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);
59 static void cpu_bringup(void)
64 cpuhp_ap_sync_alive();
67 touch_softlockup_watchdog();
69 /* PVH runs in ring 0 and allows us to do native syscalls. Yay! */
70 if (!xen_feature(XENFEAT_supervisor_mode_kernel)) {
71 xen_enable_sysenter();
74 cpu = smp_processor_id();
75 smp_store_cpu_info(cpu);
76 cpu_data(cpu).x86_max_cores = 1;
77 set_cpu_sibling_map(cpu);
79 speculative_store_bypass_ht_init();
81 xen_setup_cpu_clockevents();
83 notify_cpu_starting(cpu);
85 set_cpu_online(cpu, true);
89 /* We can take interrupts now: we're officially "up". */
93 asmlinkage __visible void cpu_bringup_and_idle(void)
96 cpu_startup_entry(CPUHP_AP_ONLINE_IDLE);
99 void xen_smp_intr_free_pv(unsigned int cpu)
101 kfree(per_cpu(xen_irq_work, cpu).name);
102 per_cpu(xen_irq_work, cpu).name = NULL;
103 if (per_cpu(xen_irq_work, cpu).irq >= 0) {
104 unbind_from_irqhandler(per_cpu(xen_irq_work, cpu).irq, NULL);
105 per_cpu(xen_irq_work, cpu).irq = -1;
108 kfree(per_cpu(xen_pmu_irq, cpu).name);
109 per_cpu(xen_pmu_irq, cpu).name = NULL;
110 if (per_cpu(xen_pmu_irq, cpu).irq >= 0) {
111 unbind_from_irqhandler(per_cpu(xen_pmu_irq, cpu).irq, NULL);
112 per_cpu(xen_pmu_irq, cpu).irq = -1;
116 int xen_smp_intr_init_pv(unsigned int cpu)
119 char *callfunc_name, *pmu_name;
121 callfunc_name = kasprintf(GFP_KERNEL, "irqwork%d", cpu);
122 per_cpu(xen_irq_work, cpu).name = callfunc_name;
123 rc = bind_ipi_to_irqhandler(XEN_IRQ_WORK_VECTOR,
125 xen_irq_work_interrupt,
126 IRQF_PERCPU|IRQF_NOBALANCING,
131 per_cpu(xen_irq_work, cpu).irq = rc;
134 pmu_name = kasprintf(GFP_KERNEL, "pmu%d", cpu);
135 per_cpu(xen_pmu_irq, cpu).name = pmu_name;
136 rc = bind_virq_to_irqhandler(VIRQ_XENPMU, cpu,
138 IRQF_PERCPU|IRQF_NOBALANCING,
142 per_cpu(xen_pmu_irq, cpu).irq = rc;
148 xen_smp_intr_free_pv(cpu);
152 static void __init _get_smp_config(unsigned int early)
155 unsigned int subtract = 0;
162 for (i = 0; i < nr_cpu_ids; i++) {
163 rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
166 set_cpu_possible(i, true);
168 set_cpu_possible(i, false);
169 set_cpu_present(i, false);
173 #ifdef CONFIG_HOTPLUG_CPU
174 /* This is akin to using 'nr_cpus' on the Linux command line.
175 * Which is OK as when we use 'dom0_max_vcpus=X' we can only
176 * have up to X, while nr_cpu_ids is greater than X. This
177 * normally is not a problem, except when CPU hotplugging
178 * is involved and then there might be more than X CPUs
179 * in the guest - which will not work as there is no
180 * hypercall to expand the max number of VCPUs an already
181 * running guest has. So cap it up to X. */
183 set_nr_cpu_ids(nr_cpu_ids - subtract);
185 /* Pretend to be a proper enumerated system */
186 smp_found_config = 1;
189 static void __init xen_pv_smp_prepare_boot_cpu(void)
191 BUG_ON(smp_processor_id() != 0);
192 native_smp_prepare_boot_cpu();
194 if (!xen_feature(XENFEAT_writable_page_tables))
195 /* We've switched to the "real" per-cpu gdt, so make
196 * sure the old memory can be recycled. */
197 make_lowmem_page_readwrite(xen_initial_gdt);
199 xen_setup_vcpu_info_placement();
202 * The alternative logic (which patches the unlock/lock) runs before
203 * the smp bootup up code is activated. Hence we need to set this up
204 * the core kernel is being patched. Otherwise we will have only
205 * modules patched but not core code.
207 xen_init_spinlocks();
210 static void __init xen_pv_smp_prepare_cpus(unsigned int max_cpus)
214 if (ioapic_is_disabled) {
215 char *m = (max_cpus == 0) ?
216 "The nosmp parameter is incompatible with Xen; " \
217 "use Xen dom0_max_vcpus=1 parameter" :
218 "The noapic parameter is incompatible with Xen";
223 xen_init_lock_cpu(0);
225 smp_prepare_cpus_common();
227 cpu_data(0).x86_max_cores = 1;
229 speculative_store_bypass_ht_init();
233 if (xen_smp_intr_init(0) || xen_smp_intr_init_pv(0))
236 if (!alloc_cpumask_var(&xen_cpu_initialized_map, GFP_KERNEL))
237 panic("could not allocate xen_cpu_initialized_map\n");
239 cpumask_copy(xen_cpu_initialized_map, cpumask_of(0));
241 /* Restrict the possible_map according to max_cpus. */
242 while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus)) {
243 for (cpu = nr_cpu_ids - 1; !cpu_possible(cpu); cpu--)
245 set_cpu_possible(cpu, false);
248 for_each_possible_cpu(cpu)
249 set_cpu_present(cpu, true);
253 cpu_initialize_context(unsigned int cpu, struct task_struct *idle)
255 struct vcpu_guest_context *ctxt;
256 struct desc_struct *gdt;
257 unsigned long gdt_mfn;
259 if (cpumask_test_and_set_cpu(cpu, xen_cpu_initialized_map))
262 ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
264 cpumask_clear_cpu(cpu, xen_cpu_initialized_map);
268 gdt = get_cpu_gdt_rw(cpu);
271 * Bring up the CPU in cpu_bringup_and_idle() with the stack
272 * pointing just below where pt_regs would be if it were a normal
275 ctxt->user_regs.eip = (unsigned long)asm_cpu_bringup_and_idle;
276 ctxt->flags = VGCF_IN_KERNEL;
277 ctxt->user_regs.eflags = 0x1000; /* IOPL_RING1 */
278 ctxt->user_regs.ds = __USER_DS;
279 ctxt->user_regs.es = __USER_DS;
280 ctxt->user_regs.ss = __KERNEL_DS;
281 ctxt->user_regs.cs = __KERNEL_CS;
282 ctxt->user_regs.esp = (unsigned long)task_pt_regs(idle);
284 xen_copy_trap_info(ctxt->trap_ctxt);
286 BUG_ON((unsigned long)gdt & ~PAGE_MASK);
288 gdt_mfn = arbitrary_virt_to_mfn(gdt);
289 make_lowmem_page_readonly(gdt);
290 make_lowmem_page_readonly(mfn_to_virt(gdt_mfn));
292 ctxt->gdt_frames[0] = gdt_mfn;
293 ctxt->gdt_ents = GDT_ENTRIES;
296 * Set SS:SP that Xen will use when entering guest kernel mode
297 * from guest user mode. Subsequent calls to load_sp0() can
300 ctxt->kernel_ss = __KERNEL_DS;
301 ctxt->kernel_sp = task_top_of_stack(idle);
303 ctxt->gs_base_kernel = per_cpu_offset(cpu);
304 ctxt->event_callback_eip =
305 (unsigned long)xen_asm_exc_xen_hypervisor_callback;
306 ctxt->failsafe_callback_eip =
307 (unsigned long)xen_failsafe_callback;
308 per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir);
310 ctxt->ctrlreg[3] = xen_pfn_to_cr3(virt_to_gfn(swapper_pg_dir));
311 if (HYPERVISOR_vcpu_op(VCPUOP_initialise, xen_vcpu_nr(cpu), ctxt))
318 static int xen_pv_kick_ap(unsigned int cpu, struct task_struct *idle)
322 rc = common_cpu_up(cpu, idle);
326 xen_setup_runstate_info(cpu);
328 /* make sure interrupts start blocked */
329 per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1;
331 rc = cpu_initialize_context(cpu, idle);
338 * Why is this a BUG? If the hypercall fails then everything can be
341 BUG_ON(HYPERVISOR_vcpu_op(VCPUOP_up, xen_vcpu_nr(cpu), NULL));
346 static void xen_pv_poll_sync_state(void)
348 HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
351 #ifdef CONFIG_HOTPLUG_CPU
352 static int xen_pv_cpu_disable(void)
354 unsigned int cpu = smp_processor_id();
358 cpu_disable_common();
360 load_cr3(swapper_pg_dir);
364 static void xen_pv_cpu_die(unsigned int cpu)
366 while (HYPERVISOR_vcpu_op(VCPUOP_is_up, xen_vcpu_nr(cpu), NULL)) {
367 __set_current_state(TASK_UNINTERRUPTIBLE);
368 schedule_timeout(HZ/10);
372 static void xen_pv_cleanup_dead_cpu(unsigned int cpu)
374 xen_smp_intr_free(cpu);
375 xen_uninit_lock_cpu(cpu);
376 xen_teardown_timer(cpu);
380 static void __noreturn xen_pv_play_dead(void) /* used only with HOTPLUG_CPU */
383 HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(smp_processor_id()), NULL);
384 xen_cpu_bringup_again((unsigned long)task_pt_regs(current));
388 #else /* !CONFIG_HOTPLUG_CPU */
389 static int xen_pv_cpu_disable(void)
394 static void xen_pv_cpu_die(unsigned int cpu)
399 static void xen_pv_cleanup_dead_cpu(unsigned int cpu)
404 static void __noreturn xen_pv_play_dead(void)
410 static void stop_self(void *v)
412 int cpu = smp_processor_id();
414 /* make sure we're not pinning something down */
415 load_cr3(swapper_pg_dir);
416 /* should set up a minimal gdt */
418 set_cpu_online(cpu, false);
420 HYPERVISOR_vcpu_op(VCPUOP_down, xen_vcpu_nr(cpu), NULL);
424 static void xen_pv_stop_other_cpus(int wait)
426 smp_call_function(stop_self, NULL, wait);
429 static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id)
432 inc_irq_stat(apic_irq_work_irqs);
437 static const struct smp_ops xen_smp_ops __initconst = {
438 .smp_prepare_boot_cpu = xen_pv_smp_prepare_boot_cpu,
439 .smp_prepare_cpus = xen_pv_smp_prepare_cpus,
440 .smp_cpus_done = xen_smp_cpus_done,
442 .kick_ap_alive = xen_pv_kick_ap,
443 .cpu_die = xen_pv_cpu_die,
444 .cleanup_dead_cpu = xen_pv_cleanup_dead_cpu,
445 .poll_sync_state = xen_pv_poll_sync_state,
446 .cpu_disable = xen_pv_cpu_disable,
447 .play_dead = xen_pv_play_dead,
449 .stop_other_cpus = xen_pv_stop_other_cpus,
450 .smp_send_reschedule = xen_smp_send_reschedule,
452 .send_call_func_ipi = xen_smp_send_call_function_ipi,
453 .send_call_func_single_ipi = xen_smp_send_call_function_single_ipi,
456 void __init xen_smp_init(void)
458 smp_ops = xen_smp_ops;
460 /* Avoid searching for BIOS MP tables */
461 x86_init.mpparse.find_smp_config = x86_init_noop;
462 x86_init.mpparse.get_smp_config = _get_smp_config;