KVM: PPC: Book3S HV P9: Remove subcore HMI handling

On POWER9 and newer, rather than the complex HMI synchronisation and
subcore state, have each thread un-apply the guest TB offset before
calling into the early HMI handler.

This allows the subcore state to be avoided, including subcore enter
/ exit guest, which includes an expensive divide that shows up
slightly in profiles.

Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20211123095231.1036501-54-npiggin@gmail.com

diff --git a/arch/powerpc/include/asm/kvm_ppc.h b/arch/powerpc/include/asm/kvm_ppc.h
index 2b76d51..33db83b 100644 (file)
--- a/arch/powerpc/include/asm/kvm_ppc.h
+++ b/arch/powerpc/include/asm/kvm_ppc.h
@@ -759,6 +759,7 @@ void kvmppc_realmode_machine_check(struct kvm_vcpu *vcpu);
 void kvmppc_subcore_enter_guest(void);
 void kvmppc_subcore_exit_guest(void);
 long kvmppc_realmode_hmi_handler(void);
+long kvmppc_p9_realmode_hmi_handler(struct kvm_vcpu *vcpu);
 long kvmppc_h_enter(struct kvm_vcpu *vcpu, unsigned long flags,
                     long pte_index, unsigned long pteh, unsigned long ptel);
 long kvmppc_h_remove(struct kvm_vcpu *vcpu, unsigned long flags,

diff --git a/arch/powerpc/kvm/book3s_hv.c b/arch/powerpc/kvm/book3s_hv.c
index 214481e..98e90bd 100644 (file)
--- a/arch/powerpc/kvm/book3s_hv.c
+++ b/arch/powerpc/kvm/book3s_hv.c
@@ -4033,8 +4033,6 @@ static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
 
        vcpu->arch.ceded = 0;
 
-       kvmppc_subcore_enter_guest();
-
        vcpu_vpa_increment_dispatch(vcpu);
 
        if (kvmhv_on_pseries()) {
@@ -4087,8 +4085,6 @@ static int kvmhv_p9_guest_entry(struct kvm_vcpu *vcpu, u64 time_limit,
 
        vcpu_vpa_increment_dispatch(vcpu);
 
-       kvmppc_subcore_exit_guest();
-
        return trap;
 }
 
@@ -6102,9 +6098,11 @@ static int kvmppc_book3s_init_hv(void)
        if (r)
                return r;
 
-       r = kvm_init_subcore_bitmap();
-       if (r)
-               return r;
+       if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
+               r = kvm_init_subcore_bitmap();
+               if (r)
+                       return r;
+       }
 
        /*
         * We need a way of accessing the XICS interrupt controller,

diff --git a/arch/powerpc/kvm/book3s_hv_hmi.c b/arch/powerpc/kvm/book3s_hv_hmi.c
index 9af6604..1ec50c6 100644 (file)
--- a/arch/powerpc/kvm/book3s_hv_hmi.c
+++ b/arch/powerpc/kvm/book3s_hv_hmi.c
@@ -20,10 +20,15 @@ void wait_for_subcore_guest_exit(void)
 
        /*
         * NULL bitmap pointer indicates that KVM module hasn't
-        * been loaded yet and hence no guests are running.
+        * been loaded yet and hence no guests are running, or running
+        * on POWER9 or newer CPU.
+        *
         * If no KVM is in use, no need to co-ordinate among threads
         * as all of them will always be in host and no one is going
         * to modify TB other than the opal hmi handler.
+        *
+        * POWER9 and newer don't need this synchronisation.
+        *
         * Hence, just return from here.
         */
        if (!local_paca->sibling_subcore_state)

diff --git a/arch/powerpc/kvm/book3s_hv_p9_entry.c b/arch/powerpc/kvm/book3s_hv_p9_entry.c
index 72119bc..ebb4781 100644 (file)
--- a/arch/powerpc/kvm/book3s_hv_p9_entry.c
+++ b/arch/powerpc/kvm/book3s_hv_p9_entry.c
@@ -1013,7 +1013,7 @@ tm_return_to_guest:
                kvmppc_realmode_machine_check(vcpu);
 
        } else if (unlikely(trap == BOOK3S_INTERRUPT_HMI)) {
-               kvmppc_realmode_hmi_handler();
+               kvmppc_p9_realmode_hmi_handler(vcpu);
 
        } else if (trap == BOOK3S_INTERRUPT_H_EMUL_ASSIST) {
                vcpu->arch.emul_inst = mfspr(SPRN_HEIR);

diff --git a/arch/powerpc/kvm/book3s_hv_ras.c b/arch/powerpc/kvm/book3s_hv_ras.c
index d4bca93..ccfd969 100644 (file)
--- a/arch/powerpc/kvm/book3s_hv_ras.c
+++ b/arch/powerpc/kvm/book3s_hv_ras.c
@@ -136,6 +136,60 @@ void kvmppc_realmode_machine_check(struct kvm_vcpu *vcpu)
        vcpu->arch.mce_evt = mce_evt;
 }
 
+
+long kvmppc_p9_realmode_hmi_handler(struct kvm_vcpu *vcpu)
+{
+       struct kvmppc_vcore *vc = vcpu->arch.vcore;
+       long ret = 0;
+
+       /*
+        * Unapply and clear the offset first. That way, if the TB was not
+        * resynced then it will remain in host-offset, and if it was resynced
+        * then it is brought into host-offset. Then the tb offset is
+        * re-applied before continuing with the KVM exit.
+        *
+        * This way, we don't need to actually know whether not OPAL resynced
+        * the timebase or do any of the complicated dance that the P7/8
+        * path requires.
+        */
+       if (vc->tb_offset_applied) {
+               u64 new_tb = mftb() - vc->tb_offset_applied;
+               mtspr(SPRN_TBU40, new_tb);
+               if ((mftb() & 0xffffff) < (new_tb & 0xffffff)) {
+                       new_tb += 0x1000000;
+                       mtspr(SPRN_TBU40, new_tb);
+               }
+               vc->tb_offset_applied = 0;
+       }
+
+       local_paca->hmi_irqs++;
+
+       if (hmi_handle_debugtrig(NULL) >= 0) {
+               ret = 1;
+               goto out;
+       }
+
+       if (ppc_md.hmi_exception_early)
+               ppc_md.hmi_exception_early(NULL);
+
+out:
+       if (vc->tb_offset) {
+               u64 new_tb = mftb() + vc->tb_offset;
+               mtspr(SPRN_TBU40, new_tb);
+               if ((mftb() & 0xffffff) < (new_tb & 0xffffff)) {
+                       new_tb += 0x1000000;
+                       mtspr(SPRN_TBU40, new_tb);
+               }
+               vc->tb_offset_applied = vc->tb_offset;
+       }
+
+       return ret;
+}
+
+/*
+ * The following subcore HMI handling is all only for pre-POWER9 CPUs.
+ */
+
 /* Check if dynamic split is in force and return subcore size accordingly. */
 static inline int kvmppc_cur_subcore_size(void)
 {