1 // SPDX-License-Identifier: GPL-2.0
5 * This file implements the Xen versions of smp_ops. SMP under Xen is
6 * very straightforward. Bringing a CPU up is simply a matter of
7 * loading its initial context and setting it running.
9 * IPIs are handled through the Xen event mechanism.
11 * Because virtual CPUs can be scheduled onto any real CPU, there's no
12 * useful topology information for the kernel to make use of. As a
13 * result, all CPUs are treated as if they're single-core and
16 #include <linux/sched.h>
17 #include <linux/sched/task_stack.h>
18 #include <linux/err.h>
19 #include <linux/slab.h>
20 #include <linux/smp.h>
21 #include <linux/irq_work.h>
22 #include <linux/tick.h>
23 #include <linux/nmi.h>
24 #include <linux/cpuhotplug.h>
26 #include <asm/paravirt.h>
28 #include <asm/pgtable.h>
31 #include <xen/interface/xen.h>
32 #include <xen/interface/vcpu.h>
33 #include <xen/interface/xenpmu.h>
35 #include <asm/spec-ctrl.h>
36 #include <asm/xen/interface.h>
37 #include <asm/xen/hypercall.h>
41 #include <xen/events.h>
43 #include <xen/hvc-console.h>
49 cpumask_var_t xen_cpu_initialized_map;
51 static DEFINE_PER_CPU(struct xen_common_irq, xen_irq_work) = { .irq = -1 };
52 static DEFINE_PER_CPU(struct xen_common_irq, xen_pmu_irq) = { .irq = -1 };
54 static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id);
56 static void cpu_bringup(void)
61 touch_softlockup_watchdog();
64 /* PVH runs in ring 0 and allows us to do native syscalls. Yay! */
65 if (!xen_feature(XENFEAT_supervisor_mode_kernel)) {
66 xen_enable_sysenter();
69 cpu = smp_processor_id();
70 smp_store_cpu_info(cpu);
71 cpu_data(cpu).x86_max_cores = 1;
72 set_cpu_sibling_map(cpu);
74 speculative_store_bypass_ht_init();
76 xen_setup_cpu_clockevents();
78 notify_cpu_starting(cpu);
80 set_cpu_online(cpu, true);
82 cpu_set_state_online(cpu); /* Implies full memory barrier. */
84 /* We can take interrupts now: we're officially "up". */
88 asmlinkage __visible void cpu_bringup_and_idle(void)
91 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
94 void xen_smp_intr_free_pv(unsigned int cpu)
96 if (per_cpu(xen_irq_work, cpu).irq >= 0) {
97 unbind_from_irqhandler(per_cpu(xen_irq_work, cpu).irq, NULL);
98 per_cpu(xen_irq_work, cpu).irq = -1;
99 kfree(per_cpu(xen_irq_work, cpu).name);
100 per_cpu(xen_irq_work, cpu).name = NULL;
103 if (per_cpu(xen_pmu_irq, cpu).irq >= 0) {
104 unbind_from_irqhandler(per_cpu(xen_pmu_irq, cpu).irq, NULL);
105 per_cpu(xen_pmu_irq, cpu).irq = -1;
106 kfree(per_cpu(xen_pmu_irq, cpu).name);
107 per_cpu(xen_pmu_irq, cpu).name = NULL;
111 int xen_smp_intr_init_pv(unsigned int cpu)
114 char *callfunc_name, *pmu_name;
116 callfunc_name = kasprintf(GFP_KERNEL, "irqwork%d", cpu);
117 rc = bind_ipi_to_irqhandler(XEN_IRQ_WORK_VECTOR,
119 xen_irq_work_interrupt,
120 IRQF_PERCPU|IRQF_NOBALANCING,
125 per_cpu(xen_irq_work, cpu).irq = rc;
126 per_cpu(xen_irq_work, cpu).name = callfunc_name;
128 if (is_xen_pmu(cpu)) {
129 pmu_name = kasprintf(GFP_KERNEL, "pmu%d", cpu);
130 rc = bind_virq_to_irqhandler(VIRQ_XENPMU, cpu,
132 IRQF_PERCPU|IRQF_NOBALANCING,
136 per_cpu(xen_pmu_irq, cpu).irq = rc;
137 per_cpu(xen_pmu_irq, cpu).name = pmu_name;
143 xen_smp_intr_free_pv(cpu);
147 static void __init xen_fill_possible_map(void)
151 if (xen_initial_domain())
154 for (i = 0; i < nr_cpu_ids; i++) {
155 rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
158 set_cpu_possible(i, true);
163 static void __init xen_filter_cpu_maps(void)
166 unsigned int subtract = 0;
168 if (!xen_initial_domain())
173 for (i = 0; i < nr_cpu_ids; i++) {
174 rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
177 set_cpu_possible(i, true);
179 set_cpu_possible(i, false);
180 set_cpu_present(i, false);
184 #ifdef CONFIG_HOTPLUG_CPU
185 /* This is akin to using 'nr_cpus' on the Linux command line.
186 * Which is OK as when we use 'dom0_max_vcpus=X' we can only
187 * have up to X, while nr_cpu_ids is greater than X. This
188 * normally is not a problem, except when CPU hotplugging
189 * is involved and then there might be more than X CPUs
190 * in the guest - which will not work as there is no
191 * hypercall to expand the max number of VCPUs an already
192 * running guest has. So cap it up to X. */
194 nr_cpu_ids = nr_cpu_ids - subtract;
199 static void __init xen_pv_smp_prepare_boot_cpu(void)
201 BUG_ON(smp_processor_id() != 0);
202 native_smp_prepare_boot_cpu();
204 if (!xen_feature(XENFEAT_writable_page_tables))
205 /* We've switched to the "real" per-cpu gdt, so make
206 * sure the old memory can be recycled. */
207 make_lowmem_page_readwrite(xen_initial_gdt);
211 * Xen starts us with XEN_FLAT_RING1_DS, but linux code
214 loadsegment(ds, __USER_DS);
215 loadsegment(es, __USER_DS);
218 xen_filter_cpu_maps();
219 xen_setup_vcpu_info_placement();
222 * The alternative logic (which patches the unlock/lock) runs before
223 * the smp bootup up code is activated. Hence we need to set this up
224 * the core kernel is being patched. Otherwise we will have only
225 * modules patched but not core code.
227 xen_init_spinlocks();
230 static void __init xen_pv_smp_prepare_cpus(unsigned int max_cpus)
235 if (skip_ioapic_setup) {
236 char *m = (max_cpus == 0) ?
237 "The nosmp parameter is incompatible with Xen; " \
238 "use Xen dom0_max_vcpus=1 parameter" :
239 "The noapic parameter is incompatible with Xen";
244 xen_init_lock_cpu(0);
246 smp_store_boot_cpu_info();
247 cpu_data(0).x86_max_cores = 1;
249 for_each_possible_cpu(i) {
250 zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
251 zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
252 zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
254 set_cpu_sibling_map(0);
256 speculative_store_bypass_ht_init();
260 if (xen_smp_intr_init(0) || xen_smp_intr_init_pv(0))
263 if (!alloc_cpumask_var(&xen_cpu_initialized_map, GFP_KERNEL))
264 panic("could not allocate xen_cpu_initialized_map\n");
266 cpumask_copy(xen_cpu_initialized_map, cpumask_of(0));
268 /* Restrict the possible_map according to max_cpus. */
269 while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus)) {
270 for (cpu = nr_cpu_ids - 1; !cpu_possible(cpu); cpu--)
272 set_cpu_possible(cpu, false);
275 for_each_possible_cpu(cpu)
276 set_cpu_present(cpu, true);
280 cpu_initialize_context(unsigned int cpu, struct task_struct *idle)
282 struct vcpu_guest_context *ctxt;
283 struct desc_struct *gdt;
284 unsigned long gdt_mfn;
286 /* used to tell cpu_init() that it can proceed with initialization */
287 cpumask_set_cpu(cpu, cpu_callout_mask);
288 if (cpumask_test_and_set_cpu(cpu, xen_cpu_initialized_map))
291 ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
295 gdt = get_cpu_gdt_rw(cpu);
298 ctxt->user_regs.fs = __KERNEL_PERCPU;
299 ctxt->user_regs.gs = __KERNEL_STACK_CANARY;
301 memset(&ctxt->fpu_ctxt, 0, sizeof(ctxt->fpu_ctxt));
304 * Bring up the CPU in cpu_bringup_and_idle() with the stack
305 * pointing just below where pt_regs would be if it were a normal
308 ctxt->user_regs.eip = (unsigned long)cpu_bringup_and_idle;
309 ctxt->flags = VGCF_IN_KERNEL;
310 ctxt->user_regs.eflags = 0x1000; /* IOPL_RING1 */
311 ctxt->user_regs.ds = __USER_DS;
312 ctxt->user_regs.es = __USER_DS;
313 ctxt->user_regs.ss = __KERNEL_DS;
314 ctxt->user_regs.cs = __KERNEL_CS;
315 ctxt->user_regs.esp = (unsigned long)task_pt_regs(idle);
317 xen_copy_trap_info(ctxt->trap_ctxt);
321 BUG_ON((unsigned long)gdt & ~PAGE_MASK);
323 gdt_mfn = arbitrary_virt_to_mfn(gdt);
324 make_lowmem_page_readonly(gdt);
325 make_lowmem_page_readonly(mfn_to_virt(gdt_mfn));
327 ctxt->gdt_frames[0] = gdt_mfn;
328 ctxt->gdt_ents = GDT_ENTRIES;
331 * Set SS:SP that Xen will use when entering guest kernel mode
332 * from guest user mode. Subsequent calls to load_sp0() can
335 ctxt->kernel_ss = __KERNEL_DS;
336 ctxt->kernel_sp = task_top_of_stack(idle);
339 ctxt->event_callback_cs = __KERNEL_CS;
340 ctxt->failsafe_callback_cs = __KERNEL_CS;
342 ctxt->gs_base_kernel = per_cpu_offset(cpu);
344 ctxt->event_callback_eip =
345 (unsigned long)xen_hypervisor_callback;
346 ctxt->failsafe_callback_eip =
347 (unsigned long)xen_failsafe_callback;
348 per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir);
350 ctxt->ctrlreg[3] = xen_pfn_to_cr3(virt_to_gfn(swapper_pg_dir));
351 if (HYPERVISOR_vcpu_op(VCPUOP_initialise, xen_vcpu_nr(cpu), ctxt))
358 static int xen_pv_cpu_up(unsigned int cpu, struct task_struct *idle)
362 common_cpu_up(cpu, idle);
364 xen_setup_runstate_info(cpu);
367 * PV VCPUs are always successfully taken down (see 'while' loop
368 * in xen_cpu_die()), so -EBUSY is an error.
370 rc = cpu_check_up_prepare(cpu);
374 /* make sure interrupts start blocked */
375 per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1;
377 rc = cpu_initialize_context(cpu, idle);
383 rc = HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL);
386 while (cpu_report_state(cpu) != CPU_ONLINE)
387 HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
392 #ifdef CONFIG_HOTPLUG_CPU
393 static int xen_pv_cpu_disable(void)
395 unsigned int cpu = smp_processor_id();
399 cpu_disable_common();
401 load_cr3(swapper_pg_dir);
405 static void xen_pv_cpu_die(unsigned int cpu)
407 while (HYPERVISOR_vcpu_op(VCPUOP_is_up,
408 xen_vcpu_nr(cpu), NULL)) {
409 __set_current_state(TASK_UNINTERRUPTIBLE);
410 schedule_timeout(HZ/10);
413 if (common_cpu_die(cpu) == 0) {
414 xen_smp_intr_free(cpu);
415 xen_uninit_lock_cpu(cpu);
416 xen_teardown_timer(cpu);
421 static void xen_pv_play_dead(void) /* used only with HOTPLUG_CPU */
424 HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(smp_processor_id()), NULL);
427 * commit 4b0c0f294 (tick: Cleanup NOHZ per cpu data on cpu down)
428 * clears certain data that the cpu_idle loop (which called us
429 * and that we return from) expects. The only way to get that
430 * data back is to call:
432 tick_nohz_idle_enter();
433 tick_nohz_idle_stop_tick_protected();
435 cpuhp_online_idle(CPUHP_AP_ONLINE_IDLE);
438 #else /* !CONFIG_HOTPLUG_CPU */
439 static int xen_pv_cpu_disable(void)
444 static void xen_pv_cpu_die(unsigned int cpu)
449 static void xen_pv_play_dead(void)
455 static void stop_self(void *v)
457 int cpu = smp_processor_id();
459 /* make sure we're not pinning something down */
460 load_cr3(swapper_pg_dir);
461 /* should set up a minimal gdt */
463 set_cpu_online(cpu, false);
465 HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(cpu), NULL);
469 static void xen_pv_stop_other_cpus(int wait)
471 smp_call_function(stop_self, NULL, wait);
474 static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id)
478 inc_irq_stat(apic_irq_work_irqs);
484 static const struct smp_ops xen_smp_ops __initconst = {
485 .smp_prepare_boot_cpu = xen_pv_smp_prepare_boot_cpu,
486 .smp_prepare_cpus = xen_pv_smp_prepare_cpus,
487 .smp_cpus_done = xen_smp_cpus_done,
489 .cpu_up = xen_pv_cpu_up,
490 .cpu_die = xen_pv_cpu_die,
491 .cpu_disable = xen_pv_cpu_disable,
492 .play_dead = xen_pv_play_dead,
494 .stop_other_cpus = xen_pv_stop_other_cpus,
495 .smp_send_reschedule = xen_smp_send_reschedule,
497 .send_call_func_ipi = xen_smp_send_call_function_ipi,
498 .send_call_func_single_ipi = xen_smp_send_call_function_single_ipi,
501 void __init xen_smp_init(void)
503 smp_ops = xen_smp_ops;
504 xen_fill_possible_map();