KVM: PPC: Book3S HV: Implement H_CLEAR_REF and H_CLEAR_MOD
authorPaul Mackerras <paulus@samba.org>
Wed, 24 Jun 2015 11:18:07 +0000 (21:18 +1000)
committerAlexander Graf <agraf@suse.de>
Sat, 22 Aug 2015 09:16:18 +0000 (11:16 +0200)
This adds implementations for the H_CLEAR_REF (test and clear reference
bit) and H_CLEAR_MOD (test and clear changed bit) hypercalls.

When clearing the reference or change bit in the guest view of the HPTE,
we also have to clear it in the real HPTE so that we can detect future
references or changes.  When we do so, we transfer the R or C bit value
to the rmap entry for the underlying host page so that kvm_age_hva_hv(),
kvm_test_age_hva_hv() and kvmppc_hv_get_dirty_log() know that the page
has been referenced and/or changed.

These hypercalls are not used by Linux guests.  These implementations
have been tested using a FreeBSD guest.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
arch/powerpc/kvm/book3s_hv_rm_mmu.c
arch/powerpc/kvm/book3s_hv_rmhandlers.S

index c7a3ab2652d0ea22a5fb6c7c0cd9e7324f439699..c1df9bb1e413a1ec76222a58cf0bc13c9bfdb280 100644 (file)
@@ -112,25 +112,38 @@ void kvmppc_update_rmap_change(unsigned long *rmap, unsigned long psize)
 }
 EXPORT_SYMBOL_GPL(kvmppc_update_rmap_change);
 
+/* Returns a pointer to the revmap entry for the page mapped by a HPTE */
+static unsigned long *revmap_for_hpte(struct kvm *kvm, unsigned long hpte_v,
+                                     unsigned long hpte_gr)
+{
+       struct kvm_memory_slot *memslot;
+       unsigned long *rmap;
+       unsigned long gfn;
+
+       gfn = hpte_rpn(hpte_gr, hpte_page_size(hpte_v, hpte_gr));
+       memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn);
+       if (!memslot)
+               return NULL;
+
+       rmap = real_vmalloc_addr(&memslot->arch.rmap[gfn - memslot->base_gfn]);
+       return rmap;
+}
+
 /* Remove this HPTE from the chain for a real page */
 static void remove_revmap_chain(struct kvm *kvm, long pte_index,
                                struct revmap_entry *rev,
                                unsigned long hpte_v, unsigned long hpte_r)
 {
        struct revmap_entry *next, *prev;
-       unsigned long gfn, ptel, head;
-       struct kvm_memory_slot *memslot;
+       unsigned long ptel, head;
        unsigned long *rmap;
        unsigned long rcbits;
 
        rcbits = hpte_r & (HPTE_R_R | HPTE_R_C);
        ptel = rev->guest_rpte |= rcbits;
-       gfn = hpte_rpn(ptel, hpte_page_size(hpte_v, ptel));
-       memslot = __gfn_to_memslot(kvm_memslots_raw(kvm), gfn);
-       if (!memslot)
+       rmap = revmap_for_hpte(kvm, hpte_v, ptel);
+       if (!rmap)
                return;
-
-       rmap = real_vmalloc_addr(&memslot->arch.rmap[gfn - memslot->base_gfn]);
        lock_rmap(rmap);
 
        head = *rmap & KVMPPC_RMAP_INDEX;
@@ -678,6 +691,105 @@ long kvmppc_h_read(struct kvm_vcpu *vcpu, unsigned long flags,
        return H_SUCCESS;
 }
 
+long kvmppc_h_clear_ref(struct kvm_vcpu *vcpu, unsigned long flags,
+                       unsigned long pte_index)
+{
+       struct kvm *kvm = vcpu->kvm;
+       __be64 *hpte;
+       unsigned long v, r, gr;
+       struct revmap_entry *rev;
+       unsigned long *rmap;
+       long ret = H_NOT_FOUND;
+
+       if (pte_index >= kvm->arch.hpt_npte)
+               return H_PARAMETER;
+
+       rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
+       hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
+       while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
+               cpu_relax();
+       v = be64_to_cpu(hpte[0]);
+       r = be64_to_cpu(hpte[1]);
+       if (!(v & (HPTE_V_VALID | HPTE_V_ABSENT)))
+               goto out;
+
+       gr = rev->guest_rpte;
+       if (rev->guest_rpte & HPTE_R_R) {
+               rev->guest_rpte &= ~HPTE_R_R;
+               note_hpte_modification(kvm, rev);
+       }
+       if (v & HPTE_V_VALID) {
+               gr |= r & (HPTE_R_R | HPTE_R_C);
+               if (r & HPTE_R_R) {
+                       kvmppc_clear_ref_hpte(kvm, hpte, pte_index);
+                       rmap = revmap_for_hpte(kvm, v, gr);
+                       if (rmap) {
+                               lock_rmap(rmap);
+                               *rmap |= KVMPPC_RMAP_REFERENCED;
+                               unlock_rmap(rmap);
+                       }
+               }
+       }
+       vcpu->arch.gpr[4] = gr;
+       ret = H_SUCCESS;
+ out:
+       unlock_hpte(hpte, v & ~HPTE_V_HVLOCK);
+       return ret;
+}
+
+long kvmppc_h_clear_mod(struct kvm_vcpu *vcpu, unsigned long flags,
+                       unsigned long pte_index)
+{
+       struct kvm *kvm = vcpu->kvm;
+       __be64 *hpte;
+       unsigned long v, r, gr;
+       struct revmap_entry *rev;
+       unsigned long *rmap;
+       long ret = H_NOT_FOUND;
+
+       if (pte_index >= kvm->arch.hpt_npte)
+               return H_PARAMETER;
+
+       rev = real_vmalloc_addr(&kvm->arch.revmap[pte_index]);
+       hpte = (__be64 *)(kvm->arch.hpt_virt + (pte_index << 4));
+       while (!try_lock_hpte(hpte, HPTE_V_HVLOCK))
+               cpu_relax();
+       v = be64_to_cpu(hpte[0]);
+       r = be64_to_cpu(hpte[1]);
+       if (!(v & (HPTE_V_VALID | HPTE_V_ABSENT)))
+               goto out;
+
+       gr = rev->guest_rpte;
+       if (gr & HPTE_R_C) {
+               rev->guest_rpte &= ~HPTE_R_C;
+               note_hpte_modification(kvm, rev);
+       }
+       if (v & HPTE_V_VALID) {
+               /* need to make it temporarily absent so C is stable */
+               hpte[0] |= cpu_to_be64(HPTE_V_ABSENT);
+               kvmppc_invalidate_hpte(kvm, hpte, pte_index);
+               r = be64_to_cpu(hpte[1]);
+               gr |= r & (HPTE_R_R | HPTE_R_C);
+               if (r & HPTE_R_C) {
+                       unsigned long psize = hpte_page_size(v, r);
+                       hpte[1] = cpu_to_be64(r & ~HPTE_R_C);
+                       eieio();
+                       rmap = revmap_for_hpte(kvm, v, gr);
+                       if (rmap) {
+                               lock_rmap(rmap);
+                               *rmap |= KVMPPC_RMAP_CHANGED;
+                               kvmppc_update_rmap_change(rmap, psize);
+                               unlock_rmap(rmap);
+                       }
+               }
+       }
+       vcpu->arch.gpr[4] = gr;
+       ret = H_SUCCESS;
+ out:
+       unlock_hpte(hpte, v & ~HPTE_V_HVLOCK);
+       return ret;
+}
+
 void kvmppc_invalidate_hpte(struct kvm *kvm, __be64 *hptep,
                        unsigned long pte_index)
 {
index db2427db44715eda7858969ccd7d306bb0ad6158..e3477667d49831fd3a3948a911705b32ecb258bd 100644 (file)
@@ -1904,8 +1904,8 @@ hcall_real_table:
        .long   DOTSYM(kvmppc_h_remove) - hcall_real_table
        .long   DOTSYM(kvmppc_h_enter) - hcall_real_table
        .long   DOTSYM(kvmppc_h_read) - hcall_real_table
-       .long   0               /* 0x10 - H_CLEAR_MOD */
-       .long   0               /* 0x14 - H_CLEAR_REF */
+       .long   DOTSYM(kvmppc_h_clear_mod) - hcall_real_table
+       .long   DOTSYM(kvmppc_h_clear_ref) - hcall_real_table
        .long   DOTSYM(kvmppc_h_protect) - hcall_real_table
        .long   DOTSYM(kvmppc_h_get_tce) - hcall_real_table
        .long   DOTSYM(kvmppc_h_put_tce) - hcall_real_table