On 5/30/2024 7:16 PM, Pankaj Gupta wrote:
[...]
+/*
+ * Currently the handling here only supports use of KVM_HC_MAP_GPA_RANGE
+ * to service guest-initiated memory attribute update requests so that
+ * KVM_SET_MEMORY_ATTRIBUTES can update whether or not a page should be
+ * backed by the private memory pool provided by guest_memfd, and as such
+ * is only applicable to guest_memfd-backed guests (e.g. SNP/TDX).
+ *
+ * Other other use-cases for KVM_HC_MAP_GPA_RANGE, such as for SEV live
^
extra "other"?
+ * migration, are not implemented here currently.
+ *
+ * For the guest_memfd use-case, these exits will generally be synthesized
+ * by KVM based on platform-specific hypercalls, like GHCB requests in the
+ * case of SEV-SNP, and not issued directly within the guest though the
+ * KVM_HC_MAP_GPA_RANGE hypercall. So in this case, KVM_HC_MAP_GPA_RANGE is
+ * not actually advertised to guests via the KVM CPUID feature bit, as
+ * opposed to SEV live migration where it would be. Since it is unlikely the
+ * SEV live migration use-case would be useful for guest-memfd backed guests,
+ * because private/shared page tracking is already provided through other
+ * means, these 2 use-cases should be treated as being mutually-exclusive.
+ */
+static int kvm_handle_hc_map_gpa_range(struct kvm_run *run)
+{
+ uint64_t gpa, size, attributes;
+
+ if (!machine_require_guest_memfd(current_machine))
+ return -EINVAL;
+
+ gpa = run->hypercall.args[0];
+ size = run->hypercall.args[1] * TARGET_PAGE_SIZE;
+ attributes = run->hypercall.args[2];
+
+ trace_kvm_hc_map_gpa_range(gpa, size, attributes, run->hypercall.flags);
+
+ return kvm_convert_memory(gpa, size, attributes & KVM_MAP_GPA_RANGE_ENCRYPTED);
run->hypercall.ret should be updated accordingly.
At least for successful case.
For failure case, QEMU will shutdown the VM, is it the expected behavior?
+}
+
+static int kvm_handle_hypercall(struct kvm_run *run)
+{
+ if (run->hypercall.nr == KVM_HC_MAP_GPA_RANGE)
+ return kvm_handle_hc_map_gpa_range(run);
+
+ return -EINVAL;
+}
+
[...]
