */
756 bl kvmhv_load_guest_pmu
758 /* Load up FP, VMX and VSX registers */
759 ld r4, HSTATE_KVM_VCPU(r13)
762 ld r14, VCPU_GPR(R14)(r4)
763 ld r15, VCPU_GPR(R15)(r4)
764 ld r16, VCPU_GPR(R16)(r4)
765 ld r17, VCPU_GPR(R17)(r4)
766 ld r18, VCPU_GPR(R18)(r4)
767 ld r19, VCPU_GPR(R19)(r4)
768 ld r20, VCPU_GPR(R20)(r4)
769 ld r21, VCPU_GPR(R21)(r4)
770 ld r22, VCPU_GPR(R22)(r4)
771 ld r23, VCPU_GPR(R23)(r4)
772 ld r24, VCPU_GPR(R24)(r4)
773 ld r25, VCPU_GPR(R25)(r4)
774 ld r26, VCPU_GPR(R26)(r4)
775 ld r27, VCPU_GPR(R27)(r4)
776 ld r28, VCPU_GPR(R28)(r4)
777 ld r29, VCPU_GPR(R29)(r4)
778 ld r30, VCPU_GPR(R30)(r4)
779 ld r31, VCPU_GPR(R31)(r4)
781 /* Switch DSCR to guest value */
786 /* Skip next section on POWER7 */
788 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
789 /* Load up POWER8-specific registers */
791 lwz r6, VCPU_PSPB(r4)
797 * Handle broken DAWR case by not writing it. This means we
798 * can still store the DAWR register for migration.
800 LOAD_REG_ADDR(r5, dawr_force_enable)
805 ld r6, VCPU_DAWRX(r4)
809 ld r7, VCPU_CIABR(r4)
814 ld r8, VCPU_EBBHR(r4)
817 ld r5, VCPU_EBBRR(r4)
818 ld r6, VCPU_BESCR(r4)
819 lwz r7, VCPU_GUEST_PID(r4)
826 /* POWER8-only registers */
827 ld r5, VCPU_TCSCR(r4)
829 ld r7, VCPU_CSIGR(r4)
837 /* POWER9-only registers */
839 ld r6, VCPU_PSSCR(r4)
840 lbz r8, HSTATE_FAKE_SUSPEND(r13)
841 oris r6, r6, PSSCR_EC@h /* This makes stop trap to HV */
842 rldimi r6, r8, PSSCR_FAKE_SUSPEND_LG, 63 - PSSCR_FAKE_SUSPEND_LG
843 ld r7, VCPU_HFSCR(r4)
847 ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
850 ld r5, VCPU_SPRG0(r4)
851 ld r6, VCPU_SPRG1(r4)
852 ld r7, VCPU_SPRG2(r4)
853 ld r8, VCPU_SPRG3(r4)
859 /* Load up DAR and DSISR */
861 lwz r6, VCPU_DSISR(r4)
865 /* Restore AMR and UAMOR, set AMOR to all 1s */
873 /* Restore state of CTRL run bit; assume 1 on entry */
881 /* Secondary threads wait for primary to have done partition switch */
882 ld r5, HSTATE_KVM_VCORE(r13)
883 lbz r6, HSTATE_PTID(r13)
886 lbz r0, VCORE_IN_GUEST(r5)
890 20: lwz r3, VCORE_ENTRY_EXIT(r5)
893 lbz r0, VCORE_IN_GUEST(r5)
904 * Set the decrementer to the guest decrementer.
906 ld r8,VCPU_DEC_EXPIRES(r4)
907 /* r8 is a host timebase value here, convert to guest TB */
908 ld r5,HSTATE_KVM_VCORE(r13)
909 ld r6,VCORE_TB_OFFSET_APPL(r5)
915 /* Check if HDEC expires soon */
918 cmpdi r3, 512 /* 1 microsecond */
921 /* For hash guest, clear out and reload the SLB */
923 lbz r0, KVM_RADIX(r6)
931 /* Load up guest SLB entries (N.B. slb_max will be 0 for radix) */
932 lwz r5,VCPU_SLB_MAX(r4)
937 1: ld r8,VCPU_SLB_E(r6)
940 addi r6,r6,VCPU_SLB_SIZE
944 #ifdef CONFIG_KVM_XICS
945 /* We are entering the guest on that thread, push VCPU to XIVE */
946 ld r11, VCPU_XIVE_SAVED_STATE(r4)
948 lwz r8, VCPU_XIVE_CAM_WORD(r4)
951 li r7, TM_QW1_OS + TM_WORD2
953 andi. r0, r0, MSR_DR /* in real mode? */
955 ld r10, HSTATE_XIVE_TIMA_VIRT(r13)
962 2: ld r10, HSTATE_XIVE_TIMA_PHYS(r13)
969 stb r9, VCPU_XIVE_PUSHED(r4)
973 * We clear the irq_pending flag. There is a small chance of a
974 * race vs. the escalation interrupt happening on another
975 * processor setting it again, but the only consequence is to
976 * cause a spurrious wakeup on the next H_CEDE which is not an
980 stb r0, VCPU_IRQ_PENDING(r4)
983 * In single escalation mode, if the escalation interrupt is
986 lbz r0, VCPU_XIVE_ESC_ON(r4)
989 li r9, XIVE_ESB_SET_PQ_01
990 beq 4f /* in real mode? */
991 ld r10, VCPU_XIVE_ESC_VADDR(r4)
994 4: ld r10, VCPU_XIVE_ESC_RADDR(r4)
998 /* We have a possible subtle race here: The escalation interrupt might
999 * have fired and be on its way to the host queue while we mask it,
1000 * and if we unmask it early enough (re-cede right away), there is
1001 * a theorical possibility that it fires again, thus landing in the
1002 * target queue more than once which is a big no-no.
1004 * Fortunately, solving this is rather easy. If the above load setting
1005 * PQ to 01 returns a previous value where P is set, then we know the
1006 * escalation interrupt is somewhere on its way to the host. In that
1007 * case we simply don't clear the xive_esc_on flag below. It will be
1008 * eventually cleared by the handler for the escalation interrupt.
1010 * Then, when doing a cede, we check that flag again before re-enabling
1011 * the escalation interrupt, and if set, we abort the cede.
1013 andi. r0, r0, XIVE_ESB_VAL_P
1016 /* Now P is 0, we can clear the flag */
1018 stb r0, VCPU_XIVE_ESC_ON(r4)
1021 #endif /* CONFIG_KVM_XICS */
1024 stw r0, STACK_SLOT_SHORT_PATH(r1)
1026 deliver_guest_interrupt: /* r4 = vcpu, r13 = paca */
1027 /* Check if we can deliver an external or decrementer interrupt now */
1028 ld r0, VCPU_PENDING_EXC(r4)
1030 /* On POWER9, also check for emulated doorbell interrupt */
1031 lbz r3, VCPU_DBELL_REQ(r4)
1033 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1037 bl kvmppc_guest_entry_inject_int
1038 ld r4, HSTATE_KVM_VCPU(r13)
1040 ld r6, VCPU_SRR0(r4)
1041 ld r7, VCPU_SRR1(r4)
1047 ld r11, VCPU_MSR(r4)
1048 /* r11 = vcpu->arch.msr & ~MSR_HV */
1049 rldicl r11, r11, 63 - MSR_HV_LG, 1
1050 rotldi r11, r11, 1 + MSR_HV_LG
1051 ori r11, r11, MSR_ME
1061 * R10: value for HSRR0
1062 * R11: value for HSRR1
1067 stb r0,VCPU_CEDED(r4) /* cancel cede */
1068 mtspr SPRN_HSRR0,r10
1069 mtspr SPRN_HSRR1,r11
1071 /* Activate guest mode, so faults get handled by KVM */
1072 li r9, KVM_GUEST_MODE_GUEST_HV
1073 stb r9, HSTATE_IN_GUEST(r13)
1075 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1076 /* Accumulate timing */
1077 addi r3, r4, VCPU_TB_GUEST
1078 bl kvmhv_accumulate_time
1084 ld r5, VCPU_CFAR(r4)
1086 END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
1089 END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1094 ld r1, VCPU_GPR(R1)(r4)
1095 ld r5, VCPU_GPR(R5)(r4)
1096 ld r8, VCPU_GPR(R8)(r4)
1097 ld r9, VCPU_GPR(R9)(r4)
1098 ld r10, VCPU_GPR(R10)(r4)
1099 ld r11, VCPU_GPR(R11)(r4)
1100 ld r12, VCPU_GPR(R12)(r4)
1101 ld r13, VCPU_GPR(R13)(r4)
1105 END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1107 /* Move canary into DSISR to check for later */
1110 mtspr SPRN_HDSISR, r0
1111 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1114 lbz r7, KVM_SECURE_GUEST(r6)
1116 ld r6, VCPU_GPR(R6)(r4)
1117 ld r7, VCPU_GPR(R7)(r4)
1123 ld r0, VCPU_GPR(R0)(r4)
1124 ld r2, VCPU_GPR(R2)(r4)
1125 ld r3, VCPU_GPR(R3)(r4)
1126 ld r4, VCPU_GPR(R4)(r4)
1130 * Use UV_RETURN ultracall to return control back to the Ultravisor after
1131 * processing an hypercall or interrupt that was forwarded (a.k.a. reflected)
1132 * to the Hypervisor.
1134 * All registers have already been loaded, except:
1136 * R2 = SRR1, so UV can detect a synthesized interrupt (if any)
1143 ld r0, VCPU_GPR(R3)(r4)
1146 ori r3, r3, UV_RETURN
1147 ld r4, VCPU_GPR(R4)(r4)
1151 * Enter the guest on a P9 or later system where we have exactly
1152 * one vcpu per vcore and we don't need to go to real mode
1153 * (which implies that host and guest are both using radix MMU mode).
1155 * Most SPRs and all the VSRs have been loaded already.
1157 _GLOBAL(__kvmhv_vcpu_entry_p9)
1158 EXPORT_SYMBOL_GPL(__kvmhv_vcpu_entry_p9)
1160 std r0, PPC_LR_STKOFF(r1)
1164 stw r0, STACK_SLOT_SHORT_PATH(r1)
1166 std r3, HSTATE_KVM_VCPU(r13)
1170 std r1, HSTATE_HOST_R1(r13)
1174 std reg, STACK_SLOT_NVGPRS + ((reg - 14) * 8)(r1)
1180 ld reg, __VCPU_GPR(reg)(r3)
1185 std r10, HSTATE_HOST_MSR(r13)
1188 b fast_guest_entry_c
1189 guest_exit_short_path:
1191 li r0, KVM_GUEST_MODE_NONE
1192 stb r0, HSTATE_IN_GUEST(r13)
1196 std reg, __VCPU_GPR(reg)(r9)
1202 ld reg, STACK_SLOT_NVGPRS + ((reg - 14) * 8)(r1)
1209 mr r3, r12 /* trap number */
1212 ld r0, PPC_LR_STKOFF(r1)
1215 /* If we are in real mode, do a rfid to get back to the caller */
1217 andi. r5, r4, MSR_IR
1219 rldicl r5, r4, 64 - MSR_TS_S_LG, 62 /* extract TS field */
1221 ld r10, HSTATE_HOST_MSR(r13)
1222 rldimi r10, r5, MSR_TS_S_LG, 63 - MSR_TS_T_LG
1223 mtspr SPRN_SRR1, r10
1229 stw r12, STACK_SLOT_TRAP(r1)
1232 stw r12, VCPU_TRAP(r4)
1233 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1234 addi r3, r4, VCPU_TB_RMEXIT
1235 bl kvmhv_accumulate_time
1237 11: b kvmhv_switch_to_host
1244 li r12, BOOK3S_INTERRUPT_HV_DECREMENTER
1245 12: stw r12, VCPU_TRAP(r4)
1247 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1248 addi r3, r4, VCPU_TB_RMEXIT
1249 bl kvmhv_accumulate_time
1253 /******************************************************************************
1257 *****************************************************************************/
1260 * We come here from the first-level interrupt handlers.
1262 .globl kvmppc_interrupt_hv
1263 kvmppc_interrupt_hv:
1265 * Register contents:
1266 * R12 = (guest CR << 32) | interrupt vector
1268 * guest R12 saved in shadow VCPU SCRATCH0
1269 * guest CTR saved in shadow VCPU SCRATCH1 if RELOCATABLE
1270 * guest R13 saved in SPRN_SCRATCH0
1272 std r9, HSTATE_SCRATCH2(r13)
1273 lbz r9, HSTATE_IN_GUEST(r13)
1274 cmpwi r9, KVM_GUEST_MODE_HOST_HV
1275 beq kvmppc_bad_host_intr
1276 #ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
1277 cmpwi r9, KVM_GUEST_MODE_GUEST
1278 ld r9, HSTATE_SCRATCH2(r13)
1279 beq kvmppc_interrupt_pr
1281 /* We're now back in the host but in guest MMU context */
1282 li r9, KVM_GUEST_MODE_HOST_HV
1283 stb r9, HSTATE_IN_GUEST(r13)
1285 ld r9, HSTATE_KVM_VCPU(r13)
1287 /* Save registers */
1289 std r0, VCPU_GPR(R0)(r9)
1290 std r1, VCPU_GPR(R1)(r9)
1291 std r2, VCPU_GPR(R2)(r9)
1292 std r3, VCPU_GPR(R3)(r9)
1293 std r4, VCPU_GPR(R4)(r9)
1294 std r5, VCPU_GPR(R5)(r9)
1295 std r6, VCPU_GPR(R6)(r9)
1296 std r7, VCPU_GPR(R7)(r9)
1297 std r8, VCPU_GPR(R8)(r9)
1298 ld r0, HSTATE_SCRATCH2(r13)
1299 std r0, VCPU_GPR(R9)(r9)
1300 std r10, VCPU_GPR(R10)(r9)
1301 std r11, VCPU_GPR(R11)(r9)
1302 ld r3, HSTATE_SCRATCH0(r13)
1303 std r3, VCPU_GPR(R12)(r9)
1304 /* CR is in the high half of r12 */
1308 ld r3, HSTATE_CFAR(r13)
1309 std r3, VCPU_CFAR(r9)
1310 END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
1312 ld r4, HSTATE_PPR(r13)
1313 std r4, VCPU_PPR(r9)
1314 END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
1316 /* Restore R1/R2 so we can handle faults */
1317 ld r1, HSTATE_HOST_R1(r13)
1320 mfspr r10, SPRN_SRR0
1321 mfspr r11, SPRN_SRR1
1322 std r10, VCPU_SRR0(r9)
1323 std r11, VCPU_SRR1(r9)
1324 /* trap is in the low half of r12, clear CR from the high half */
1326 andi. r0, r12, 2 /* need to read HSRR0/1? */
1328 mfspr r10, SPRN_HSRR0
1329 mfspr r11, SPRN_HSRR1
1331 1: std r10, VCPU_PC(r9)
1332 std r11, VCPU_MSR(r9)
1336 std r3, VCPU_GPR(R13)(r9)
1339 stw r12,VCPU_TRAP(r9)
1342 * Now that we have saved away SRR0/1 and HSRR0/1,
1343 * interrupts are recoverable in principle, so set MSR_RI.
1344 * This becomes important for relocation-on interrupts from
1345 * the guest, which we can get in radix mode on POWER9.
1350 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1351 addi r3, r9, VCPU_TB_RMINTR
1353 bl kvmhv_accumulate_time
1354 ld r5, VCPU_GPR(R5)(r9)
1355 ld r6, VCPU_GPR(R6)(r9)
1356 ld r7, VCPU_GPR(R7)(r9)
1357 ld r8, VCPU_GPR(R8)(r9)
1360 /* Save HEIR (HV emulation assist reg) in emul_inst
1361 if this is an HEI (HV emulation interrupt, e40) */
1362 li r3,KVM_INST_FETCH_FAILED
1363 stw r3,VCPU_LAST_INST(r9)
1364 cmpwi r12,BOOK3S_INTERRUPT_H_EMUL_ASSIST
1367 11: stw r3,VCPU_HEIR(r9)
1369 /* these are volatile across C function calls */
1370 #ifdef CONFIG_RELOCATABLE
1371 ld r3, HSTATE_SCRATCH1(r13)
1377 std r3, VCPU_CTR(r9)
1378 std r4, VCPU_XER(r9)
1380 /* Save more register state */
1383 std r3, VCPU_DAR(r9)
1384 stw r4, VCPU_DSISR(r9)
1386 /* If this is a page table miss then see if it's theirs or ours */
1387 cmpwi r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
1389 std r3, VCPU_FAULT_DAR(r9)
1390 stw r4, VCPU_FAULT_DSISR(r9)
1391 cmpwi r12, BOOK3S_INTERRUPT_H_INST_STORAGE
1394 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1395 /* For softpatch interrupt, go off and do TM instruction emulation */
1396 cmpwi r12, BOOK3S_INTERRUPT_HV_SOFTPATCH
1400 /* See if this is a leftover HDEC interrupt */
1401 cmpwi r12,BOOK3S_INTERRUPT_HV_DECREMENTER
1407 bge fast_guest_return
1409 /* See if this is an hcall we can handle in real mode */
1410 cmpwi r12,BOOK3S_INTERRUPT_SYSCALL
1411 beq hcall_try_real_mode
1413 /* Hypervisor doorbell - exit only if host IPI flag set */
1414 cmpwi r12, BOOK3S_INTERRUPT_H_DOORBELL
1419 /* always exit if we're running a nested guest */
1420 ld r0, VCPU_NESTED(r9)
1423 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1424 lbz r0, HSTATE_HOST_IPI(r13)
1426 beq maybe_reenter_guest
1429 /* If it's a hypervisor facility unavailable interrupt, save HFSCR */
1430 cmpwi r12, BOOK3S_INTERRUPT_H_FAC_UNAVAIL
1432 mfspr r3, SPRN_HFSCR
1433 std r3, VCPU_HFSCR(r9)
1436 /* External interrupt ? */
1437 cmpwi r12, BOOK3S_INTERRUPT_EXTERNAL
1438 beq kvmppc_guest_external
1439 /* See if it is a machine check */
1440 cmpwi r12, BOOK3S_INTERRUPT_MACHINE_CHECK
1441 beq machine_check_realmode
1442 /* Or a hypervisor maintenance interrupt */
1443 cmpwi r12, BOOK3S_INTERRUPT_HMI
1446 guest_exit_cont: /* r9 = vcpu, r12 = trap, r13 = paca */
1448 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1449 addi r3, r9, VCPU_TB_RMEXIT
1451 bl kvmhv_accumulate_time
1453 #ifdef CONFIG_KVM_XICS
1454 /* We are exiting, pull the VP from the XIVE */
1455 lbz r0, VCPU_XIVE_PUSHED(r9)
1458 li r7, TM_SPC_PULL_OS_CTX
1461 andi. r0, r0, MSR_DR /* in real mode? */
1463 ld r10, HSTATE_XIVE_TIMA_VIRT(r13)
1466 /* First load to pull the context, we ignore the value */
1469 /* Second load to recover the context state (Words 0 and 1) */
1472 2: ld r10, HSTATE_XIVE_TIMA_PHYS(r13)
1475 /* First load to pull the context, we ignore the value */
1478 /* Second load to recover the context state (Words 0 and 1) */
1480 3: std r11, VCPU_XIVE_SAVED_STATE(r9)
1481 /* Fixup some of the state for the next load */
1484 stb r10, VCPU_XIVE_PUSHED(r9)
1485 stb r10, (VCPU_XIVE_SAVED_STATE+3)(r9)
1486 stb r0, (VCPU_XIVE_SAVED_STATE+4)(r9)
1489 #endif /* CONFIG_KVM_XICS */
1492 * Possibly flush the link stack here, before we do a blr in
1493 * guest_exit_short_path.
1496 patch_site 1b patch__call_kvm_flush_link_stack
1498 /* If we came in through the P9 short path, go back out to C now */
1499 lwz r0, STACK_SLOT_SHORT_PATH(r1)
1501 bne guest_exit_short_path
1503 /* For hash guest, read the guest SLB and save it away */
1505 lbz r0, KVM_RADIX(r5)
1508 bne 3f /* for radix, save 0 entries */
1509 lwz r0,VCPU_SLB_NR(r9) /* number of entries in SLB */
1514 andis. r0,r8,SLB_ESID_V@h
1516 add r8,r8,r6 /* put index in */
1518 std r8,VCPU_SLB_E(r7)
1519 std r3,VCPU_SLB_V(r7)
1520 addi r7,r7,VCPU_SLB_SIZE
1524 /* Finally clear out the SLB */
1529 3: stw r5,VCPU_SLB_MAX(r9)
1531 /* load host SLB entries */
1532 BEGIN_MMU_FTR_SECTION
1534 END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
1535 ld r8,PACA_SLBSHADOWPTR(r13)
1537 .rept SLB_NUM_BOLTED
1538 li r3, SLBSHADOW_SAVEAREA
1542 andis. r7,r5,SLB_ESID_V@h
1550 stw r12, STACK_SLOT_TRAP(r1)
1553 /* Do this before kvmhv_commence_exit so we know TB is guest TB */
1554 ld r3, HSTATE_KVM_VCORE(r13)
1557 /* On P9, if the guest has large decr enabled, don't sign extend */
1559 ld r4, VCORE_LPCR(r3)
1560 andis. r4, r4, LPCR_LD@h
1562 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1565 /* r5 is a guest timebase value here, convert to host TB */
1566 ld r4,VCORE_TB_OFFSET_APPL(r3)
1568 std r5,VCPU_DEC_EXPIRES(r9)
1570 /* Increment exit count, poke other threads to exit */
1572 bl kvmhv_commence_exit
1574 ld r9, HSTATE_KVM_VCPU(r13)
1576 /* Stop others sending VCPU interrupts to this physical CPU */
1578 stw r0, VCPU_CPU(r9)
1579 stw r0, VCPU_THREAD_CPU(r9)
1581 /* Save guest CTRL register, set runlatch to 1 */
1583 stw r6,VCPU_CTRL(r9)
1590 * Save the guest PURR/SPURR
1595 ld r8,VCPU_SPURR(r9)
1596 std r5,VCPU_PURR(r9)
1597 std r6,VCPU_SPURR(r9)
1602 * Restore host PURR/SPURR and add guest times
1603 * so that the time in the guest gets accounted.
1605 ld r3,HSTATE_PURR(r13)
1606 ld r4,HSTATE_SPURR(r13)
1614 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
1615 /* Save POWER8-specific registers */
1619 std r5, VCPU_IAMR(r9)
1620 stw r6, VCPU_PSPB(r9)
1621 std r7, VCPU_FSCR(r9)
1625 std r7, VCPU_TAR(r9)
1626 mfspr r8, SPRN_EBBHR
1627 std r8, VCPU_EBBHR(r9)
1628 mfspr r5, SPRN_EBBRR
1629 mfspr r6, SPRN_BESCR
1632 std r5, VCPU_EBBRR(r9)
1633 std r6, VCPU_BESCR(r9)
1634 stw r7, VCPU_GUEST_PID(r9)
1635 std r8, VCPU_WORT(r9)
1637 mfspr r5, SPRN_TCSCR
1639 mfspr r7, SPRN_CSIGR
1641 std r5, VCPU_TCSCR(r9)
1642 std r6, VCPU_ACOP(r9)
1643 std r7, VCPU_CSIGR(r9)
1644 std r8, VCPU_TACR(r9)
1647 mfspr r6, SPRN_PSSCR
1648 std r5, VCPU_TID(r9)
1649 rldicl r6, r6, 4, 50 /* r6 &= PSSCR_GUEST_VIS */
1651 std r6, VCPU_PSSCR(r9)
1652 /* Restore host HFSCR value */
1653 ld r7, STACK_SLOT_HFSCR(r1)
1654 mtspr SPRN_HFSCR, r7
1655 ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
1657 * Restore various registers to 0, where non-zero values
1658 * set by the guest could disrupt the host.
1664 mtspr SPRN_TCSCR, r0
1665 /* Set MMCRS to 1<<31 to freeze and disable the SPMC counters */
1668 mtspr SPRN_MMCRS, r0
1669 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
1671 /* Save and restore AMR, IAMR and UAMOR before turning on the MMU */
1672 ld r8, STACK_SLOT_IAMR(r1)
1675 8: /* Power7 jumps back in here */
1679 std r6,VCPU_UAMOR(r9)
1680 ld r5,STACK_SLOT_AMR(r1)
1681 ld r6,STACK_SLOT_UAMOR(r1)
1683 mtspr SPRN_UAMOR, r6
1685 /* Switch DSCR back to host value */
1687 ld r7, HSTATE_DSCR(r13)
1688 std r8, VCPU_DSCR(r9)
1691 /* Save non-volatile GPRs */
1692 std r14, VCPU_GPR(R14)(r9)
1693 std r15, VCPU_GPR(R15)(r9)
1694 std r16, VCPU_GPR(R16)(r9)
1695 std r17, VCPU_GPR(R17)(r9)
1696 std r18, VCPU_GPR(R18)(r9)
1697 std r19, VCPU_GPR(R19)(r9)
1698 std r20, VCPU_GPR(R20)(r9)
1699 std r21, VCPU_GPR(R21)(r9)
1700 std r22, VCPU_GPR(R22)(r9)
1701 std r23, VCPU_GPR(R23)(r9)
1702 std r24, VCPU_GPR(R24)(r9)
1703 std r25, VCPU_GPR(R25)(r9)
1704 std r26, VCPU_GPR(R26)(r9)
1705 std r27, VCPU_GPR(R27)(r9)
1706 std r28, VCPU_GPR(R28)(r9)
1707 std r29, VCPU_GPR(R29)(r9)
1708 std r30, VCPU_GPR(R30)(r9)
1709 std r31, VCPU_GPR(R31)(r9)
1712 mfspr r3, SPRN_SPRG0
1713 mfspr r4, SPRN_SPRG1
1714 mfspr r5, SPRN_SPRG2
1715 mfspr r6, SPRN_SPRG3
1716 std r3, VCPU_SPRG0(r9)
1717 std r4, VCPU_SPRG1(r9)
1718 std r5, VCPU_SPRG2(r9)
1719 std r6, VCPU_SPRG3(r9)
1725 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1727 * Branch around the call if both CPU_FTR_TM and
1728 * CPU_FTR_P9_TM_HV_ASSIST are off.
1732 END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
1734 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
1738 li r5, 0 /* don't preserve non-vol regs */
1739 bl kvmppc_save_tm_hv
1741 ld r9, HSTATE_KVM_VCPU(r13)
1745 /* Increment yield count if they have a VPA */
1746 ld r8, VCPU_VPA(r9) /* do they have a VPA? */
1749 li r4, LPPACA_YIELDCOUNT
1754 stb r3, VCPU_VPA_DIRTY(r9)
1756 /* Save PMU registers if requested */
1757 /* r8 and cr0.eq are live here */
1760 beq 21f /* if no VPA, save PMU stuff anyway */
1761 lbz r4, LPPACA_PMCINUSE(r8)
1762 21: bl kvmhv_save_guest_pmu
1763 ld r9, HSTATE_KVM_VCPU(r13)
1765 /* Restore host values of some registers */
1767 ld r5, STACK_SLOT_CIABR(r1)
1768 ld r6, STACK_SLOT_DAWR(r1)
1769 ld r7, STACK_SLOT_DAWRX(r1)
1770 mtspr SPRN_CIABR, r5
1772 * If the DAWR doesn't work, it's ok to write these here as
1773 * this value should always be zero
1776 mtspr SPRN_DAWRX, r7
1777 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1779 ld r5, STACK_SLOT_TID(r1)
1780 ld r6, STACK_SLOT_PSSCR(r1)
1781 ld r7, STACK_SLOT_PID(r1)
1783 mtspr SPRN_PSSCR, r6
1785 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1787 #ifdef CONFIG_PPC_RADIX_MMU
1789 * Are we running hash or radix ?
1792 lbz r0, KVM_RADIX(r5)
1797 * Radix: do eieio; tlbsync; ptesync sequence in case we
1798 * interrupted the guest between a tlbie and a ptesync.
1804 /* Radix: Handle the case where the guest used an illegal PID */
1805 LOAD_REG_ADDR(r4, mmu_base_pid)
1806 lwz r3, VCPU_GUEST_PID(r9)
1812 * Illegal PID, the HW might have prefetched and cached in the TLB
1813 * some translations for the LPID 0 / guest PID combination which
1814 * Linux doesn't know about, so we need to flush that PID out of
1815 * the TLB. First we need to set LPIDR to 0 so tlbiel applies to
1816 * the right context.
1822 /* Then do a congruence class local flush */
1824 lwz r0,KVM_TLB_SETS(r6)
1826 li r7,0x400 /* IS field = 0b01 */
1828 sldi r0,r3,32 /* RS has PID */
1829 1: PPC_TLBIEL(7,0,2,1,1) /* RIC=2, PRS=1, R=1 */
1835 #endif /* CONFIG_PPC_RADIX_MMU */
1838 * POWER7/POWER8 guest -> host partition switch code.
1839 * We don't have to lock against tlbies but we do
1840 * have to coordinate the hardware threads.
1841 * Here STACK_SLOT_TRAP(r1) contains the trap number.
1843 kvmhv_switch_to_host:
1844 /* Secondary threads wait for primary to do partition switch */
1845 ld r5,HSTATE_KVM_VCORE(r13)
1846 ld r4,VCORE_KVM(r5) /* pointer to struct kvm */
1847 lbz r3,HSTATE_PTID(r13)
1851 13: lbz r3,VCORE_IN_GUEST(r5)
1857 /* Primary thread waits for all the secondaries to exit guest */
1858 15: lwz r3,VCORE_ENTRY_EXIT(r5)
1859 rlwinm r0,r3,32-8,0xff
1865 /* Did we actually switch to the guest at all? */
1866 lbz r6, VCORE_IN_GUEST(r5)
1870 /* Primary thread switches back to host partition */
1871 lwz r7,KVM_HOST_LPID(r4)
1873 ld r6,KVM_HOST_SDR1(r4)
1874 li r8,LPID_RSVD /* switch to reserved LPID */
1877 mtspr SPRN_SDR1,r6 /* switch to host page table */
1878 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
1883 /* DPDES and VTB are shared between threads */
1884 mfspr r7, SPRN_DPDES
1886 std r7, VCORE_DPDES(r5)
1887 std r8, VCORE_VTB(r5)
1888 /* clear DPDES so we don't get guest doorbells in the host */
1890 mtspr SPRN_DPDES, r8
1891 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
1893 /* Subtract timebase offset from timebase */
1894 ld r8, VCORE_TB_OFFSET_APPL(r5)
1898 std r0, VCORE_TB_OFFSET_APPL(r5)
1899 mftb r6 /* current guest timebase */
1901 mtspr SPRN_TBU40,r8 /* update upper 40 bits */
1902 mftb r7 /* check if lower 24 bits overflowed */
1907 addis r8,r8,0x100 /* if so, increment upper 40 bits */
1912 * If this is an HMI, we called kvmppc_realmode_hmi_handler
1913 * above, which may or may not have already called
1914 * kvmppc_subcore_exit_guest. Fortunately, all that
1915 * kvmppc_subcore_exit_guest does is clear a flag, so calling
1916 * it again here is benign even if kvmppc_realmode_hmi_handler
1917 * has already called it.
1919 bl kvmppc_subcore_exit_guest
1921 30: ld r5,HSTATE_KVM_VCORE(r13)
1922 ld r4,VCORE_KVM(r5) /* pointer to struct kvm */
1925 ld r0, VCORE_PCR(r5)
1926 LOAD_REG_IMMEDIATE(r6, PCR_MASK)
1931 /* Signal secondary CPUs to continue */
1933 stb r0,VCORE_IN_GUEST(r5)
1934 19: lis r8,0x7fff /* MAX_INT@h */
1939 /* On POWER9 with HPT-on-radix we need to wait for all other threads */
1940 ld r3, HSTATE_SPLIT_MODE(r13)
1943 lwz r8, KVM_SPLIT_DO_RESTORE(r3)
1946 bl kvmhv_p9_restore_lpcr
1950 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1951 ld r8,KVM_HOST_LPCR(r4)
1955 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
1956 /* Finish timing, if we have a vcpu */
1957 ld r4, HSTATE_KVM_VCPU(r13)
1961 bl kvmhv_accumulate_time
1964 /* Unset guest mode */
1965 li r0, KVM_GUEST_MODE_NONE
1966 stb r0, HSTATE_IN_GUEST(r13)
1968 lwz r12, STACK_SLOT_TRAP(r1) /* return trap # in r12 */
1969 ld r0, SFS+PPC_LR_STKOFF(r1)
1975 .global kvm_flush_link_stack
1976 kvm_flush_link_stack:
1977 /* Save LR into r0 */
1980 /* Flush the link stack. On Power8 it's up to 32 entries in size. */
1985 /* And on Power9 it's up to 64. */
1990 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
1996 kvmppc_guest_external:
1997 /* External interrupt, first check for host_ipi. If this is
1998 * set, we know the host wants us out so let's do it now
2003 * Restore the active volatile registers after returning from
2006 ld r9, HSTATE_KVM_VCPU(r13)
2007 li r12, BOOK3S_INTERRUPT_EXTERNAL
2010 * kvmppc_read_intr return codes:
2012 * Exit to host (r3 > 0)
2013 * 1 An interrupt is pending that needs to be handled by the host
2014 * Exit guest and return to host by branching to guest_exit_cont
2016 * 2 Passthrough that needs completion in the host
2017 * Exit guest and return to host by branching to guest_exit_cont
2018 * However, we also set r12 to BOOK3S_INTERRUPT_HV_RM_HARD
2019 * to indicate to the host to complete handling the interrupt
2021 * Before returning to guest, we check if any CPU is heading out
2022 * to the host and if so, we head out also. If no CPUs are heading
2023 * check return values <= 0.
2025 * Return to guest (r3 <= 0)
2026 * 0 No external interrupt is pending
2027 * -1 A guest wakeup IPI (which has now been cleared)
2028 * In either case, we return to guest to deliver any pending
2031 * -2 A PCI passthrough external interrupt was handled
2032 * (interrupt was delivered directly to guest)
2033 * Return to guest to deliver any pending guest interrupts.
2039 /* Return code = 2 */
2040 li r12, BOOK3S_INTERRUPT_HV_RM_HARD
2041 stw r12, VCPU_TRAP(r9)
2044 1: /* Return code <= 1 */
2048 /* Return code <= 0 */
2049 maybe_reenter_guest:
2050 ld r5, HSTATE_KVM_VCORE(r13)
2051 lwz r0, VCORE_ENTRY_EXIT(r5)
2054 blt deliver_guest_interrupt
2057 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2059 * Softpatch interrupt for transactional memory emulation cases
2060 * on POWER9 DD2.2. This is early in the guest exit path - we
2061 * haven't saved registers or done a treclaim yet.
2064 /* Save instruction image in HEIR */
2066 stw r3, VCPU_HEIR(r9)
2069 * The cases we want to handle here are those where the guest
2070 * is in real suspend mode and is trying to transition to
2071 * transactional mode.
2073 lbz r0, HSTATE_FAKE_SUSPEND(r13)
2074 cmpwi r0, 0 /* keep exiting guest if in fake suspend */
2076 rldicl r3, r11, 64 - MSR_TS_S_LG, 62
2077 cmpwi r3, 1 /* or if not in suspend state */
2080 /* Call C code to do the emulation */
2082 bl kvmhv_p9_tm_emulation_early
2084 ld r9, HSTATE_KVM_VCPU(r13)
2085 li r12, BOOK3S_INTERRUPT_HV_SOFTPATCH
2087 beq guest_exit_cont /* continue exiting if not handled */
2089 ld r11, VCPU_MSR(r9)
2090 b fast_interrupt_c_return /* go back to guest if handled */
2091 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
2094 * Check whether an HDSI is an HPTE not found fault or something else.
2095 * If it is an HPTE not found fault that is due to the guest accessing
2096 * a page that they have mapped but which we have paged out, then
2097 * we continue on with the guest exit path. In all other cases,
2098 * reflect the HDSI to the guest as a DSI.
2102 lbz r0, KVM_RADIX(r3)
2104 mfspr r6, SPRN_HDSISR
2106 /* Look for DSISR canary. If we find it, retry instruction */
2109 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2111 bne .Lradix_hdsi /* on radix, just save DAR/DSISR/ASDR */
2112 /* HPTE not found fault or protection fault? */
2113 andis. r0, r6, (DSISR_NOHPTE | DSISR_PROTFAULT)@h
2114 beq 1f /* if not, send it to the guest */
2115 andi. r0, r11, MSR_DR /* data relocation enabled? */
2118 mfspr r5, SPRN_ASDR /* on POWER9, use ASDR to get VSID */
2120 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2122 PPC_SLBFEE_DOT(R5, R0) /* if so, look up SLB */
2123 li r0, BOOK3S_INTERRUPT_DATA_SEGMENT
2124 bne 7f /* if no SLB entry found */
2125 4: std r4, VCPU_FAULT_DAR(r9)
2126 stw r6, VCPU_FAULT_DSISR(r9)
2128 /* Search the hash table. */
2129 mr r3, r9 /* vcpu pointer */
2130 li r7, 1 /* data fault */
2131 bl kvmppc_hpte_hv_fault
2132 ld r9, HSTATE_KVM_VCPU(r13)
2134 ld r11, VCPU_MSR(r9)
2135 li r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
2136 cmpdi r3, 0 /* retry the instruction */
2138 cmpdi r3, -1 /* handle in kernel mode */
2140 cmpdi r3, -2 /* MMIO emulation; need instr word */
2143 /* Synthesize a DSI (or DSegI) for the guest */
2144 ld r4, VCPU_FAULT_DAR(r9)
2146 1: li r0, BOOK3S_INTERRUPT_DATA_STORAGE
2147 mtspr SPRN_DSISR, r6
2148 7: mtspr SPRN_DAR, r4
2149 mtspr SPRN_SRR0, r10
2150 mtspr SPRN_SRR1, r11
2152 bl kvmppc_msr_interrupt
2153 fast_interrupt_c_return:
2154 6: ld r7, VCPU_CTR(r9)
2161 3: ld r5, VCPU_KVM(r9) /* not relocated, use VRMA */
2162 ld r5, KVM_VRMA_SLB_V(r5)
2165 /* If this is for emulated MMIO, load the instruction word */
2166 2: li r8, KVM_INST_FETCH_FAILED /* In case lwz faults */
2168 /* Set guest mode to 'jump over instruction' so if lwz faults
2169 * we'll just continue at the next IP. */
2170 li r0, KVM_GUEST_MODE_SKIP
2171 stb r0, HSTATE_IN_GUEST(r13)
2173 /* Do the access with MSR:DR enabled */
2175 ori r4, r3, MSR_DR /* Enable paging for data */
2180 /* Store the result */
2181 stw r8, VCPU_LAST_INST(r9)
2183 /* Unset guest mode. */
2184 li r0, KVM_GUEST_MODE_HOST_HV
2185 stb r0, HSTATE_IN_GUEST(r13)
2189 std r4, VCPU_FAULT_DAR(r9)
2190 stw r6, VCPU_FAULT_DSISR(r9)
2193 std r5, VCPU_FAULT_GPA(r9)
2197 * Similarly for an HISI, reflect it to the guest as an ISI unless
2198 * it is an HPTE not found fault for a page that we have paged out.
2202 lbz r0, KVM_RADIX(r3)
2204 bne .Lradix_hisi /* for radix, just save ASDR */
2205 andis. r0, r11, SRR1_ISI_NOPT@h
2207 andi. r0, r11, MSR_IR /* instruction relocation enabled? */
2210 mfspr r5, SPRN_ASDR /* on POWER9, use ASDR to get VSID */
2212 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2214 PPC_SLBFEE_DOT(R5, R0) /* if so, look up SLB */
2215 li r0, BOOK3S_INTERRUPT_INST_SEGMENT
2216 bne 7f /* if no SLB entry found */
2218 /* Search the hash table. */
2219 mr r3, r9 /* vcpu pointer */
2222 li r7, 0 /* instruction fault */
2223 bl kvmppc_hpte_hv_fault
2224 ld r9, HSTATE_KVM_VCPU(r13)
2226 ld r11, VCPU_MSR(r9)
2227 li r12, BOOK3S_INTERRUPT_H_INST_STORAGE
2228 cmpdi r3, 0 /* retry the instruction */
2229 beq fast_interrupt_c_return
2230 cmpdi r3, -1 /* handle in kernel mode */
2233 /* Synthesize an ISI (or ISegI) for the guest */
2235 1: li r0, BOOK3S_INTERRUPT_INST_STORAGE
2236 7: mtspr SPRN_SRR0, r10
2237 mtspr SPRN_SRR1, r11
2239 bl kvmppc_msr_interrupt
2240 b fast_interrupt_c_return
2242 3: ld r6, VCPU_KVM(r9) /* not relocated, use VRMA */
2243 ld r5, KVM_VRMA_SLB_V(r6)
2247 * Try to handle an hcall in real mode.
2248 * Returns to the guest if we handle it, or continues on up to
2249 * the kernel if we can't (i.e. if we don't have a handler for
2250 * it, or if the handler returns H_TOO_HARD).
2252 * r5 - r8 contain hcall args,
2253 * r9 = vcpu, r10 = pc, r11 = msr, r12 = trap, r13 = paca
2255 hcall_try_real_mode:
2256 ld r3,VCPU_GPR(R3)(r9)
2258 /* sc 1 from userspace - reflect to guest syscall */
2259 bne sc_1_fast_return
2260 /* sc 1 from nested guest - give it to L1 to handle */
2261 ld r0, VCPU_NESTED(r9)
2265 cmpldi r3,hcall_real_table_end - hcall_real_table
2267 /* See if this hcall is enabled for in-kernel handling */
2269 srdi r0, r3, 8 /* r0 = (r3 / 4) >> 6 */
2270 sldi r0, r0, 3 /* index into kvm->arch.enabled_hcalls[] */
2272 ld r0, KVM_ENABLED_HCALLS(r4)
2273 rlwinm r4, r3, 32-2, 0x3f /* r4 = (r3 / 4) & 0x3f */
2277 /* Get pointer to handler, if any, and call it */
2278 LOAD_REG_ADDR(r4, hcall_real_table)
2284 mr r3,r9 /* get vcpu pointer */
2285 ld r4,VCPU_GPR(R4)(r9)
2288 beq hcall_real_fallback
2289 ld r4,HSTATE_KVM_VCPU(r13)
2290 std r3,VCPU_GPR(R3)(r4)
2298 li r10, BOOK3S_INTERRUPT_SYSCALL
2299 bl kvmppc_msr_interrupt
2303 /* We've attempted a real mode hcall, but it's punted it back
2304 * to userspace. We need to restore some clobbered volatiles
2305 * before resuming the pass-it-to-qemu path */
2306 hcall_real_fallback:
2307 li r12,BOOK3S_INTERRUPT_SYSCALL
2308 ld r9, HSTATE_KVM_VCPU(r13)
2312 .globl hcall_real_table
2314 .long 0 /* 0 - unused */
2315 .long DOTSYM(kvmppc_h_remove) - hcall_real_table
2316 .long DOTSYM(kvmppc_h_enter) - hcall_real_table
2317 .long DOTSYM(kvmppc_h_read) - hcall_real_table
2318 .long DOTSYM(kvmppc_h_clear_mod) - hcall_real_table
2319 .long DOTSYM(kvmppc_h_clear_ref) - hcall_real_table
2320 .long DOTSYM(kvmppc_h_protect) - hcall_real_table
2321 #ifdef CONFIG_SPAPR_TCE_IOMMU
2322 .long DOTSYM(kvmppc_h_get_tce) - hcall_real_table
2323 .long DOTSYM(kvmppc_rm_h_put_tce) - hcall_real_table
2328 .long 0 /* 0x24 - H_SET_SPRG0 */
2329 .long DOTSYM(kvmppc_h_set_dabr) - hcall_real_table
2330 .long DOTSYM(kvmppc_rm_h_page_init) - hcall_real_table
2344 #ifdef CONFIG_KVM_XICS
2345 .long DOTSYM(kvmppc_rm_h_eoi) - hcall_real_table
2346 .long DOTSYM(kvmppc_rm_h_cppr) - hcall_real_table
2347 .long DOTSYM(kvmppc_rm_h_ipi) - hcall_real_table
2348 .long DOTSYM(kvmppc_rm_h_ipoll) - hcall_real_table
2349 .long DOTSYM(kvmppc_rm_h_xirr) - hcall_real_table
2351 .long 0 /* 0x64 - H_EOI */
2352 .long 0 /* 0x68 - H_CPPR */
2353 .long 0 /* 0x6c - H_IPI */
2354 .long 0 /* 0x70 - H_IPOLL */
2355 .long 0 /* 0x74 - H_XIRR */
2383 .long DOTSYM(kvmppc_h_cede) - hcall_real_table
2384 .long DOTSYM(kvmppc_rm_h_confer) - hcall_real_table
2400 .long DOTSYM(kvmppc_h_bulk_remove) - hcall_real_table
2404 .long DOTSYM(kvmppc_h_set_xdabr) - hcall_real_table
2405 #ifdef CONFIG_SPAPR_TCE_IOMMU
2406 .long DOTSYM(kvmppc_rm_h_stuff_tce) - hcall_real_table
2407 .long DOTSYM(kvmppc_rm_h_put_tce_indirect) - hcall_real_table
2523 #ifdef CONFIG_KVM_XICS
2524 .long DOTSYM(kvmppc_rm_h_xirr_x) - hcall_real_table
2526 .long 0 /* 0x2fc - H_XIRR_X*/
2528 .long DOTSYM(kvmppc_h_random) - hcall_real_table
2529 .globl hcall_real_table_end
2530 hcall_real_table_end:
2532 _GLOBAL(kvmppc_h_set_xdabr)
2533 EXPORT_SYMBOL_GPL(kvmppc_h_set_xdabr)
2534 andi. r0, r5, DABRX_USER | DABRX_KERNEL
2536 li r0, DABRX_USER | DABRX_KERNEL | DABRX_BTI
2539 6: li r3, H_PARAMETER
2542 _GLOBAL(kvmppc_h_set_dabr)
2543 EXPORT_SYMBOL_GPL(kvmppc_h_set_dabr)
2544 li r5, DABRX_USER | DABRX_KERNEL
2548 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2549 std r4,VCPU_DABR(r3)
2550 stw r5, VCPU_DABRX(r3)
2551 mtspr SPRN_DABRX, r5
2552 /* Work around P7 bug where DABR can get corrupted on mtspr */
2553 1: mtspr SPRN_DABR,r4
2562 LOAD_REG_ADDR(r11, dawr_force_enable)
2569 /* Emulate H_SET_DABR/X on P8 for the sake of compat mode guests */
2570 rlwimi r5, r4, 5, DAWRX_DR | DAWRX_DW
2571 rlwimi r5, r4, 2, DAWRX_WT
2573 std r4, VCPU_DAWR(r3)
2574 std r5, VCPU_DAWRX(r3)
2576 * If came in through the real mode hcall handler then it is necessary
2577 * to write the registers since the return path won't. Otherwise it is
2578 * sufficient to store then in the vcpu struct as they will be loaded
2579 * next time the vcpu is run.
2582 andi. r6, r6, MSR_DR /* in real mode? */
2585 mtspr SPRN_DAWRX, r5
2589 _GLOBAL(kvmppc_h_cede) /* r3 = vcpu pointer, r11 = msr, r13 = paca */
2591 std r11,VCPU_MSR(r3)
2593 stb r0,VCPU_CEDED(r3)
2594 sync /* order setting ceded vs. testing prodded */
2595 lbz r5,VCPU_PRODDED(r3)
2597 bne kvm_cede_prodded
2598 li r12,0 /* set trap to 0 to say hcall is handled */
2599 stw r12,VCPU_TRAP(r3)
2601 std r0,VCPU_GPR(R3)(r3)
2604 * Set our bit in the bitmask of napping threads unless all the
2605 * other threads are already napping, in which case we send this
2608 ld r5,HSTATE_KVM_VCORE(r13)
2609 lbz r6,HSTATE_PTID(r13)
2610 lwz r8,VCORE_ENTRY_EXIT(r5)
2614 addi r6,r5,VCORE_NAPPING_THREADS
2621 /* order napping_threads update vs testing entry_exit_map */
2624 stb r0,HSTATE_NAPPING(r13)
2625 lwz r7,VCORE_ENTRY_EXIT(r5)
2627 bge 33f /* another thread already exiting */
2630 * Although not specifically required by the architecture, POWER7
2631 * preserves the following registers in nap mode, even if an SMT mode
2632 * switch occurs: SLB entries, PURR, SPURR, AMOR, UAMOR, AMR, SPRG0-3,
2633 * DAR, DSISR, DABR, DABRX, DSCR, PMCx, MMCRx, SIAR, SDAR.
2635 /* Save non-volatile GPRs */
2636 std r14, VCPU_GPR(R14)(r3)
2637 std r15, VCPU_GPR(R15)(r3)
2638 std r16, VCPU_GPR(R16)(r3)
2639 std r17, VCPU_GPR(R17)(r3)
2640 std r18, VCPU_GPR(R18)(r3)
2641 std r19, VCPU_GPR(R19)(r3)
2642 std r20, VCPU_GPR(R20)(r3)
2643 std r21, VCPU_GPR(R21)(r3)
2644 std r22, VCPU_GPR(R22)(r3)
2645 std r23, VCPU_GPR(R23)(r3)
2646 std r24, VCPU_GPR(R24)(r3)
2647 std r25, VCPU_GPR(R25)(r3)
2648 std r26, VCPU_GPR(R26)(r3)
2649 std r27, VCPU_GPR(R27)(r3)
2650 std r28, VCPU_GPR(R28)(r3)
2651 std r29, VCPU_GPR(R29)(r3)
2652 std r30, VCPU_GPR(R30)(r3)
2653 std r31, VCPU_GPR(R31)(r3)
2658 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2660 * Branch around the call if both CPU_FTR_TM and
2661 * CPU_FTR_P9_TM_HV_ASSIST are off.
2665 END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
2667 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
2669 ld r3, HSTATE_KVM_VCPU(r13)
2671 li r5, 0 /* don't preserve non-vol regs */
2672 bl kvmppc_save_tm_hv
2678 * Set DEC to the smaller of DEC and HDEC, so that we wake
2679 * no later than the end of our timeslice (HDEC interrupts
2680 * don't wake us from nap).
2686 /* On P9 check whether the guest has large decrementer mode enabled */
2687 ld r6, HSTATE_KVM_VCORE(r13)
2688 ld r6, VCORE_LPCR(r6)
2689 andis. r6, r6, LPCR_LD@h
2691 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
2698 /* save expiry time of guest decrementer */
2700 ld r4, HSTATE_KVM_VCPU(r13)
2701 ld r5, HSTATE_KVM_VCORE(r13)
2702 ld r6, VCORE_TB_OFFSET_APPL(r5)
2703 subf r3, r6, r3 /* convert to host TB value */
2704 std r3, VCPU_DEC_EXPIRES(r4)
2706 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2707 ld r4, HSTATE_KVM_VCPU(r13)
2708 addi r3, r4, VCPU_TB_CEDE
2709 bl kvmhv_accumulate_time
2712 lis r3, LPCR_PECEDP@h /* Do wake on privileged doorbell */
2714 /* Go back to host stack */
2715 ld r1, HSTATE_HOST_R1(r13)
2718 * Take a nap until a decrementer or external or doobell interrupt
2719 * occurs, with PECE1 and PECE0 set in LPCR.
2720 * On POWER8, set PECEDH, and if we are ceding, also set PECEDP.
2721 * Also clear the runlatch bit before napping.
2724 mfspr r0, SPRN_CTRLF
2726 mtspr SPRN_CTRLT, r0
2729 stb r0,HSTATE_HWTHREAD_REQ(r13)
2731 ori r5,r5,LPCR_PECE0 | LPCR_PECE1
2733 ori r5, r5, LPCR_PECEDH
2734 rlwimi r5, r3, 0, LPCR_PECEDP
2735 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2737 kvm_nap_sequence: /* desired LPCR value in r5 */
2740 * PSSCR bits: exit criterion = 1 (wakeup based on LPCR at sreset)
2741 * enable state loss = 1 (allow SMT mode switch)
2742 * requested level = 0 (just stop dispatching)
2744 lis r3, (PSSCR_EC | PSSCR_ESL)@h
2745 /* Set LPCR_PECE_HVEE bit to enable wakeup by HV interrupts */
2746 li r4, LPCR_PECE_HVEE@higher
2750 li r3, PNV_THREAD_NAP
2751 ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_300)
2756 bl isa300_idle_stop_mayloss
2758 bl isa206_idle_insn_mayloss
2759 ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_300)
2761 mfspr r0, SPRN_CTRLF
2763 mtspr SPRN_CTRLT, r0
2768 stb r0, PACA_FTRACE_ENABLED(r13)
2770 li r0, KVM_HWTHREAD_IN_KVM
2771 stb r0, HSTATE_HWTHREAD_STATE(r13)
2773 lbz r0, HSTATE_NAPPING(r13)
2774 cmpwi r0, NAPPING_CEDE
2776 cmpwi r0, NAPPING_NOVCPU
2777 beq kvm_novcpu_wakeup
2778 cmpwi r0, NAPPING_UNSPLIT
2779 beq kvm_unsplit_wakeup
2780 twi 31,0,0 /* Nap state must not be zero */
2788 /* Woken by external or decrementer interrupt */
2790 /* get vcpu pointer */
2791 ld r4, HSTATE_KVM_VCPU(r13)
2793 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
2794 addi r3, r4, VCPU_TB_RMINTR
2795 bl kvmhv_accumulate_time
2798 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
2800 * Branch around the call if both CPU_FTR_TM and
2801 * CPU_FTR_P9_TM_HV_ASSIST are off.
2805 END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
2807 * NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
2811 li r5, 0 /* don't preserve non-vol regs */
2812 bl kvmppc_restore_tm_hv
2814 ld r4, HSTATE_KVM_VCPU(r13)
2818 /* load up FP state */
2821 /* Restore guest decrementer */
2822 ld r3, VCPU_DEC_EXPIRES(r4)
2823 ld r5, HSTATE_KVM_VCORE(r13)
2824 ld r6, VCORE_TB_OFFSET_APPL(r5)
2825 add r3, r3, r6 /* convert host TB to guest TB value */
2831 ld r14, VCPU_GPR(R14)(r4)
2832 ld r15, VCPU_GPR(R15)(r4)
2833 ld r16, VCPU_GPR(R16)(r4)
2834 ld r17, VCPU_GPR(R17)(r4)
2835 ld r18, VCPU_GPR(R18)(r4)
2836 ld r19, VCPU_GPR(R19)(r4)
2837 ld r20, VCPU_GPR(R20)(r4)
2838 ld r21, VCPU_GPR(R21)(r4)
2839 ld r22, VCPU_GPR(R22)(r4)
2840 ld r23, VCPU_GPR(R23)(r4)
2841 ld r24, VCPU_GPR(R24)(r4)
2842 ld r25, VCPU_GPR(R25)(r4)
2843 ld r26, VCPU_GPR(R26)(r4)
2844 ld r27, VCPU_GPR(R27)(r4)
2845 ld r28, VCPU_GPR(R28)(r4)
2846 ld r29, VCPU_GPR(R29)(r4)
2847 ld r30, VCPU_GPR(R30)(r4)
2848 ld r31, VCPU_GPR(R31)(r4)
2850 /* Check the wake reason in SRR1 to see why we got here */
2851 bl kvmppc_check_wake_reason
2854 * Restore volatile registers since we could have called a
2855 * C routine in kvmppc_check_wake_reason
2857 * r3 tells us whether we need to return to host or not
2858 * WARNING: it gets checked further down:
2859 * should not modify r3 until this check is done.
2861 ld r4, HSTATE_KVM_VCPU(r13)
2863 /* clear our bit in vcore->napping_threads */
2864 34: ld r5,HSTATE_KVM_VCORE(r13)
2865 lbz r7,HSTATE_PTID(r13)
2868 addi r6,r5,VCORE_NAPPING_THREADS
2874 stb r0,HSTATE_NAPPING(r13)
2876 /* See if the wake reason saved in r3 means we need to exit */
2877 stw r12, VCPU_TRAP(r4)
2881 b maybe_reenter_guest
2883 /* cede when already previously prodded case */
2886 stb r0,VCPU_PRODDED(r3)
2887 sync /* order testing prodded vs. clearing ceded */
2888 stb r0,VCPU_CEDED(r3)
2892 /* we've ceded but we want to give control to the host */
2894 ld r9, HSTATE_KVM_VCPU(r13)
2895 #ifdef CONFIG_KVM_XICS
2896 /* are we using XIVE with single escalation? */
2897 ld r10, VCPU_XIVE_ESC_VADDR(r9)
2900 li r6, XIVE_ESB_SET_PQ_00
2902 * If we still have a pending escalation, abort the cede,
2903 * and we must set PQ to 10 rather than 00 so that we don't
2904 * potentially end up with two entries for the escalation
2905 * interrupt in the XIVE interrupt queue. In that case
2906 * we also don't want to set xive_esc_on to 1 here in
2907 * case we race with xive_esc_irq().
2909 lbz r5, VCPU_XIVE_ESC_ON(r9)
2913 stb r0, VCPU_CEDED(r9)
2914 li r6, XIVE_ESB_SET_PQ_10
2917 stb r0, VCPU_XIVE_ESC_ON(r9)
2918 /* make sure store to xive_esc_on is seen before xive_esc_irq runs */
2920 5: /* Enable XIVE escalation */
2922 andi. r0, r0, MSR_DR /* in real mode? */
2926 1: ld r10, VCPU_XIVE_ESC_RADDR(r9)
2929 #endif /* CONFIG_KVM_XICS */
2930 3: b guest_exit_cont
2932 /* Try to do machine check recovery in real mode */
2933 machine_check_realmode:
2934 mr r3, r9 /* get vcpu pointer */
2935 bl kvmppc_realmode_machine_check
2937 /* all machine checks go to virtual mode for further handling */
2938 ld r9, HSTATE_KVM_VCPU(r13)
2939 li r12, BOOK3S_INTERRUPT_MACHINE_CHECK
2943 * Call C code to handle a HMI in real mode.
2944 * Only the primary thread does the call, secondary threads are handled
2945 * by calling hmi_exception_realmode() after kvmppc_hv_entry returns.
2946 * r9 points to the vcpu on entry
2949 lbz r0, HSTATE_PTID(r13)
2952 bl kvmppc_realmode_hmi_handler
2953 ld r9, HSTATE_KVM_VCPU(r13)
2954 li r12, BOOK3S_INTERRUPT_HMI
2958 * Check the reason we woke from nap, and take appropriate action.
2960 * 0 if nothing needs to be done
2961 * 1 if something happened that needs to be handled by the host
2962 * -1 if there was a guest wakeup (IPI or msgsnd)
2963 * -2 if we handled a PCI passthrough interrupt (returned by
2964 * kvmppc_read_intr only)
2966 * Also sets r12 to the interrupt vector for any interrupt that needs
2967 * to be handled now by the host (0x500 for external interrupt), or zero.
2968 * Modifies all volatile registers (since it may call a C function).
2969 * This routine calls kvmppc_read_intr, a C function, if an external
2970 * interrupt is pending.
2972 kvmppc_check_wake_reason:
2975 rlwinm r6, r6, 45-31, 0xf /* extract wake reason field (P8) */
2977 rlwinm r6, r6, 45-31, 0xe /* P7 wake reason field is 3 bits */
2978 ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_207S)
2979 cmpwi r6, 8 /* was it an external interrupt? */
2980 beq 7f /* if so, see what it was */
2983 cmpwi r6, 6 /* was it the decrementer? */
2986 cmpwi r6, 5 /* privileged doorbell? */
2988 cmpwi r6, 3 /* hypervisor doorbell? */
2990 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
2991 cmpwi r6, 0xa /* Hypervisor maintenance ? */
2993 li r3, 1 /* anything else, return 1 */
2996 /* hypervisor doorbell */
2997 3: li r12, BOOK3S_INTERRUPT_H_DOORBELL
3000 * Clear the doorbell as we will invoke the handler
3001 * explicitly in the guest exit path.
3003 lis r6, (PPC_DBELL_SERVER << (63-36))@h
3005 /* see if it's a host IPI */
3010 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
3011 lbz r0, HSTATE_HOST_IPI(r13)
3014 /* if not, return -1 */
3018 /* Woken up due to Hypervisor maintenance interrupt */
3019 4: li r12, BOOK3S_INTERRUPT_HMI
3023 /* external interrupt - create a stack frame so we can call C */
3025 std r0, PPC_LR_STKOFF(r1)
3026 stdu r1, -PPC_MIN_STKFRM(r1)
3029 li r12, BOOK3S_INTERRUPT_EXTERNAL
3034 * Return code of 2 means PCI passthrough interrupt, but
3035 * we need to return back to host to complete handling the
3036 * interrupt. Trap reason is expected in r12 by guest
3039 li r12, BOOK3S_INTERRUPT_HV_RM_HARD
3041 ld r0, PPC_MIN_STKFRM+PPC_LR_STKOFF(r1)
3042 addi r1, r1, PPC_MIN_STKFRM
3047 * Save away FP, VMX and VSX registers.
3049 * N.B. r30 and r31 are volatile across this function,
3050 * thus it is not callable from C.
3057 #ifdef CONFIG_ALTIVEC
3059 oris r8,r8,MSR_VEC@h
3060 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
3064 oris r8,r8,MSR_VSX@h
3065 END_FTR_SECTION_IFSET(CPU_FTR_VSX)
3068 addi r3,r3,VCPU_FPRS
3070 #ifdef CONFIG_ALTIVEC
3072 addi r3,r31,VCPU_VRS
3074 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
3076 mfspr r6,SPRN_VRSAVE
3077 stw r6,VCPU_VRSAVE(r31)
3082 * Load up FP, VMX and VSX registers
3084 * N.B. r30 and r31 are volatile across this function,
3085 * thus it is not callable from C.
3092 #ifdef CONFIG_ALTIVEC
3094 oris r8,r8,MSR_VEC@h
3095 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
3099 oris r8,r8,MSR_VSX@h
3100 END_FTR_SECTION_IFSET(CPU_FTR_VSX)
3103 addi r3,r4,VCPU_FPRS
3105 #ifdef CONFIG_ALTIVEC
3107 addi r3,r31,VCPU_VRS
3109 END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
3111 lwz r7,VCPU_VRSAVE(r31)
3112 mtspr SPRN_VRSAVE,r7
3117 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
3119 * Save transactional state and TM-related registers.
3120 * Called with r3 pointing to the vcpu struct and r4 containing
3121 * the guest MSR value.
3122 * r5 is non-zero iff non-volatile register state needs to be maintained.
3123 * If r5 == 0, this can modify all checkpointed registers, but
3124 * restores r1 and r2 before exit.
3126 _GLOBAL_TOC(kvmppc_save_tm_hv)
3127 EXPORT_SYMBOL_GPL(kvmppc_save_tm_hv)
3128 /* See if we need to handle fake suspend mode */
3131 END_FTR_SECTION_IFCLR(CPU_FTR_P9_TM_HV_ASSIST)
3133 lbz r0, HSTATE_FAKE_SUSPEND(r13) /* Were we fake suspended? */
3135 beq __kvmppc_save_tm
3137 /* The following code handles the fake_suspend = 1 case */
3139 std r0, PPC_LR_STKOFF(r1)
3140 stdu r1, -PPC_MIN_STKFRM(r1)
3145 rldimi r8, r0, MSR_TM_LG, 63-MSR_TM_LG
3148 rldicl. r8, r8, 64 - MSR_TS_S_LG, 62 /* Did we actually hrfid? */
3151 bl pnv_power9_force_smt4_catch
3152 END_FTR_SECTION_IFSET(CPU_FTR_P9_TM_XER_SO_BUG)
3155 /* We have to treclaim here because that's the only way to do S->N */
3156 li r3, TM_CAUSE_KVM_RESCHED
3160 * We were in fake suspend, so we are not going to save the
3161 * register state as the guest checkpointed state (since
3162 * we already have it), therefore we can now use any volatile GPR.
3163 * In fact treclaim in fake suspend state doesn't modify
3168 bl pnv_power9_force_smt4_release
3169 END_FTR_SECTION_IFSET(CPU_FTR_P9_TM_XER_SO_BUG)
3173 mfspr r3, SPRN_PSSCR
3174 /* PSSCR_FAKE_SUSPEND is a write-only bit, but clear it anyway */
3175 li r0, PSSCR_FAKE_SUSPEND
3177 mtspr SPRN_PSSCR, r3
3179 /* Don't save TEXASR, use value from last exit in real suspend state */
3180 ld r9, HSTATE_KVM_VCPU(r13)
3181 mfspr r5, SPRN_TFHAR
3182 mfspr r6, SPRN_TFIAR
3183 std r5, VCPU_TFHAR(r9)
3184 std r6, VCPU_TFIAR(r9)
3186 addi r1, r1, PPC_MIN_STKFRM
3187 ld r0, PPC_LR_STKOFF(r1)
3192 * Restore transactional state and TM-related registers.
3193 * Called with r3 pointing to the vcpu struct
3194 * and r4 containing the guest MSR value.
3195 * r5 is non-zero iff non-volatile register state needs to be maintained.
3196 * This potentially modifies all checkpointed registers.
3197 * It restores r1 and r2 from the PACA.
3199 _GLOBAL_TOC(kvmppc_restore_tm_hv)
3200 EXPORT_SYMBOL_GPL(kvmppc_restore_tm_hv)
3202 * If we are doing TM emulation for the guest on a POWER9 DD2,
3203 * then we don't actually do a trechkpt -- we either set up
3204 * fake-suspend mode, or emulate a TM rollback.
3207 b __kvmppc_restore_tm
3208 END_FTR_SECTION_IFCLR(CPU_FTR_P9_TM_HV_ASSIST)
3210 std r0, PPC_LR_STKOFF(r1)
3213 stb r0, HSTATE_FAKE_SUSPEND(r13)
3215 /* Turn on TM so we can restore TM SPRs */
3218 rldimi r5, r0, MSR_TM_LG, 63-MSR_TM_LG
3222 * The user may change these outside of a transaction, so they must
3223 * always be context switched.
3225 ld r5, VCPU_TFHAR(r3)
3226 ld r6, VCPU_TFIAR(r3)
3227 ld r7, VCPU_TEXASR(r3)
3228 mtspr SPRN_TFHAR, r5
3229 mtspr SPRN_TFIAR, r6
3230 mtspr SPRN_TEXASR, r7
3232 rldicl. r5, r4, 64 - MSR_TS_S_LG, 62
3233 beqlr /* TM not active in guest */
3235 /* Make sure the failure summary is set */
3236 oris r7, r7, (TEXASR_FS)@h
3237 mtspr SPRN_TEXASR, r7
3239 cmpwi r5, 1 /* check for suspended state */
3241 stb r5, HSTATE_FAKE_SUSPEND(r13)
3242 b 9f /* and return */
3243 10: stdu r1, -PPC_MIN_STKFRM(r1)
3244 /* guest is in transactional state, so simulate rollback */
3245 bl kvmhv_emulate_tm_rollback
3247 addi r1, r1, PPC_MIN_STKFRM
3248 9: ld r0, PPC_LR_STKOFF(r1)
3251 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
3254 * We come here if we get any exception or interrupt while we are
3255 * executing host real mode code while in guest MMU context.
3256 * r12 is (CR << 32) | vector
3257 * r13 points to our PACA
3258 * r12 is saved in HSTATE_SCRATCH0(r13)
3259 * ctr is saved in HSTATE_SCRATCH1(r13) if RELOCATABLE
3260 * r9 is saved in HSTATE_SCRATCH2(r13)
3261 * r13 is saved in HSPRG1
3262 * cfar is saved in HSTATE_CFAR(r13)
3263 * ppr is saved in HSTATE_PPR(r13)
3265 kvmppc_bad_host_intr:
3267 * Switch to the emergency stack, but start half-way down in
3268 * case we were already on it.
3272 ld r1, PACAEMERGSP(r13)
3273 subi r1, r1, THREAD_SIZE/2 + INT_FRAME_SIZE
3286 mfspr r3, SPRN_HSRR0
3287 mfspr r4, SPRN_HSRR1
3289 mfspr r6, SPRN_HDSISR
3291 1: mfspr r3, SPRN_SRR0
3294 mfspr r6, SPRN_DSISR
3299 ld r9, HSTATE_SCRATCH2(r13)
3300 ld r12, HSTATE_SCRATCH0(r13)
3305 ld r5, HSTATE_CFAR(r13)
3306 std r5, ORIG_GPR3(r1)
3308 #ifdef CONFIG_RELOCATABLE
3309 ld r4, HSTATE_SCRATCH1(r13)
3314 lbz r6, PACAIRQSOFTMASK(r13)
3320 LOAD_REG_IMMEDIATE(3, 0x7265677368657265)
3321 std r3, STACK_FRAME_OVERHEAD-16(r1)
3324 * On POWER9 do a minimal restore of the MMU and call C code,
3325 * which will print a message and panic.
3326 * XXX On POWER7 and POWER8, we just spin here since we don't
3327 * know what the other threads are doing (and we don't want to
3328 * coordinate with them) - but at least we now have register state
3329 * in memory that we might be able to look at from another CPU.
3333 END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
3334 ld r9, HSTATE_KVM_VCPU(r13)
3335 ld r10, VCPU_KVM(r9)
3340 mtspr SPRN_CIABR, r0
3341 mtspr SPRN_DAWRX, r0
3343 BEGIN_MMU_FTR_SECTION
3345 END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
3350 ld r8, PACA_SLBSHADOWPTR(r13)
3351 .rept SLB_NUM_BOLTED
3352 li r3, SLBSHADOW_SAVEAREA
3356 andis. r7, r5, SLB_ESID_V@h
3362 4: lwz r7, KVM_HOST_LPID(r10)
3365 ld r8, KVM_HOST_LPCR(r10)
3368 li r0, KVM_GUEST_MODE_NONE
3369 stb r0, HSTATE_IN_GUEST(r13)
3372 * Turn on the MMU and jump to C code
3376 addi r3, r3, 9f - 5b
3378 rldimi r3, r4, 62, 0 /* ensure 0xc000000000000000 bits are set */
3379 ld r4, PACAKMSR(r13)
3383 9: addi r3, r1, STACK_FRAME_OVERHEAD
3384 bl kvmppc_bad_interrupt
3388 * This mimics the MSR transition on IRQ delivery. The new guest MSR is taken
3389 * from VCPU_INTR_MSR and is modified based on the required TM state changes.
3390 * r11 has the guest MSR value (in/out)
3391 * r9 has a vcpu pointer (in)
3392 * r0 is used as a scratch register
3394 kvmppc_msr_interrupt:
3395 rldicl r0, r11, 64 - MSR_TS_S_LG, 62
3396 cmpwi r0, 2 /* Check if we are in transactional state.. */
3397 ld r11, VCPU_INTR_MSR(r9)
3399 /* ... if transactional, change to suspended */
3401 1: rldimi r11, r0, MSR_TS_S_LG, 63 - MSR_TS_T_LG
3405 * Load up guest PMU state. R3 points to the vcpu struct.
3407 _GLOBAL(kvmhv_load_guest_pmu)
3408 EXPORT_SYMBOL_GPL(kvmhv_load_guest_pmu)
3412 sldi r3, r3, 31 /* MMCR0_FC (freeze counters) bit */
3413 mtspr SPRN_MMCR0, r3 /* freeze all counters, disable ints */
3416 ld r3, VCPU_MMCR(r4)
3417 andi. r5, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
3418 cmpwi r5, MMCR0_PMAO
3419 beql kvmppc_fix_pmao
3420 END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
3421 lwz r3, VCPU_PMC(r4) /* always load up guest PMU registers */
3422 lwz r5, VCPU_PMC + 4(r4) /* to prevent information leak */
3423 lwz r6, VCPU_PMC + 8(r4)
3424 lwz r7, VCPU_PMC + 12(r4)
3425 lwz r8, VCPU_PMC + 16(r4)
3426 lwz r9, VCPU_PMC + 20(r4)
3433 ld r3, VCPU_MMCR(r4)
3434 ld r5, VCPU_MMCR + 8(r4)
3435 ld r6, VCPU_MMCR + 16(r4)
3436 ld r7, VCPU_SIAR(r4)
3437 ld r8, VCPU_SDAR(r4)
3438 mtspr SPRN_MMCR1, r5
3439 mtspr SPRN_MMCRA, r6
3443 ld r5, VCPU_MMCR + 24(r4)
3444 ld r6, VCPU_SIER(r4)
3445 mtspr SPRN_MMCR2, r5
3447 BEGIN_FTR_SECTION_NESTED(96)
3448 lwz r7, VCPU_PMC + 24(r4)
3449 lwz r8, VCPU_PMC + 28(r4)
3450 ld r9, VCPU_MMCR + 32(r4)
3451 mtspr SPRN_SPMC1, r7
3452 mtspr SPRN_SPMC2, r8
3453 mtspr SPRN_MMCRS, r9
3454 END_FTR_SECTION_NESTED(CPU_FTR_ARCH_300, 0, 96)
3455 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3456 mtspr SPRN_MMCR0, r3
3462 * Reload host PMU state saved in the PACA by kvmhv_save_host_pmu.
3464 _GLOBAL(kvmhv_load_host_pmu)
3465 EXPORT_SYMBOL_GPL(kvmhv_load_host_pmu)
3467 lbz r4, PACA_PMCINUSE(r13) /* is the host using the PMU? */
3469 beq 23f /* skip if not */
3471 ld r3, HSTATE_MMCR0(r13)
3472 andi. r4, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
3473 cmpwi r4, MMCR0_PMAO
3474 beql kvmppc_fix_pmao
3475 END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
3476 lwz r3, HSTATE_PMC1(r13)
3477 lwz r4, HSTATE_PMC2(r13)
3478 lwz r5, HSTATE_PMC3(r13)
3479 lwz r6, HSTATE_PMC4(r13)
3480 lwz r8, HSTATE_PMC5(r13)
3481 lwz r9, HSTATE_PMC6(r13)
3488 ld r3, HSTATE_MMCR0(r13)
3489 ld r4, HSTATE_MMCR1(r13)
3490 ld r5, HSTATE_MMCRA(r13)
3491 ld r6, HSTATE_SIAR(r13)
3492 ld r7, HSTATE_SDAR(r13)
3493 mtspr SPRN_MMCR1, r4
3494 mtspr SPRN_MMCRA, r5
3498 ld r8, HSTATE_MMCR2(r13)
3499 ld r9, HSTATE_SIER(r13)
3500 mtspr SPRN_MMCR2, r8
3502 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3503 mtspr SPRN_MMCR0, r3
3509 * Save guest PMU state into the vcpu struct.
3510 * r3 = vcpu, r4 = full save flag (PMU in use flag set in VPA)
3512 _GLOBAL(kvmhv_save_guest_pmu)
3513 EXPORT_SYMBOL_GPL(kvmhv_save_guest_pmu)
3518 * POWER8 seems to have a hardware bug where setting
3519 * MMCR0[PMAE] along with MMCR0[PMC1CE] and/or MMCR0[PMCjCE]
3520 * when some counters are already negative doesn't seem
3521 * to cause a performance monitor alert (and hence interrupt).
3522 * The effect of this is that when saving the PMU state,
3523 * if there is no PMU alert pending when we read MMCR0
3524 * before freezing the counters, but one becomes pending
3525 * before we read the counters, we lose it.
3526 * To work around this, we need a way to freeze the counters
3527 * before reading MMCR0. Normally, freezing the counters
3528 * is done by writing MMCR0 (to set MMCR0[FC]) which
3529 * unavoidably writes MMCR0[PMA0] as well. On POWER8,
3530 * we can also freeze the counters using MMCR2, by writing
3531 * 1s to all the counter freeze condition bits (there are
3532 * 9 bits each for 6 counters).
3534 li r3, -1 /* set all freeze bits */
3536 mfspr r10, SPRN_MMCR2
3537 mtspr SPRN_MMCR2, r3
3539 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3541 sldi r3, r3, 31 /* MMCR0_FC (freeze counters) bit */
3542 mfspr r4, SPRN_MMCR0 /* save MMCR0 */
3543 mtspr SPRN_MMCR0, r3 /* freeze all counters, disable ints */
3544 mfspr r6, SPRN_MMCRA
3545 /* Clear MMCRA in order to disable SDAR updates */
3547 mtspr SPRN_MMCRA, r7
3549 cmpwi r8, 0 /* did they ask for PMU stuff to be saved? */
3551 std r3, VCPU_MMCR(r9) /* if not, set saved MMCR0 to FC */
3553 21: mfspr r5, SPRN_MMCR1
3556 std r4, VCPU_MMCR(r9)
3557 std r5, VCPU_MMCR + 8(r9)
3558 std r6, VCPU_MMCR + 16(r9)
3560 std r10, VCPU_MMCR + 24(r9)
3561 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3562 std r7, VCPU_SIAR(r9)
3563 std r8, VCPU_SDAR(r9)
3570 stw r3, VCPU_PMC(r9)
3571 stw r4, VCPU_PMC + 4(r9)
3572 stw r5, VCPU_PMC + 8(r9)
3573 stw r6, VCPU_PMC + 12(r9)
3574 stw r7, VCPU_PMC + 16(r9)
3575 stw r8, VCPU_PMC + 20(r9)
3578 std r5, VCPU_SIER(r9)
3579 BEGIN_FTR_SECTION_NESTED(96)
3580 mfspr r6, SPRN_SPMC1
3581 mfspr r7, SPRN_SPMC2
3582 mfspr r8, SPRN_MMCRS
3583 stw r6, VCPU_PMC + 24(r9)
3584 stw r7, VCPU_PMC + 28(r9)
3585 std r8, VCPU_MMCR + 32(r9)
3587 mtspr SPRN_MMCRS, r4
3588 END_FTR_SECTION_NESTED(CPU_FTR_ARCH_300, 0, 96)
3589 END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
3593 * This works around a hardware bug on POWER8E processors, where
3594 * writing a 1 to the MMCR0[PMAO] bit doesn't generate a
3595 * performance monitor interrupt. Instead, when we need to have
3596 * an interrupt pending, we have to arrange for a counter to overflow.
3600 mtspr SPRN_MMCR2, r3
3601 lis r3, (MMCR0_PMXE | MMCR0_FCECE)@h
3602 ori r3, r3, MMCR0_PMCjCE | MMCR0_C56RUN
3603 mtspr SPRN_MMCR0, r3
3610 #ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
3612 * Start timing an activity
3613 * r3 = pointer to time accumulation struct, r4 = vcpu
3616 ld r5, HSTATE_KVM_VCORE(r13)
3617 ld r6, VCORE_TB_OFFSET_APPL(r5)
3619 subf r5, r6, r5 /* subtract current timebase offset */
3620 std r3, VCPU_CUR_ACTIVITY(r4)
3621 std r5, VCPU_ACTIVITY_START(r4)
3625 * Accumulate time to one activity and start another.
3626 * r3 = pointer to new time accumulation struct, r4 = vcpu
3628 kvmhv_accumulate_time:
3629 ld r5, HSTATE_KVM_VCORE(r13)
3630 ld r8, VCORE_TB_OFFSET_APPL(r5)
3631 ld r5, VCPU_CUR_ACTIVITY(r4)
3632 ld r6, VCPU_ACTIVITY_START(r4)
3633 std r3, VCPU_CUR_ACTIVITY(r4)
3635 subf r7, r8, r7 /* subtract current timebase offset */
3636 std r7, VCPU_ACTIVITY_START(r4)
3640 ld r8, TAS_SEQCOUNT(r5)
3643 std r8, TAS_SEQCOUNT(r5)
3645 ld r7, TAS_TOTAL(r5)
3647 std r7, TAS_TOTAL(r5)
3653 3: std r3, TAS_MIN(r5)
3659 std r8, TAS_SEQCOUNT(r5)
projects / linux.git / blob