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>
25 #include <linux/stackprotector.h>
26 #include <linux/pgtable.h>
28 #include <asm/paravirt.h>
29 #include <asm/idtentry.h>
32 #include <asm/io_apic.h>
34 #include <xen/interface/xen.h>
35 #include <xen/interface/vcpu.h>
36 #include <xen/interface/xenpmu.h>
38 #include <asm/spec-ctrl.h>
39 #include <asm/xen/interface.h>
40 #include <asm/xen/hypercall.h>
44 #include <xen/events.h>
46 #include <xen/hvc-console.h>
52 cpumask_var_t xen_cpu_initialized_map;
54 static DEFINE_PER_CPU(struct xen_common_irq, xen_irq_work) = { .irq = -1 };
55 static DEFINE_PER_CPU(struct xen_common_irq, xen_pmu_irq) = { .irq = -1 };
57 static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id);
58 void asm_cpu_bringup_and_idle(void);
60 static void cpu_bringup(void)
66 touch_softlockup_watchdog();
68 /* PVH runs in ring 0 and allows us to do native syscalls. Yay! */
69 if (!xen_feature(XENFEAT_supervisor_mode_kernel)) {
70 xen_enable_sysenter();
73 cpu = smp_processor_id();
74 smp_store_cpu_info(cpu);
75 cpu_data(cpu).x86_max_cores = 1;
76 set_cpu_sibling_map(cpu);
78 speculative_store_bypass_ht_init();
80 xen_setup_cpu_clockevents();
82 notify_cpu_starting(cpu);
84 set_cpu_online(cpu, true);
86 cpu_set_state_online(cpu); /* Implies full memory barrier. */
88 /* We can take interrupts now: we're officially "up". */
92 asmlinkage __visible void cpu_bringup_and_idle(void)
95 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
98 void xen_smp_intr_free_pv(unsigned int cpu)
100 if (per_cpu(xen_irq_work, cpu).irq >= 0) {
101 unbind_from_irqhandler(per_cpu(xen_irq_work, cpu).irq, NULL);
102 per_cpu(xen_irq_work, cpu).irq = -1;
103 kfree(per_cpu(xen_irq_work, cpu).name);
104 per_cpu(xen_irq_work, cpu).name = NULL;
107 if (per_cpu(xen_pmu_irq, cpu).irq >= 0) {
108 unbind_from_irqhandler(per_cpu(xen_pmu_irq, cpu).irq, NULL);
109 per_cpu(xen_pmu_irq, cpu).irq = -1;
110 kfree(per_cpu(xen_pmu_irq, cpu).name);
111 per_cpu(xen_pmu_irq, cpu).name = NULL;
115 int xen_smp_intr_init_pv(unsigned int cpu)
118 char *callfunc_name, *pmu_name;
120 callfunc_name = kasprintf(GFP_KERNEL, "irqwork%d", cpu);
121 rc = bind_ipi_to_irqhandler(XEN_IRQ_WORK_VECTOR,
123 xen_irq_work_interrupt,
124 IRQF_PERCPU|IRQF_NOBALANCING,
129 per_cpu(xen_irq_work, cpu).irq = rc;
130 per_cpu(xen_irq_work, cpu).name = callfunc_name;
133 pmu_name = kasprintf(GFP_KERNEL, "pmu%d", cpu);
134 rc = bind_virq_to_irqhandler(VIRQ_XENPMU, cpu,
136 IRQF_PERCPU|IRQF_NOBALANCING,
140 per_cpu(xen_pmu_irq, cpu).irq = rc;
141 per_cpu(xen_pmu_irq, cpu).name = pmu_name;
147 xen_smp_intr_free_pv(cpu);
151 static void __init _get_smp_config(unsigned int early)
154 unsigned int subtract = 0;
161 for (i = 0; i < nr_cpu_ids; i++) {
162 rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
165 set_cpu_possible(i, true);
167 set_cpu_possible(i, false);
168 set_cpu_present(i, false);
172 #ifdef CONFIG_HOTPLUG_CPU
173 /* This is akin to using 'nr_cpus' on the Linux command line.
174 * Which is OK as when we use 'dom0_max_vcpus=X' we can only
175 * have up to X, while nr_cpu_ids is greater than X. This
176 * normally is not a problem, except when CPU hotplugging
177 * is involved and then there might be more than X CPUs
178 * in the guest - which will not work as there is no
179 * hypercall to expand the max number of VCPUs an already
180 * running guest has. So cap it up to X. */
182 nr_cpu_ids = nr_cpu_ids - subtract;
187 static void __init xen_pv_smp_prepare_boot_cpu(void)
189 BUG_ON(smp_processor_id() != 0);
190 native_smp_prepare_boot_cpu();
192 if (!xen_feature(XENFEAT_writable_page_tables))
193 /* We've switched to the "real" per-cpu gdt, so make
194 * sure the old memory can be recycled. */
195 make_lowmem_page_readwrite(xen_initial_gdt);
197 xen_setup_vcpu_info_placement();
200 * The alternative logic (which patches the unlock/lock) runs before
201 * the smp bootup up code is activated. Hence we need to set this up
202 * the core kernel is being patched. Otherwise we will have only
203 * modules patched but not core code.
205 xen_init_spinlocks();
208 static void __init xen_pv_smp_prepare_cpus(unsigned int max_cpus)
213 if (skip_ioapic_setup) {
214 char *m = (max_cpus == 0) ?
215 "The nosmp parameter is incompatible with Xen; " \
216 "use Xen dom0_max_vcpus=1 parameter" :
217 "The noapic parameter is incompatible with Xen";
222 xen_init_lock_cpu(0);
224 smp_store_boot_cpu_info();
225 cpu_data(0).x86_max_cores = 1;
227 for_each_possible_cpu(i) {
228 zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
229 zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
230 zalloc_cpumask_var(&per_cpu(cpu_die_map, i), GFP_KERNEL);
231 zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
233 set_cpu_sibling_map(0);
235 speculative_store_bypass_ht_init();
239 if (xen_smp_intr_init(0) || xen_smp_intr_init_pv(0))
242 if (!alloc_cpumask_var(&xen_cpu_initialized_map, GFP_KERNEL))
243 panic("could not allocate xen_cpu_initialized_map\n");
245 cpumask_copy(xen_cpu_initialized_map, cpumask_of(0));
247 /* Restrict the possible_map according to max_cpus. */
248 while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus)) {
249 for (cpu = nr_cpu_ids - 1; !cpu_possible(cpu); cpu--)
251 set_cpu_possible(cpu, false);
254 for_each_possible_cpu(cpu)
255 set_cpu_present(cpu, true);
259 cpu_initialize_context(unsigned int cpu, struct task_struct *idle)
261 struct vcpu_guest_context *ctxt;
262 struct desc_struct *gdt;
263 unsigned long gdt_mfn;
265 /* used to tell cpu_init() that it can proceed with initialization */
266 cpumask_set_cpu(cpu, cpu_callout_mask);
267 if (cpumask_test_and_set_cpu(cpu, xen_cpu_initialized_map))
270 ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
274 gdt = get_cpu_gdt_rw(cpu);
277 * Bring up the CPU in cpu_bringup_and_idle() with the stack
278 * pointing just below where pt_regs would be if it were a normal
281 ctxt->user_regs.eip = (unsigned long)asm_cpu_bringup_and_idle;
282 ctxt->flags = VGCF_IN_KERNEL;
283 ctxt->user_regs.eflags = 0x1000; /* IOPL_RING1 */
284 ctxt->user_regs.ds = __USER_DS;
285 ctxt->user_regs.es = __USER_DS;
286 ctxt->user_regs.ss = __KERNEL_DS;
287 ctxt->user_regs.cs = __KERNEL_CS;
288 ctxt->user_regs.esp = (unsigned long)task_pt_regs(idle);
290 xen_copy_trap_info(ctxt->trap_ctxt);
292 BUG_ON((unsigned long)gdt & ~PAGE_MASK);
294 gdt_mfn = arbitrary_virt_to_mfn(gdt);
295 make_lowmem_page_readonly(gdt);
296 make_lowmem_page_readonly(mfn_to_virt(gdt_mfn));
298 ctxt->gdt_frames[0] = gdt_mfn;
299 ctxt->gdt_ents = GDT_ENTRIES;
302 * Set SS:SP that Xen will use when entering guest kernel mode
303 * from guest user mode. Subsequent calls to load_sp0() can
306 ctxt->kernel_ss = __KERNEL_DS;
307 ctxt->kernel_sp = task_top_of_stack(idle);
309 ctxt->gs_base_kernel = per_cpu_offset(cpu);
310 ctxt->event_callback_eip =
311 (unsigned long)xen_asm_exc_xen_hypervisor_callback;
312 ctxt->failsafe_callback_eip =
313 (unsigned long)xen_failsafe_callback;
314 per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir);
316 ctxt->ctrlreg[3] = xen_pfn_to_cr3(virt_to_gfn(swapper_pg_dir));
317 if (HYPERVISOR_vcpu_op(VCPUOP_initialise, xen_vcpu_nr(cpu), ctxt))
324 static int xen_pv_cpu_up(unsigned int cpu, struct task_struct *idle)
328 rc = common_cpu_up(cpu, idle);
332 xen_setup_runstate_info(cpu);
335 * PV VCPUs are always successfully taken down (see 'while' loop
336 * in xen_cpu_die()), so -EBUSY is an error.
338 rc = cpu_check_up_prepare(cpu);
342 /* make sure interrupts start blocked */
343 per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1;
345 rc = cpu_initialize_context(cpu, idle);
351 rc = HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL);
354 while (cpu_report_state(cpu) != CPU_ONLINE)
355 HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
360 #ifdef CONFIG_HOTPLUG_CPU
361 static int xen_pv_cpu_disable(void)
363 unsigned int cpu = smp_processor_id();
367 cpu_disable_common();
369 load_cr3(swapper_pg_dir);
373 static void xen_pv_cpu_die(unsigned int cpu)
375 while (HYPERVISOR_vcpu_op(VCPUOP_is_up,
376 xen_vcpu_nr(cpu), NULL)) {
377 __set_current_state(TASK_UNINTERRUPTIBLE);
378 schedule_timeout(HZ/10);
381 if (common_cpu_die(cpu) == 0) {
382 xen_smp_intr_free(cpu);
383 xen_uninit_lock_cpu(cpu);
384 xen_teardown_timer(cpu);
389 static void xen_pv_play_dead(void) /* used only with HOTPLUG_CPU */
392 HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(smp_processor_id()), NULL);
395 * commit 4b0c0f294 (tick: Cleanup NOHZ per cpu data on cpu down)
396 * clears certain data that the cpu_idle loop (which called us
397 * and that we return from) expects. The only way to get that
398 * data back is to call:
400 tick_nohz_idle_enter();
401 tick_nohz_idle_stop_tick_protected();
403 cpuhp_online_idle(CPUHP_AP_ONLINE_IDLE);
406 #else /* !CONFIG_HOTPLUG_CPU */
407 static int xen_pv_cpu_disable(void)
412 static void xen_pv_cpu_die(unsigned int cpu)
417 static void xen_pv_play_dead(void)
423 static void stop_self(void *v)
425 int cpu = smp_processor_id();
427 /* make sure we're not pinning something down */
428 load_cr3(swapper_pg_dir);
429 /* should set up a minimal gdt */
431 set_cpu_online(cpu, false);
433 HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(cpu), NULL);
437 static void xen_pv_stop_other_cpus(int wait)
439 smp_call_function(stop_self, NULL, wait);
442 static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id)
446 inc_irq_stat(apic_irq_work_irqs);
452 static const struct smp_ops xen_smp_ops __initconst = {
453 .smp_prepare_boot_cpu = xen_pv_smp_prepare_boot_cpu,
454 .smp_prepare_cpus = xen_pv_smp_prepare_cpus,
455 .smp_cpus_done = xen_smp_cpus_done,
457 .cpu_up = xen_pv_cpu_up,
458 .cpu_die = xen_pv_cpu_die,
459 .cpu_disable = xen_pv_cpu_disable,
460 .play_dead = xen_pv_play_dead,
462 .stop_other_cpus = xen_pv_stop_other_cpus,
463 .smp_send_reschedule = xen_smp_send_reschedule,
465 .send_call_func_ipi = xen_smp_send_call_function_ipi,
466 .send_call_func_single_ipi = xen_smp_send_call_function_single_ipi,
469 void __init xen_smp_init(void)
471 smp_ops = xen_smp_ops;
473 /* Avoid searching for BIOS MP tables */
474 x86_init.mpparse.find_smp_config = x86_init_noop;
475 x86_init.mpparse.get_smp_config = _get_smp_config;