kvm_call_hyp(__kvm_tlb_flush_vmid_ipa, kvm, ipa);
}
+/*
+ * D-Cache management functions. They take the page table entries by
+ * value, as they are flushing the cache using the kernel mapping (or
+ * kmap on 32bit).
+ */
+static void kvm_flush_dcache_pte(pte_t pte)
+{
+ __kvm_flush_dcache_pte(pte);
+}
+
+static void kvm_flush_dcache_pmd(pmd_t pmd)
+{
+ __kvm_flush_dcache_pmd(pmd);
+}
+
+static void kvm_flush_dcache_pud(pud_t pud)
+{
+ __kvm_flush_dcache_pud(pud);
+}
+
/**
* stage2_dissolve_pmd() - clear and flush huge PMD entry
* @kvm: pointer to kvm structure.
put_page(virt_to_page(pmd));
}
+/*
+ * Unmapping vs dcache management:
+ *
+ * If a guest maps certain memory pages as uncached, all writes will
+ * bypass the data cache and go directly to RAM. However, the CPUs
+ * can still speculate reads (not writes) and fill cache lines with
+ * data.
+ *
+ * Those cache lines will be *clean* cache lines though, so a
+ * clean+invalidate operation is equivalent to an invalidate
+ * operation, because no cache lines are marked dirty.
+ *
+ * Those clean cache lines could be filled prior to an uncached write
+ * by the guest, and the cache coherent IO subsystem would therefore
+ * end up writing old data to disk.
+ *
+ * This is why right after unmapping a page/section and invalidating
+ * the corresponding TLBs, we call kvm_flush_dcache_p*() to make sure
+ * the IO subsystem will never hit in the cache.
+ */
static void unmap_ptes(struct kvm *kvm, pmd_t *pmd,
phys_addr_t addr, phys_addr_t end)
{
start_pte = pte = pte_offset_kernel(pmd, addr);
do {
if (!pte_none(*pte)) {
+ pte_t old_pte = *pte;
+
kvm_set_pte(pte, __pte(0));
- put_page(virt_to_page(pte));
kvm_tlb_flush_vmid_ipa(kvm, addr);
+
+ /* No need to invalidate the cache for device mappings */
+ if ((pte_val(old_pte) & PAGE_S2_DEVICE) != PAGE_S2_DEVICE)
+ kvm_flush_dcache_pte(old_pte);
+
+ put_page(virt_to_page(pte));
}
} while (pte++, addr += PAGE_SIZE, addr != end);
next = kvm_pmd_addr_end(addr, end);
if (!pmd_none(*pmd)) {
if (kvm_pmd_huge(*pmd)) {
+ pmd_t old_pmd = *pmd;
+
pmd_clear(pmd);
kvm_tlb_flush_vmid_ipa(kvm, addr);
+
+ kvm_flush_dcache_pmd(old_pmd);
+
put_page(virt_to_page(pmd));
} else {
unmap_ptes(kvm, pmd, addr, next);
next = kvm_pud_addr_end(addr, end);
if (!pud_none(*pud)) {
if (pud_huge(*pud)) {
+ pud_t old_pud = *pud;
+
pud_clear(pud);
kvm_tlb_flush_vmid_ipa(kvm, addr);
+
+ kvm_flush_dcache_pud(old_pud);
+
put_page(virt_to_page(pud));
} else {
unmap_pmds(kvm, pud, addr, next);
pte = pte_offset_kernel(pmd, addr);
do {
- if (!pte_none(*pte)) {
- hva_t hva = gfn_to_hva(kvm, addr >> PAGE_SHIFT);
- kvm_flush_dcache_to_poc((void*)hva, PAGE_SIZE);
- }
+ if (!pte_none(*pte) &&
+ (pte_val(*pte) & PAGE_S2_DEVICE) != PAGE_S2_DEVICE)
+ kvm_flush_dcache_pte(*pte);
} while (pte++, addr += PAGE_SIZE, addr != end);
}
do {
next = kvm_pmd_addr_end(addr, end);
if (!pmd_none(*pmd)) {
- if (kvm_pmd_huge(*pmd)) {
- hva_t hva = gfn_to_hva(kvm, addr >> PAGE_SHIFT);
- kvm_flush_dcache_to_poc((void*)hva, PMD_SIZE);
- } else {
+ if (kvm_pmd_huge(*pmd))
+ kvm_flush_dcache_pmd(*pmd);
+ else
stage2_flush_ptes(kvm, pmd, addr, next);
- }
}
} while (pmd++, addr = next, addr != end);
}
do {
next = kvm_pud_addr_end(addr, end);
if (!pud_none(*pud)) {
- if (pud_huge(*pud)) {
- hva_t hva = gfn_to_hva(kvm, addr >> PAGE_SHIFT);
- kvm_flush_dcache_to_poc((void*)hva, PUD_SIZE);
- } else {
+ if (pud_huge(*pud))
+ kvm_flush_dcache_pud(*pud);
+ else
stage2_flush_pmds(kvm, pud, addr, next);
- }
}
} while (pud++, addr = next, addr != end);
}
* Go through the stage 2 page tables and invalidate any cache lines
* backing memory already mapped to the VM.
*/
-void stage2_flush_vm(struct kvm *kvm)
+static void stage2_flush_vm(struct kvm *kvm)
{
struct kvm_memslots *slots;
struct kvm_memory_slot *memslot;
kvm_mmu_write_protect_pt_masked(kvm, slot, gfn_offset, mask);
}
+static void coherent_cache_guest_page(struct kvm_vcpu *vcpu, pfn_t pfn,
+ unsigned long size, bool uncached)
+{
+ __coherent_cache_guest_page(vcpu, pfn, size, uncached);
+}
+
static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
struct kvm_memory_slot *memslot, unsigned long hva,
unsigned long fault_status)
kvm_set_s2pmd_writable(&new_pmd);
kvm_set_pfn_dirty(pfn);
}
- coherent_cache_guest_page(vcpu, hva & PMD_MASK, PMD_SIZE,
- fault_ipa_uncached);
+ coherent_cache_guest_page(vcpu, pfn, PMD_SIZE, fault_ipa_uncached);
ret = stage2_set_pmd_huge(kvm, memcache, fault_ipa, &new_pmd);
} else {
pte_t new_pte = pfn_pte(pfn, mem_type);
kvm_set_pfn_dirty(pfn);
mark_page_dirty(kvm, gfn);
}
- coherent_cache_guest_page(vcpu, hva, PAGE_SIZE,
- fault_ipa_uncached);
+ coherent_cache_guest_page(vcpu, pfn, PAGE_SIZE, fault_ipa_uncached);
ret = stage2_set_pte(kvm, memcache, fault_ipa, &new_pte, flags);
}
unmap_stage2_range(kvm, gpa, size);
spin_unlock(&kvm->mmu_lock);
}
+
+/*
+ * See note at ARMv7 ARM B1.14.4 (TL;DR: S/W ops are not easily virtualized).
+ *
+ * Main problems:
+ * - S/W ops are local to a CPU (not broadcast)
+ * - We have line migration behind our back (speculation)
+ * - System caches don't support S/W at all (damn!)
+ *
+ * In the face of the above, the best we can do is to try and convert
+ * S/W ops to VA ops. Because the guest is not allowed to infer the
+ * S/W to PA mapping, it can only use S/W to nuke the whole cache,
+ * which is a rather good thing for us.
+ *
+ * Also, it is only used when turning caches on/off ("The expected
+ * usage of the cache maintenance instructions that operate by set/way
+ * is associated with the cache maintenance instructions associated
+ * with the powerdown and powerup of caches, if this is required by
+ * the implementation.").
+ *
+ * We use the following policy:
+ *
+ * - If we trap a S/W operation, we enable VM trapping to detect
+ * caches being turned on/off, and do a full clean.
+ *
+ * - We flush the caches on both caches being turned on and off.
+ *
+ * - Once the caches are enabled, we stop trapping VM ops.
+ */
+void kvm_set_way_flush(struct kvm_vcpu *vcpu)
+{
+ unsigned long hcr = vcpu_get_hcr(vcpu);
+
+ /*
+ * If this is the first time we do a S/W operation
+ * (i.e. HCR_TVM not set) flush the whole memory, and set the
+ * VM trapping.
+ *
+ * Otherwise, rely on the VM trapping to wait for the MMU +
+ * Caches to be turned off. At that point, we'll be able to
+ * clean the caches again.
+ */
+ if (!(hcr & HCR_TVM)) {
+ trace_kvm_set_way_flush(*vcpu_pc(vcpu),
+ vcpu_has_cache_enabled(vcpu));
+ stage2_flush_vm(vcpu->kvm);
+ vcpu_set_hcr(vcpu, hcr | HCR_TVM);
+ }
+}
+
+void kvm_toggle_cache(struct kvm_vcpu *vcpu, bool was_enabled)
+{
+ bool now_enabled = vcpu_has_cache_enabled(vcpu);
+
+ /*
+ * If switching the MMU+caches on, need to invalidate the caches.
+ * If switching it off, need to clean the caches.
+ * Clean + invalidate does the trick always.
+ */
+ if (now_enabled != was_enabled)
+ stage2_flush_vm(vcpu->kvm);
+
+ /* Caches are now on, stop trapping VM ops (until a S/W op) */
+ if (now_enabled)
+ vcpu_set_hcr(vcpu, vcpu_get_hcr(vcpu) & ~HCR_TVM);
+
+ trace_kvm_toggle_cache(*vcpu_pc(vcpu), was_enabled, now_enabled);
+}
projects / karo-tx-linux.git / blobdiff