KVM: PPC: KVM PV guest stubs
authorAlexander Graf <agraf@suse.de>
Thu, 29 Jul 2010 12:47:58 +0000 (14:47 +0200)
committerAvi Kivity <avi@redhat.com>
Sun, 24 Oct 2010 08:50:51 +0000 (10:50 +0200)
We will soon start and replace instructions from the text section with
other, paravirtualized versions. To ease the readability of those patches
I split out the generic looping and magic page mapping code out.

This patch still only contains stubs. But at least it loops through the
text section :).

Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
arch/powerpc/kernel/kvm.c

index a5ece71..e93366f 100644 (file)
 #define KVM_MAGIC_PAGE         (-4096L)
 #define magic_var(x) KVM_MAGIC_PAGE + offsetof(struct kvm_vcpu_arch_shared, x)
 
+#define KVM_MASK_RT            0x03e00000
+
+static bool kvm_patching_worked = true;
+
+static inline void kvm_patch_ins(u32 *inst, u32 new_inst)
+{
+       *inst = new_inst;
+       flush_icache_range((ulong)inst, (ulong)inst + 4);
+}
+
+static void kvm_map_magic_page(void *data)
+{
+       kvm_hypercall2(KVM_HC_PPC_MAP_MAGIC_PAGE,
+                      KVM_MAGIC_PAGE,  /* Physical Address */
+                      KVM_MAGIC_PAGE); /* Effective Address */
+}
+
+static void kvm_check_ins(u32 *inst)
+{
+       u32 _inst = *inst;
+       u32 inst_no_rt = _inst & ~KVM_MASK_RT;
+       u32 inst_rt = _inst & KVM_MASK_RT;
+
+       switch (inst_no_rt) {
+       }
+
+       switch (_inst) {
+       }
+}
+
+static void kvm_use_magic_page(void)
+{
+       u32 *p;
+       u32 *start, *end;
+       u32 tmp;
+
+       /* Tell the host to map the magic page to -4096 on all CPUs */
+       on_each_cpu(kvm_map_magic_page, NULL, 1);
+
+       /* Quick self-test to see if the mapping works */
+       if (__get_user(tmp, (u32*)KVM_MAGIC_PAGE)) {
+               kvm_patching_worked = false;
+               return;
+       }
+
+       /* Now loop through all code and find instructions */
+       start = (void*)_stext;
+       end = (void*)_etext;
+
+       for (p = start; p < end; p++)
+               kvm_check_ins(p);
+
+       printk(KERN_INFO "KVM: Live patching for a fast VM %s\n",
+                        kvm_patching_worked ? "worked" : "failed");
+}
+
 unsigned long kvm_hypercall(unsigned long *in,
                            unsigned long *out,
                            unsigned long nr)
@@ -69,3 +125,42 @@ unsigned long kvm_hypercall(unsigned long *in,
        return r3;
 }
 EXPORT_SYMBOL_GPL(kvm_hypercall);
+
+static int kvm_para_setup(void)
+{
+       extern u32 kvm_hypercall_start;
+       struct device_node *hyper_node;
+       u32 *insts;
+       int len, i;
+
+       hyper_node = of_find_node_by_path("/hypervisor");
+       if (!hyper_node)
+               return -1;
+
+       insts = (u32*)of_get_property(hyper_node, "hcall-instructions", &len);
+       if (len % 4)
+               return -1;
+       if (len > (4 * 4))
+               return -1;
+
+       for (i = 0; i < (len / 4); i++)
+               kvm_patch_ins(&(&kvm_hypercall_start)[i], insts[i]);
+
+       return 0;
+}
+
+static int __init kvm_guest_init(void)
+{
+       if (!kvm_para_available())
+               return 0;
+
+       if (kvm_para_setup())
+               return 0;
+
+       if (kvm_para_has_feature(KVM_FEATURE_MAGIC_PAGE))
+               kvm_use_magic_page();
+
+       return 0;
+}
+
+postcore_initcall(kvm_guest_init);