x86/kvm: Add kexec support for SEV Live Migration.
authorAshish Kalra <ashish.kalra@amd.com>
Tue, 24 Aug 2021 11:07:45 +0000 (11:07 +0000)
committerPaolo Bonzini <pbonzini@redhat.com>
Thu, 11 Nov 2021 12:37:44 +0000 (07:37 -0500)
Reset the host's shared pages list related to kernel
specific page encryption status settings before we load a
new kernel by kexec. We cannot reset the complete
shared pages list here as we need to retain the
UEFI/OVMF firmware specific settings.

The host's shared pages list is maintained for the
guest to keep track of all unencrypted guest memory regions,
therefore we need to explicitly mark all shared pages as
encrypted again before rebooting into the new guest kernel.

Signed-off-by: Ashish Kalra <ashish.kalra@amd.com>
Reviewed-by: Steve Rutherford <srutherford@google.com>
Message-Id: <3e051424ab839ea470f88333273d7a185006754f.1629726117.git.ashish.kalra@amd.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
arch/x86/kernel/kvm.c

index a672a2ef65b34ac8fd9c374a783a2c512a0b0271..3910419fae6cbcaed87e6fd2444693c51757f9a2 100644 (file)
@@ -869,10 +869,35 @@ static void __init kvm_init_platform(void)
        if (sev_active() &&
            kvm_para_has_feature(KVM_FEATURE_MIGRATION_CONTROL)) {
                unsigned long nr_pages;
+               int i;
 
                pv_ops.mmu.notify_page_enc_status_changed =
                        kvm_sev_hc_page_enc_status;
 
+               /*
+                * Reset the host's shared pages list related to kernel
+                * specific page encryption status settings before we load a
+                * new kernel by kexec. Reset the page encryption status
+                * during early boot intead of just before kexec to avoid SMP
+                * races during kvm_pv_guest_cpu_reboot().
+                * NOTE: We cannot reset the complete shared pages list
+                * here as we need to retain the UEFI/OVMF firmware
+                * specific settings.
+                */
+
+               for (i = 0; i < e820_table->nr_entries; i++) {
+                       struct e820_entry *entry = &e820_table->entries[i];
+
+                       if (entry->type != E820_TYPE_RAM)
+                               continue;
+
+                       nr_pages = DIV_ROUND_UP(entry->size, PAGE_SIZE);
+
+                       kvm_sev_hypercall3(KVM_HC_MAP_GPA_RANGE, entry->addr,
+                                      nr_pages,
+                                      KVM_MAP_GPA_RANGE_ENCRYPTED | KVM_MAP_GPA_RANGE_PAGE_SZ_4K);
+               }
+
                /*
                 * Ensure that _bss_decrypted section is marked as decrypted in the
                 * shared pages list.