When running with independent threads off the main thread is responsible
for collecting secondary threads on the core and giving them vcpus to
run. These secondary threads spin in kvm_no_guest until they are given a
vcore to run and then take the kvmppc_hv_entry entry path from
kvm_secondary_got_guest. However this entry path doesn't load the guest
LPID for a radix guest, assuming it has already been loaded in
kvmppc_radix_check_need_tlb_flush() which would be called from
kvmppc_run_core when running in independent threads on mode. This means
that the guest runs with the host LPID which leads to an unhandleable mmu
fault giving:
KVM: Got unsupported MMU fault
error: kvm run failed Bad address
NIP c000000000028d50 LR c00000000004cad4 CTR 0000000000000000 XER 0000000000000000 CPU#0
MSR 8000000002089033 HID0 0000000000000000 HF 8000000000000000 iidx 3 didx 3
TB 00000000 00000000 DECR 0
GPR00 c00000000004cad4 c00000003e543a10 c00000000105ea00 000000000000c800
GPR04 0000000000000002 0000000000000000 0000000000000000 0000000000000000
GPR08 0000000000000000 000000000000a001 00000000a77784ea 0000000002001001
GPR12 0000000000000000 c000000001240000 c000000000010b50 0000000000000000
GPR16 0000000000000000 0000000000000000 0000000000000000 0000000000000000
GPR20 0000000000000000 0000000000000000 0000000000000000 c00000003fe7c800
GPR24 0000000000000000 0000000000000004 c0000000011044f8 0000000000000001
GPR28 0000000000000001 0000000000000000 c0000000011044fc 000000000000c336
CR 48000424 [ G L - - - G E G ] RES ffffffffffffffff
SRR0 c00000000000e708 SRR1 8000000002009033 PVR 00000000004e1202 VRSAVE 0000000000000000
SPRG0 0000000000000000 SPRG1 c000000001240000 SPRG2 0000000000000000 SPRG3 0000000000000000
SPRG4 0000000000000000 SPRG5 0000000000000000 SPRG6 0000000000000000 SPRG7 0000000000000000
HSRR0 0000000000000000 HSRR1 0000000000000000
CFAR 0000000000000000
LPCR 0000000003d6f41f
PTCR 0000000000000000 DAR 0000000000000000 DSISR 0000000000000000
Also we fail to perform the tlb flush if this vcpu has moved to this
core from another physical core which could result in stale tlb entries
being used.
To fix this we call kvmppc_radix_check_need_tlb_flush() from the
secondary entry path as well. This will both set the guest LPID and
check if the tlb needs to be flushed.
Signed-off-by: Suraj Jitindar Singh <sjitindarsingh@xxxxxxxxx>
---
arch/powerpc/include/asm/kvm_ppc.h | 2 ++
arch/powerpc/kernel/asm-offsets.c | 1 +
arch/powerpc/kvm/book3s_hv.c | 41 +++++----------------------------
arch/powerpc/kvm/book3s_hv_builtin.c | 32 +++++++++++++++++++++++++
arch/powerpc/kvm/book3s_hv_rmhandlers.S | 25 ++++++++++++--------
5 files changed, 57 insertions(+), 44 deletions(-)
diff --git a/arch/powerpc/include/asm/kvm_ppc.h b/arch/powerpc/include/asm/kvm_ppc.h
index eb0d79f0ca45..0defe17ff3b3 100644
--- a/arch/powerpc/include/asm/kvm_ppc.h
+++ b/arch/powerpc/include/asm/kvm_ppc.h
@@ -476,6 +476,8 @@ static inline void kvmppc_fast_vcpu_kick(struct kvm_vcpu *vcpu)
extern void kvm_hv_vm_activated(void);
extern void kvm_hv_vm_deactivated(void);
extern bool kvm_hv_mode_active(void);
+extern void kvmppc_radix_check_need_tlb_flush(struct kvm *kvm, int pcpu,
+ struct kvm_nested_guest *nested);
#else
static inline void __init kvm_cma_reserve(void)
diff --git a/arch/powerpc/kernel/asm-offsets.c b/arch/powerpc/kernel/asm-offsets.c
index 9ffc72ded73a..4af4401d50bd 100644
--- a/arch/powerpc/kernel/asm-offsets.c
+++ b/arch/powerpc/kernel/asm-offsets.c
@@ -557,6 +557,7 @@ int main(void)
OFFSET(VCPU_PSSCR, kvm_vcpu, arch.psscr);
OFFSET(VCPU_HFSCR, kvm_vcpu, arch.hfscr);
OFFSET(VCORE_ENTRY_EXIT, kvmppc_vcore, entry_exit_map);
+ OFFSET(VCORE_PCPU, kvmppc_vcore, pcpu);
OFFSET(VCORE_IN_GUEST, kvmppc_vcore, in_guest);
OFFSET(VCORE_NAPPING_THREADS, kvmppc_vcore, napping_threads);
OFFSET(VCORE_KVM, kvmppc_vcore, kvm);
diff --git a/arch/powerpc/kvm/book3s_hv.c b/arch/powerpc/kvm/book3s_hv.c
index 5a066fc299e1..1fd0920ae57d 100644
--- a/arch/powerpc/kvm/book3s_hv.c
+++ b/arch/powerpc/kvm/book3s_hv.c
@@ -2464,6 +2464,10 @@ static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
else
prev_cpu = vcpu->arch.prev_cpu;
+ /* TLB shared between threads so we use a single bit for all threads */
+ pcpu = cpu_first_thread_sibling(pcpu);
+ prev_cpu = cpu_first_thread_sibling(prev_cpu);
+
/*
* With radix, the guest can do TLB invalidations itself,
* and it could choose to use the local form (tlbiel) if
@@ -2477,9 +2481,7 @@ static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
* so we use a single bit in .need_tlb_flush for all 4 threads.
*/
if (prev_cpu != pcpu) {
- if (prev_cpu >= 0 &&
- cpu_first_thread_sibling(prev_cpu) !=
- cpu_first_thread_sibling(pcpu))
+ if (prev_cpu >= 0)
radix_flush_cpu(kvm, prev_cpu, vcpu);
if (nested)
nested->prev_cpu[vcpu->arch.nested_vcpu_id] = pcpu;
@@ -2488,37 +2490,6 @@ static void kvmppc_prepare_radix_vcpu(struct kvm_vcpu *vcpu, int pcpu)
}
}
-static void kvmppc_radix_check_need_tlb_flush(struct kvm *kvm, int pcpu,
- struct kvm_nested_guest *nested)
-{
- cpumask_t *need_tlb_flush;
- int lpid;
-
- if (!cpu_has_feature(CPU_FTR_HVMODE))
- return;
-
- if (cpu_has_feature(CPU_FTR_ARCH_300))
- pcpu &= ~0x3UL;
-
- if (nested) {
- lpid = nested->shadow_lpid;
- need_tlb_flush = &nested->need_tlb_flush;
- } else {
- lpid = kvm->arch.lpid;
- need_tlb_flush = &kvm->arch.need_tlb_flush;
- }
-
- mtspr(SPRN_LPID, lpid);
- isync();
- smp_mb();
-
- if (cpumask_test_cpu(pcpu, need_tlb_flush)) {
- radix__local_flush_tlb_lpid_guest(lpid);
- /* Clear the bit after the TLB flush */
- cpumask_clear_cpu(pcpu, need_tlb_flush);
- }
-}
-
static void kvmppc_start_thread(struct kvm_vcpu *vcpu, struct kvmppc_vcore *vc)
{
int cpu;
@@ -3044,7 +3015,7 @@ static noinline void kvmppc_run_core(struct kvmppc_vcore *vc)
if (kvm_is_radix(vc->kvm)) {
for (sub = 0; sub < core_info.n_subcores; ++sub)
for_each_runnable_thread(i, vcpu, core_info.vc[sub])
- kvmppc_prepare_radix_vcpu(vcpu, pcpu);
+ kvmppc_prepare_radix_vcpu(vcpu, pcpu + sub);
}
/*
diff --git a/arch/powerpc/kvm/book3s_hv_builtin.c b/arch/powerpc/kvm/book3s_hv_builtin.c
index a71e2fc00a4e..e1fb00e647a5 100644
--- a/arch/powerpc/kvm/book3s_hv_builtin.c
+++ b/arch/powerpc/kvm/book3s_hv_builtin.c
@@ -805,3 +805,35 @@ void kvmppc_guest_entry_inject_int(struct kvm_vcpu *vcpu)
vcpu->arch.doorbell_request = 0;
}
}
+
+void kvmppc_radix_check_need_tlb_flush(struct kvm *kvm, int pcpu,
+ struct kvm_nested_guest *nested)
+{
+ cpumask_t *need_tlb_flush;
+ int lpid;
+
+ if (!cpu_has_feature(CPU_FTR_HVMODE))
+ return;
+
+ if (cpu_has_feature(CPU_FTR_ARCH_300))
+ pcpu = cpu_first_thread_sibling(pcpu);
+
+ if (nested) {
+ lpid = nested->shadow_lpid;
+ need_tlb_flush = &nested->need_tlb_flush;
+ } else {
+ lpid = kvm->arch.lpid;
+ need_tlb_flush = &kvm->arch.need_tlb_flush;
+ }
+
+ mtspr(SPRN_LPID, lpid);
+ isync();
+ smp_mb();
+
+ if (cpumask_test_cpu(pcpu, need_tlb_flush)) {
+ radix__local_flush_tlb_lpid_guest(lpid);
+ /* Clear the bit after the TLB flush */
+ cpumask_clear_cpu(pcpu, need_tlb_flush);
+ }
+}
+EXPORT_SYMBOL_GPL(kvmppc_radix_check_need_tlb_flush);
diff --git a/arch/powerpc/kvm/book3s_hv_rmhandlers.S b/arch/powerpc/kvm/book3s_hv_rmhandlers.S
index 9b8d50a7cbaf..41628b67b88c 100644
--- a/arch/powerpc/kvm/book3s_hv_rmhandlers.S
+++ b/arch/powerpc/kvm/book3s_hv_rmhandlers.S
@@ -369,14 +369,6 @@ kvm_secondary_got_guest:
cmpdi r6, 0
beq 63f
BEGIN_FTR_SECTION
- ld r0, KVM_SPLIT_RPR(r6)
- mtspr SPRN_RPR, r0
- ld r0, KVM_SPLIT_PMMAR(r6)
- mtspr SPRN_PMMAR, r0
- ld r0, KVM_SPLIT_LDBAR(r6)
- mtspr SPRN_LDBAR, r0
- isync
-FTR_SECTION_ELSE
/* On P9 we use the split_info for coordinating LPCR changes */
lwz r4, KVM_SPLIT_DO_SET(r6)
cmpwi r4, 0
@@ -384,8 +376,23 @@ FTR_SECTION_ELSE
mr r3, r6
bl kvmhv_p9_set_lpcr
nop
+ b 2f
1:
-ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
+ ld r3, VCORE_KVM(r5) /* pointer to struct kvm */
+ ld r4, VCORE_PCPU(r5) /* physical cpu number */
+ li r5, 0 /* nested = NULL */
+ bl kvmppc_radix_check_need_tlb_flush
+ nop
+2:
+FTR_SECTION_ELSE
+ ld r0, KVM_SPLIT_RPR(r6)
+ mtspr SPRN_RPR, r0
+ ld r0, KVM_SPLIT_PMMAR(r6)
+ mtspr SPRN_PMMAR, r0
+ ld r0, KVM_SPLIT_LDBAR(r6)
+ mtspr SPRN_LDBAR, r0
+ isync
+ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_300)
63:
/* Order load of vcpu after load of vcore */
lwsync
--
2.13.6
