]> www.pilppa.org Git - linux-2.6-omap-h63xx.git/blobdiff - drivers/kvm/kvm_main.c
Driver core: change sysdev classes to use dynamic kobject names
[linux-2.6-omap-h63xx.git] / drivers / kvm / kvm_main.c
index 9bfa1bcd26e9e1b4fd6780fc425c1d1751c1e130..c0f372f1d761312bc648d9deccd8e74609fbc513 100644 (file)
@@ -38,6 +38,7 @@
 #include <linux/cpumask.h>
 #include <linux/smp.h>
 #include <linux/anon_inodes.h>
+#include <linux/profile.h>
 
 #include <asm/processor.h>
 #include <asm/msr.h>
@@ -197,21 +198,15 @@ static void vcpu_put(struct kvm_vcpu *vcpu)
 
 static void ack_flush(void *_completed)
 {
-       atomic_t *completed = _completed;
-
-       atomic_inc(completed);
 }
 
 void kvm_flush_remote_tlbs(struct kvm *kvm)
 {
-       int i, cpu, needed;
+       int i, cpu;
        cpumask_t cpus;
        struct kvm_vcpu *vcpu;
-       atomic_t completed;
 
-       atomic_set(&completed, 0);
        cpus_clear(cpus);
-       needed = 0;
        for (i = 0; i < KVM_MAX_VCPUS; ++i) {
                vcpu = kvm->vcpus[i];
                if (!vcpu)
@@ -220,23 +215,9 @@ void kvm_flush_remote_tlbs(struct kvm *kvm)
                        continue;
                cpu = vcpu->cpu;
                if (cpu != -1 && cpu != raw_smp_processor_id())
-                       if (!cpu_isset(cpu, cpus)) {
-                               cpu_set(cpu, cpus);
-                               ++needed;
-                       }
-       }
-
-       /*
-        * We really want smp_call_function_mask() here.  But that's not
-        * available, so ipi all cpus in parallel and wait for them
-        * to complete.
-        */
-       for (cpu = first_cpu(cpus); cpu != NR_CPUS; cpu = next_cpu(cpu, cpus))
-               smp_call_function_single(cpu, ack_flush, &completed, 1, 0);
-       while (atomic_read(&completed) != needed) {
-               cpu_relax();
-               barrier();
+                       cpu_set(cpu, cpus);
        }
+       smp_call_function_mask(cpus, ack_flush, NULL, 1);
 }
 
 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
@@ -1207,8 +1188,7 @@ int emulate_invlpg(struct kvm_vcpu *vcpu, gva_t address)
 
 int emulate_clts(struct kvm_vcpu *vcpu)
 {
-       vcpu->cr0 &= ~X86_CR0_TS;
-       kvm_x86_ops->set_cr0(vcpu, vcpu->cr0);
+       kvm_x86_ops->set_cr0(vcpu, vcpu->cr0 & ~X86_CR0_TS);
        return X86EMUL_CONTINUE;
 }
 
@@ -1239,25 +1219,25 @@ int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr, unsigned long value)
        return X86EMUL_CONTINUE;
 }
 
-static void report_emulation_failure(struct x86_emulate_ctxt *ctxt)
+void kvm_report_emulation_failure(struct kvm_vcpu *vcpu, const char *context)
 {
        static int reported;
        u8 opcodes[4];
-       unsigned long rip = ctxt->vcpu->rip;
+       unsigned long rip = vcpu->rip;
        unsigned long rip_linear;
 
-       rip_linear = rip + get_segment_base(ctxt->vcpu, VCPU_SREG_CS);
+       rip_linear = rip + get_segment_base(vcpu, VCPU_SREG_CS);
 
        if (reported)
                return;
 
-       emulator_read_std(rip_linear, (void *)opcodes, 4, ctxt->vcpu);
+       emulator_read_std(rip_linear, (void *)opcodes, 4, vcpu);
 
-       printk(KERN_ERR "emulation failed but !mmio_needed?"
-              " rip %lx %02x %02x %02x %02x\n",
-              rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
+       printk(KERN_ERR "emulation failed (%s) rip %lx %02x %02x %02x %02x\n",
+              context, rip, opcodes[0], opcodes[1], opcodes[2], opcodes[3]);
        reported = 1;
 }
+EXPORT_SYMBOL_GPL(kvm_report_emulation_failure);
 
 struct x86_emulate_ops emulate_ops = {
        .read_std            = emulator_read_std,
@@ -1322,7 +1302,7 @@ int emulate_instruction(struct kvm_vcpu *vcpu,
                if (kvm_mmu_unprotect_page_virt(vcpu, cr2))
                        return EMULATE_DONE;
                if (!vcpu->mmio_needed) {
-                       report_emulation_failure(&emulate_ctxt);
+                       kvm_report_emulation_failure(vcpu, "mmio");
                        return EMULATE_FAIL;
                }
                return EMULATE_DO_MMIO;
@@ -1814,8 +1794,6 @@ static int complete_pio(struct kvm_vcpu *vcpu)
        io->count -= io->cur_count;
        io->cur_count = 0;
 
-       if (!io->count)
-               kvm_x86_ops->skip_emulated_instruction(vcpu);
        return 0;
 }
 
@@ -1875,6 +1853,8 @@ int kvm_emulate_pio (struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
        memcpy(vcpu->pio_data, &vcpu->regs[VCPU_REGS_RAX], 4);
        kvm_x86_ops->decache_regs(vcpu);
 
+       kvm_x86_ops->skip_emulated_instruction(vcpu);
+
        pio_dev = vcpu_find_pio_dev(vcpu, port);
        if (pio_dev) {
                kernel_pio(pio_dev, vcpu, vcpu->pio_data);
@@ -1937,6 +1917,9 @@ int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
        vcpu->run->io.count = now;
        vcpu->pio.cur_count = now;
 
+       if (vcpu->pio.cur_count == vcpu->pio.count)
+               kvm_x86_ops->skip_emulated_instruction(vcpu);
+
        for (i = 0; i < nr_pages; ++i) {
                mutex_lock(&vcpu->kvm->lock);
                page = gva_to_page(vcpu, address + i * PAGE_SIZE);
@@ -1970,6 +1953,138 @@ int kvm_emulate_pio_string(struct kvm_vcpu *vcpu, struct kvm_run *run, int in,
 }
 EXPORT_SYMBOL_GPL(kvm_emulate_pio_string);
 
+/*
+ * Check if userspace requested an interrupt window, and that the
+ * interrupt window is open.
+ *
+ * No need to exit to userspace if we already have an interrupt queued.
+ */
+static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
+                                         struct kvm_run *kvm_run)
+{
+       return (!vcpu->irq_summary &&
+               kvm_run->request_interrupt_window &&
+               vcpu->interrupt_window_open &&
+               (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF));
+}
+
+static void post_kvm_run_save(struct kvm_vcpu *vcpu,
+                             struct kvm_run *kvm_run)
+{
+       kvm_run->if_flag = (kvm_x86_ops->get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
+       kvm_run->cr8 = get_cr8(vcpu);
+       kvm_run->apic_base = kvm_get_apic_base(vcpu);
+       if (irqchip_in_kernel(vcpu->kvm))
+               kvm_run->ready_for_interrupt_injection = 1;
+       else
+               kvm_run->ready_for_interrupt_injection =
+                                       (vcpu->interrupt_window_open &&
+                                        vcpu->irq_summary == 0);
+}
+
+static int __vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
+{
+       int r;
+
+       if (unlikely(vcpu->mp_state == VCPU_MP_STATE_SIPI_RECEIVED)) {
+               printk("vcpu %d received sipi with vector # %x\n",
+                      vcpu->vcpu_id, vcpu->sipi_vector);
+               kvm_lapic_reset(vcpu);
+               kvm_x86_ops->vcpu_reset(vcpu);
+               vcpu->mp_state = VCPU_MP_STATE_RUNNABLE;
+       }
+
+preempted:
+       if (vcpu->guest_debug.enabled)
+               kvm_x86_ops->guest_debug_pre(vcpu);
+
+again:
+       r = kvm_mmu_reload(vcpu);
+       if (unlikely(r))
+               goto out;
+
+       preempt_disable();
+
+       kvm_x86_ops->prepare_guest_switch(vcpu);
+       kvm_load_guest_fpu(vcpu);
+
+       local_irq_disable();
+
+       if (signal_pending(current)) {
+               local_irq_enable();
+               preempt_enable();
+               r = -EINTR;
+               kvm_run->exit_reason = KVM_EXIT_INTR;
+               ++vcpu->stat.signal_exits;
+               goto out;
+       }
+
+       if (irqchip_in_kernel(vcpu->kvm))
+               kvm_x86_ops->inject_pending_irq(vcpu);
+       else if (!vcpu->mmio_read_completed)
+               kvm_x86_ops->inject_pending_vectors(vcpu, kvm_run);
+
+       vcpu->guest_mode = 1;
+       kvm_guest_enter();
+
+       if (vcpu->requests)
+               if (test_and_clear_bit(KVM_TLB_FLUSH, &vcpu->requests))
+                       kvm_x86_ops->tlb_flush(vcpu);
+
+       kvm_x86_ops->run(vcpu, kvm_run);
+
+       vcpu->guest_mode = 0;
+       local_irq_enable();
+
+       ++vcpu->stat.exits;
+
+       /*
+        * We must have an instruction between local_irq_enable() and
+        * kvm_guest_exit(), so the timer interrupt isn't delayed by
+        * the interrupt shadow.  The stat.exits increment will do nicely.
+        * But we need to prevent reordering, hence this barrier():
+        */
+       barrier();
+
+       kvm_guest_exit();
+
+       preempt_enable();
+
+       /*
+        * Profile KVM exit RIPs:
+        */
+       if (unlikely(prof_on == KVM_PROFILING)) {
+               kvm_x86_ops->cache_regs(vcpu);
+               profile_hit(KVM_PROFILING, (void *)vcpu->rip);
+       }
+
+       r = kvm_x86_ops->handle_exit(kvm_run, vcpu);
+
+       if (r > 0) {
+               if (dm_request_for_irq_injection(vcpu, kvm_run)) {
+                       r = -EINTR;
+                       kvm_run->exit_reason = KVM_EXIT_INTR;
+                       ++vcpu->stat.request_irq_exits;
+                       goto out;
+               }
+               if (!need_resched()) {
+                       ++vcpu->stat.light_exits;
+                       goto again;
+               }
+       }
+
+out:
+       if (r > 0) {
+               kvm_resched(vcpu);
+               goto preempted;
+       }
+
+       post_kvm_run_save(vcpu, kvm_run);
+
+       return r;
+}
+
+
 static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
 {
        int r;
@@ -2017,7 +2132,7 @@ static int kvm_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
                kvm_x86_ops->decache_regs(vcpu);
        }
 
-       r = kvm_x86_ops->run(vcpu, kvm_run);
+       r = __vcpu_run(vcpu, kvm_run);
 
 out:
        if (vcpu->sigset_active)
@@ -3336,7 +3451,7 @@ static int kvm_resume(struct sys_device *dev)
 }
 
 static struct sysdev_class kvm_sysdev_class = {
-       set_kset_name("kvm"),
+       .name = "kvm",
        .suspend = kvm_suspend,
        .resume = kvm_resume,
 };