x86/xen: split off smp_pv.c
[muen/linux.git] / arch / x86 / xen / smp.c
index 148e62cfc22ff9661b7c4ef2e8bff8fdef57ea34..82ac611f2fc123d2a4209ade33247ee510661d2e 100644 (file)
@@ -1,63 +1,21 @@
-/*
- * Xen SMP support
- *
- * This file implements the Xen versions of smp_ops.  SMP under Xen is
- * very straightforward.  Bringing a CPU up is simply a matter of
- * loading its initial context and setting it running.
- *
- * IPIs are handled through the Xen event mechanism.
- *
- * Because virtual CPUs can be scheduled onto any real CPU, there's no
- * useful topology information for the kernel to make use of.  As a
- * result, all CPUs are treated as if they're single-core and
- * single-threaded.
- */
-#include <linux/sched.h>
-#include <linux/err.h>
-#include <linux/slab.h>
 #include <linux/smp.h>
-#include <linux/irq_work.h>
-#include <linux/tick.h>
-#include <linux/nmi.h>
-
-#include <asm/paravirt.h>
-#include <asm/desc.h>
-#include <asm/pgtable.h>
-#include <asm/cpu.h>
-
-#include <xen/interface/xen.h>
-#include <xen/interface/vcpu.h>
-#include <xen/interface/xenpmu.h>
-
-#include <asm/xen/interface.h>
-#include <asm/xen/hypercall.h>
+#include <linux/slab.h>
+#include <linux/cpumask.h>
+#include <linux/percpu.h>
 
-#include <xen/xen.h>
-#include <xen/page.h>
 #include <xen/events.h>
 
 #include <xen/hvc-console.h>
 #include "xen-ops.h"
-#include "mmu.h"
 #include "smp.h"
-#include "pmu.h"
-
-cpumask_var_t xen_cpu_initialized_map;
 
-struct xen_common_irq {
-       int irq;
-       char *name;
-};
 static DEFINE_PER_CPU(struct xen_common_irq, xen_resched_irq) = { .irq = -1 };
 static DEFINE_PER_CPU(struct xen_common_irq, xen_callfunc_irq) = { .irq = -1 };
 static DEFINE_PER_CPU(struct xen_common_irq, xen_callfuncsingle_irq) = { .irq = -1 };
-static DEFINE_PER_CPU(struct xen_common_irq, xen_irq_work) = { .irq = -1 };
 static DEFINE_PER_CPU(struct xen_common_irq, xen_debug_irq) = { .irq = -1 };
-static DEFINE_PER_CPU(struct xen_common_irq, xen_pmu_irq) = { .irq = -1 };
 
 static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id);
 static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id);
-static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id);
 
 /*
  * Reschedule call back.
@@ -70,42 +28,6 @@ static irqreturn_t xen_reschedule_interrupt(int irq, void *dev_id)
        return IRQ_HANDLED;
 }
 
-static void cpu_bringup(void)
-{
-       int cpu;
-
-       cpu_init();
-       touch_softlockup_watchdog();
-       preempt_disable();
-
-       /* PVH runs in ring 0 and allows us to do native syscalls. Yay! */
-       if (!xen_feature(XENFEAT_supervisor_mode_kernel)) {
-               xen_enable_sysenter();
-               xen_enable_syscall();
-       }
-       cpu = smp_processor_id();
-       smp_store_cpu_info(cpu);
-       cpu_data(cpu).x86_max_cores = 1;
-       set_cpu_sibling_map(cpu);
-
-       xen_setup_cpu_clockevents();
-
-       notify_cpu_starting(cpu);
-
-       set_cpu_online(cpu, true);
-
-       cpu_set_state_online(cpu);  /* Implies full memory barrier. */
-
-       /* We can take interrupts now: we're officially "up". */
-       local_irq_enable();
-}
-
-asmlinkage __visible void cpu_bringup_and_idle(void)
-{
-       cpu_bringup();
-       cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
-}
-
 void xen_smp_intr_free(unsigned int cpu)
 {
        if (per_cpu(xen_resched_irq, cpu).irq >= 0) {
@@ -135,23 +57,6 @@ void xen_smp_intr_free(unsigned int cpu)
        }
 }
 
-void xen_smp_intr_free_pv(unsigned int cpu)
-{
-       if (per_cpu(xen_irq_work, cpu).irq >= 0) {
-               unbind_from_irqhandler(per_cpu(xen_irq_work, cpu).irq, NULL);
-               per_cpu(xen_irq_work, cpu).irq = -1;
-               kfree(per_cpu(xen_irq_work, cpu).name);
-               per_cpu(xen_irq_work, cpu).name = NULL;
-       }
-
-       if (per_cpu(xen_pmu_irq, cpu).irq >= 0) {
-               unbind_from_irqhandler(per_cpu(xen_pmu_irq, cpu).irq, NULL);
-               per_cpu(xen_pmu_irq, cpu).irq = -1;
-               kfree(per_cpu(xen_pmu_irq, cpu).name);
-               per_cpu(xen_pmu_irq, cpu).name = NULL;
-       }
-}
-
 int xen_smp_intr_init(unsigned int cpu)
 {
        int rc;
@@ -209,360 +114,6 @@ int xen_smp_intr_init(unsigned int cpu)
        return rc;
 }
 
-int xen_smp_intr_init_pv(unsigned int cpu)
-{
-       int rc;
-       char *callfunc_name, *pmu_name;
-
-       callfunc_name = kasprintf(GFP_KERNEL, "irqwork%d", cpu);
-       rc = bind_ipi_to_irqhandler(XEN_IRQ_WORK_VECTOR,
-                                   cpu,
-                                   xen_irq_work_interrupt,
-                                   IRQF_PERCPU|IRQF_NOBALANCING,
-                                   callfunc_name,
-                                   NULL);
-       if (rc < 0)
-               goto fail;
-       per_cpu(xen_irq_work, cpu).irq = rc;
-       per_cpu(xen_irq_work, cpu).name = callfunc_name;
-
-       if (is_xen_pmu(cpu)) {
-               pmu_name = kasprintf(GFP_KERNEL, "pmu%d", cpu);
-               rc = bind_virq_to_irqhandler(VIRQ_XENPMU, cpu,
-                                            xen_pmu_irq_handler,
-                                            IRQF_PERCPU|IRQF_NOBALANCING,
-                                            pmu_name, NULL);
-               if (rc < 0)
-                       goto fail;
-               per_cpu(xen_pmu_irq, cpu).irq = rc;
-               per_cpu(xen_pmu_irq, cpu).name = pmu_name;
-       }
-
-       return 0;
-
- fail:
-       xen_smp_intr_free_pv(cpu);
-       return rc;
-}
-
-static void __init xen_fill_possible_map(void)
-{
-       int i, rc;
-
-       if (xen_initial_domain())
-               return;
-
-       for (i = 0; i < nr_cpu_ids; i++) {
-               rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
-               if (rc >= 0) {
-                       num_processors++;
-                       set_cpu_possible(i, true);
-               }
-       }
-}
-
-static void __init xen_filter_cpu_maps(void)
-{
-       int i, rc;
-       unsigned int subtract = 0;
-
-       if (!xen_initial_domain())
-               return;
-
-       num_processors = 0;
-       disabled_cpus = 0;
-       for (i = 0; i < nr_cpu_ids; i++) {
-               rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
-               if (rc >= 0) {
-                       num_processors++;
-                       set_cpu_possible(i, true);
-               } else {
-                       set_cpu_possible(i, false);
-                       set_cpu_present(i, false);
-                       subtract++;
-               }
-       }
-#ifdef CONFIG_HOTPLUG_CPU
-       /* This is akin to using 'nr_cpus' on the Linux command line.
-        * Which is OK as when we use 'dom0_max_vcpus=X' we can only
-        * have up to X, while nr_cpu_ids is greater than X. This
-        * normally is not a problem, except when CPU hotplugging
-        * is involved and then there might be more than X CPUs
-        * in the guest - which will not work as there is no
-        * hypercall to expand the max number of VCPUs an already
-        * running guest has. So cap it up to X. */
-       if (subtract)
-               nr_cpu_ids = nr_cpu_ids - subtract;
-#endif
-
-}
-
-static void __init xen_pv_smp_prepare_boot_cpu(void)
-{
-       BUG_ON(smp_processor_id() != 0);
-       native_smp_prepare_boot_cpu();
-
-       if (!xen_feature(XENFEAT_writable_page_tables))
-               /* We've switched to the "real" per-cpu gdt, so make
-                * sure the old memory can be recycled. */
-               make_lowmem_page_readwrite(xen_initial_gdt);
-
-#ifdef CONFIG_X86_32
-       /*
-        * Xen starts us with XEN_FLAT_RING1_DS, but linux code
-        * expects __USER_DS
-        */
-       loadsegment(ds, __USER_DS);
-       loadsegment(es, __USER_DS);
-#endif
-
-       xen_filter_cpu_maps();
-       xen_setup_vcpu_info_placement();
-
-       /*
-        * The alternative logic (which patches the unlock/lock) runs before
-        * the smp bootup up code is activated. Hence we need to set this up
-        * the core kernel is being patched. Otherwise we will have only
-        * modules patched but not core code.
-        */
-       xen_init_spinlocks();
-}
-
-static void __init xen_smp_prepare_cpus(unsigned int max_cpus)
-{
-       unsigned cpu;
-       unsigned int i;
-
-       if (skip_ioapic_setup) {
-               char *m = (max_cpus == 0) ?
-                       "The nosmp parameter is incompatible with Xen; " \
-                       "use Xen dom0_max_vcpus=1 parameter" :
-                       "The noapic parameter is incompatible with Xen";
-
-               xen_raw_printk(m);
-               panic(m);
-       }
-       xen_init_lock_cpu(0);
-
-       smp_store_boot_cpu_info();
-       cpu_data(0).x86_max_cores = 1;
-
-       for_each_possible_cpu(i) {
-               zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
-               zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
-               zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
-       }
-       set_cpu_sibling_map(0);
-
-       xen_pmu_init(0);
-
-       if (xen_smp_intr_init(0))
-               BUG();
-
-       if (!alloc_cpumask_var(&xen_cpu_initialized_map, GFP_KERNEL))
-               panic("could not allocate xen_cpu_initialized_map\n");
-
-       cpumask_copy(xen_cpu_initialized_map, cpumask_of(0));
-
-       /* Restrict the possible_map according to max_cpus. */
-       while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus)) {
-               for (cpu = nr_cpu_ids - 1; !cpu_possible(cpu); cpu--)
-                       continue;
-               set_cpu_possible(cpu, false);
-       }
-
-       for_each_possible_cpu(cpu)
-               set_cpu_present(cpu, true);
-}
-
-static int
-cpu_initialize_context(unsigned int cpu, struct task_struct *idle)
-{
-       struct vcpu_guest_context *ctxt;
-       struct desc_struct *gdt;
-       unsigned long gdt_mfn;
-
-       /* used to tell cpu_init() that it can proceed with initialization */
-       cpumask_set_cpu(cpu, cpu_callout_mask);
-       if (cpumask_test_and_set_cpu(cpu, xen_cpu_initialized_map))
-               return 0;
-
-       ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
-       if (ctxt == NULL)
-               return -ENOMEM;
-
-       gdt = get_cpu_gdt_rw(cpu);
-
-#ifdef CONFIG_X86_32
-       ctxt->user_regs.fs = __KERNEL_PERCPU;
-       ctxt->user_regs.gs = __KERNEL_STACK_CANARY;
-#endif
-       memset(&ctxt->fpu_ctxt, 0, sizeof(ctxt->fpu_ctxt));
-
-       ctxt->user_regs.eip = (unsigned long)cpu_bringup_and_idle;
-       ctxt->flags = VGCF_IN_KERNEL;
-       ctxt->user_regs.eflags = 0x1000; /* IOPL_RING1 */
-       ctxt->user_regs.ds = __USER_DS;
-       ctxt->user_regs.es = __USER_DS;
-       ctxt->user_regs.ss = __KERNEL_DS;
-
-       xen_copy_trap_info(ctxt->trap_ctxt);
-
-       ctxt->ldt_ents = 0;
-
-       BUG_ON((unsigned long)gdt & ~PAGE_MASK);
-
-       gdt_mfn = arbitrary_virt_to_mfn(gdt);
-       make_lowmem_page_readonly(gdt);
-       make_lowmem_page_readonly(mfn_to_virt(gdt_mfn));
-
-       ctxt->gdt_frames[0] = gdt_mfn;
-       ctxt->gdt_ents      = GDT_ENTRIES;
-
-       ctxt->kernel_ss = __KERNEL_DS;
-       ctxt->kernel_sp = idle->thread.sp0;
-
-#ifdef CONFIG_X86_32
-       ctxt->event_callback_cs     = __KERNEL_CS;
-       ctxt->failsafe_callback_cs  = __KERNEL_CS;
-#else
-       ctxt->gs_base_kernel = per_cpu_offset(cpu);
-#endif
-       ctxt->event_callback_eip    =
-               (unsigned long)xen_hypervisor_callback;
-       ctxt->failsafe_callback_eip =
-               (unsigned long)xen_failsafe_callback;
-       ctxt->user_regs.cs = __KERNEL_CS;
-       per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir);
-
-       ctxt->user_regs.esp = idle->thread.sp0 - sizeof(struct pt_regs);
-       ctxt->ctrlreg[3] = xen_pfn_to_cr3(virt_to_gfn(swapper_pg_dir));
-       if (HYPERVISOR_vcpu_op(VCPUOP_initialise, xen_vcpu_nr(cpu), ctxt))
-               BUG();
-
-       kfree(ctxt);
-       return 0;
-}
-
-static int xen_cpu_up(unsigned int cpu, struct task_struct *idle)
-{
-       int rc;
-
-       common_cpu_up(cpu, idle);
-
-       xen_setup_runstate_info(cpu);
-
-       /*
-        * PV VCPUs are always successfully taken down (see 'while' loop
-        * in xen_cpu_die()), so -EBUSY is an error.
-        */
-       rc = cpu_check_up_prepare(cpu);
-       if (rc)
-               return rc;
-
-       /* make sure interrupts start blocked */
-       per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1;
-
-       rc = cpu_initialize_context(cpu, idle);
-       if (rc)
-               return rc;
-
-       xen_pmu_init(cpu);
-
-       rc = HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL);
-       BUG_ON(rc);
-
-       while (cpu_report_state(cpu) != CPU_ONLINE)
-               HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
-
-       return 0;
-}
-
-static void xen_smp_cpus_done(unsigned int max_cpus)
-{
-}
-
-#ifdef CONFIG_HOTPLUG_CPU
-static int xen_cpu_disable(void)
-{
-       unsigned int cpu = smp_processor_id();
-       if (cpu == 0)
-               return -EBUSY;
-
-       cpu_disable_common();
-
-       load_cr3(swapper_pg_dir);
-       return 0;
-}
-
-static void xen_pv_cpu_die(unsigned int cpu)
-{
-       while (HYPERVISOR_vcpu_op(VCPUOP_is_up,
-                                 xen_vcpu_nr(cpu), NULL)) {
-               __set_current_state(TASK_UNINTERRUPTIBLE);
-               schedule_timeout(HZ/10);
-       }
-
-       if (common_cpu_die(cpu) == 0) {
-               xen_smp_intr_free(cpu);
-               xen_uninit_lock_cpu(cpu);
-               xen_teardown_timer(cpu);
-               xen_pmu_finish(cpu);
-       }
-}
-
-static void xen_play_dead(void) /* used only with HOTPLUG_CPU */
-{
-       play_dead_common();
-       HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(smp_processor_id()), NULL);
-       cpu_bringup();
-       /*
-        * commit 4b0c0f294 (tick: Cleanup NOHZ per cpu data on cpu down)
-        * clears certain data that the cpu_idle loop (which called us
-        * and that we return from) expects. The only way to get that
-        * data back is to call:
-        */
-       tick_nohz_idle_enter();
-
-       cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
-}
-
-#else /* !CONFIG_HOTPLUG_CPU */
-static int xen_cpu_disable(void)
-{
-       return -ENOSYS;
-}
-
-static void xen_pv_cpu_die(unsigned int cpu)
-{
-       BUG();
-}
-
-static void xen_play_dead(void)
-{
-       BUG();
-}
-
-#endif
-static void stop_self(void *v)
-{
-       int cpu = smp_processor_id();
-
-       /* make sure we're not pinning something down */
-       load_cr3(swapper_pg_dir);
-       /* should set up a minimal gdt */
-
-       set_cpu_online(cpu, false);
-
-       HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(cpu), NULL);
-       BUG();
-}
-
-static void xen_stop_other_cpus(int wait)
-{
-       smp_call_function(stop_self, NULL, wait);
-}
-
 void xen_smp_send_reschedule(int cpu)
 {
        xen_send_IPI_one(cpu, XEN_RESCHEDULE_VECTOR);
@@ -697,36 +248,3 @@ static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id)
 
        return IRQ_HANDLED;
 }
-
-static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id)
-{
-       irq_enter();
-       irq_work_run();
-       inc_irq_stat(apic_irq_work_irqs);
-       irq_exit();
-
-       return IRQ_HANDLED;
-}
-
-static const struct smp_ops xen_smp_ops __initconst = {
-       .smp_prepare_boot_cpu = xen_pv_smp_prepare_boot_cpu,
-       .smp_prepare_cpus = xen_smp_prepare_cpus,
-       .smp_cpus_done = xen_smp_cpus_done,
-
-       .cpu_up = xen_cpu_up,
-       .cpu_die = xen_pv_cpu_die,
-       .cpu_disable = xen_cpu_disable,
-       .play_dead = xen_play_dead,
-
-       .stop_other_cpus = xen_stop_other_cpus,
-       .smp_send_reschedule = xen_smp_send_reschedule,
-
-       .send_call_func_ipi = xen_smp_send_call_function_ipi,
-       .send_call_func_single_ipi = xen_smp_send_call_function_single_ipi,
-};
-
-void __init xen_smp_init(void)
-{
-       smp_ops = xen_smp_ops;
-       xen_fill_possible_map();
-}