Loading...
1// SPDX-License-Identifier: GPL-2.0
2#include <linux/smp.h>
3#include <linux/cpu.h>
4#include <linux/slab.h>
5#include <linux/cpumask.h>
6#include <linux/percpu.h>
7
8#include <xen/events.h>
9
10#include <xen/hvc-console.h>
11#include "xen-ops.h"
12#include "smp.h"
13
14static DEFINE_PER_CPU(struct xen_common_irq, xen_resched_irq) = { .irq = -1 };
15static DEFINE_PER_CPU(struct xen_common_irq, xen_callfunc_irq) = { .irq = -1 };
16static DEFINE_PER_CPU(struct xen_common_irq, xen_callfuncsingle_irq) = { .irq = -1 };
17static DEFINE_PER_CPU(struct xen_common_irq, xen_debug_irq) = { .irq = -1 };
18
19static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id);
20static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id);
21
22/*
23 * Reschedule call back.
24 */
25static irqreturn_t xen_reschedule_interrupt(int irq, void *dev_id)
26{
27 inc_irq_stat(irq_resched_count);
28 scheduler_ipi();
29
30 return IRQ_HANDLED;
31}
32
33void xen_smp_intr_free(unsigned int cpu)
34{
35 if (per_cpu(xen_resched_irq, cpu).irq >= 0) {
36 unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu).irq, NULL);
37 per_cpu(xen_resched_irq, cpu).irq = -1;
38 kfree(per_cpu(xen_resched_irq, cpu).name);
39 per_cpu(xen_resched_irq, cpu).name = NULL;
40 }
41 if (per_cpu(xen_callfunc_irq, cpu).irq >= 0) {
42 unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu).irq, NULL);
43 per_cpu(xen_callfunc_irq, cpu).irq = -1;
44 kfree(per_cpu(xen_callfunc_irq, cpu).name);
45 per_cpu(xen_callfunc_irq, cpu).name = NULL;
46 }
47 if (per_cpu(xen_debug_irq, cpu).irq >= 0) {
48 unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu).irq, NULL);
49 per_cpu(xen_debug_irq, cpu).irq = -1;
50 kfree(per_cpu(xen_debug_irq, cpu).name);
51 per_cpu(xen_debug_irq, cpu).name = NULL;
52 }
53 if (per_cpu(xen_callfuncsingle_irq, cpu).irq >= 0) {
54 unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu).irq,
55 NULL);
56 per_cpu(xen_callfuncsingle_irq, cpu).irq = -1;
57 kfree(per_cpu(xen_callfuncsingle_irq, cpu).name);
58 per_cpu(xen_callfuncsingle_irq, cpu).name = NULL;
59 }
60}
61
62int xen_smp_intr_init(unsigned int cpu)
63{
64 int rc;
65 char *resched_name, *callfunc_name, *debug_name;
66
67 resched_name = kasprintf(GFP_KERNEL, "resched%d", cpu);
68 rc = bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR,
69 cpu,
70 xen_reschedule_interrupt,
71 IRQF_PERCPU|IRQF_NOBALANCING,
72 resched_name,
73 NULL);
74 if (rc < 0)
75 goto fail;
76 per_cpu(xen_resched_irq, cpu).irq = rc;
77 per_cpu(xen_resched_irq, cpu).name = resched_name;
78
79 callfunc_name = kasprintf(GFP_KERNEL, "callfunc%d", cpu);
80 rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_VECTOR,
81 cpu,
82 xen_call_function_interrupt,
83 IRQF_PERCPU|IRQF_NOBALANCING,
84 callfunc_name,
85 NULL);
86 if (rc < 0)
87 goto fail;
88 per_cpu(xen_callfunc_irq, cpu).irq = rc;
89 per_cpu(xen_callfunc_irq, cpu).name = callfunc_name;
90
91 debug_name = kasprintf(GFP_KERNEL, "debug%d", cpu);
92 rc = bind_virq_to_irqhandler(VIRQ_DEBUG, cpu, xen_debug_interrupt,
93 IRQF_PERCPU | IRQF_NOBALANCING,
94 debug_name, NULL);
95 if (rc < 0)
96 goto fail;
97 per_cpu(xen_debug_irq, cpu).irq = rc;
98 per_cpu(xen_debug_irq, cpu).name = debug_name;
99
100 callfunc_name = kasprintf(GFP_KERNEL, "callfuncsingle%d", cpu);
101 rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_SINGLE_VECTOR,
102 cpu,
103 xen_call_function_single_interrupt,
104 IRQF_PERCPU|IRQF_NOBALANCING,
105 callfunc_name,
106 NULL);
107 if (rc < 0)
108 goto fail;
109 per_cpu(xen_callfuncsingle_irq, cpu).irq = rc;
110 per_cpu(xen_callfuncsingle_irq, cpu).name = callfunc_name;
111
112 return 0;
113
114 fail:
115 xen_smp_intr_free(cpu);
116 return rc;
117}
118
119void __init xen_smp_cpus_done(unsigned int max_cpus)
120{
121 int cpu, rc, count = 0;
122
123 if (xen_hvm_domain())
124 native_smp_cpus_done(max_cpus);
125 else
126 calculate_max_logical_packages();
127
128 if (xen_have_vcpu_info_placement)
129 return;
130
131 for_each_online_cpu(cpu) {
132 if (xen_vcpu_nr(cpu) < MAX_VIRT_CPUS)
133 continue;
134
135 rc = remove_cpu(cpu);
136
137 if (rc == 0) {
138 /*
139 * Reset vcpu_info so this cpu cannot be onlined again.
140 */
141 xen_vcpu_info_reset(cpu);
142 count++;
143 } else {
144 pr_warn("%s: failed to bring CPU %d down, error %d\n",
145 __func__, cpu, rc);
146 }
147 }
148 WARN(count, "%s: brought %d CPUs offline\n", __func__, count);
149}
150
151void xen_smp_send_reschedule(int cpu)
152{
153 xen_send_IPI_one(cpu, XEN_RESCHEDULE_VECTOR);
154}
155
156static void __xen_send_IPI_mask(const struct cpumask *mask,
157 int vector)
158{
159 unsigned cpu;
160
161 for_each_cpu_and(cpu, mask, cpu_online_mask)
162 xen_send_IPI_one(cpu, vector);
163}
164
165void xen_smp_send_call_function_ipi(const struct cpumask *mask)
166{
167 int cpu;
168
169 __xen_send_IPI_mask(mask, XEN_CALL_FUNCTION_VECTOR);
170
171 /* Make sure other vcpus get a chance to run if they need to. */
172 for_each_cpu(cpu, mask) {
173 if (xen_vcpu_stolen(cpu)) {
174 HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
175 break;
176 }
177 }
178}
179
180void xen_smp_send_call_function_single_ipi(int cpu)
181{
182 __xen_send_IPI_mask(cpumask_of(cpu),
183 XEN_CALL_FUNCTION_SINGLE_VECTOR);
184}
185
186static inline int xen_map_vector(int vector)
187{
188 int xen_vector;
189
190 switch (vector) {
191 case RESCHEDULE_VECTOR:
192 xen_vector = XEN_RESCHEDULE_VECTOR;
193 break;
194 case CALL_FUNCTION_VECTOR:
195 xen_vector = XEN_CALL_FUNCTION_VECTOR;
196 break;
197 case CALL_FUNCTION_SINGLE_VECTOR:
198 xen_vector = XEN_CALL_FUNCTION_SINGLE_VECTOR;
199 break;
200 case IRQ_WORK_VECTOR:
201 xen_vector = XEN_IRQ_WORK_VECTOR;
202 break;
203#ifdef CONFIG_X86_64
204 case NMI_VECTOR:
205 case APIC_DM_NMI: /* Some use that instead of NMI_VECTOR */
206 xen_vector = XEN_NMI_VECTOR;
207 break;
208#endif
209 default:
210 xen_vector = -1;
211 printk(KERN_ERR "xen: vector 0x%x is not implemented\n",
212 vector);
213 }
214
215 return xen_vector;
216}
217
218void xen_send_IPI_mask(const struct cpumask *mask,
219 int vector)
220{
221 int xen_vector = xen_map_vector(vector);
222
223 if (xen_vector >= 0)
224 __xen_send_IPI_mask(mask, xen_vector);
225}
226
227void xen_send_IPI_all(int vector)
228{
229 int xen_vector = xen_map_vector(vector);
230
231 if (xen_vector >= 0)
232 __xen_send_IPI_mask(cpu_online_mask, xen_vector);
233}
234
235void xen_send_IPI_self(int vector)
236{
237 int xen_vector = xen_map_vector(vector);
238
239 if (xen_vector >= 0)
240 xen_send_IPI_one(smp_processor_id(), xen_vector);
241}
242
243void xen_send_IPI_mask_allbutself(const struct cpumask *mask,
244 int vector)
245{
246 unsigned cpu;
247 unsigned int this_cpu = smp_processor_id();
248 int xen_vector = xen_map_vector(vector);
249
250 if (!(num_online_cpus() > 1) || (xen_vector < 0))
251 return;
252
253 for_each_cpu_and(cpu, mask, cpu_online_mask) {
254 if (this_cpu == cpu)
255 continue;
256
257 xen_send_IPI_one(cpu, xen_vector);
258 }
259}
260
261void xen_send_IPI_allbutself(int vector)
262{
263 xen_send_IPI_mask_allbutself(cpu_online_mask, vector);
264}
265
266static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id)
267{
268 irq_enter();
269 generic_smp_call_function_interrupt();
270 inc_irq_stat(irq_call_count);
271 irq_exit();
272
273 return IRQ_HANDLED;
274}
275
276static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id)
277{
278 irq_enter();
279 generic_smp_call_function_single_interrupt();
280 inc_irq_stat(irq_call_count);
281 irq_exit();
282
283 return IRQ_HANDLED;
284}
1/*
2 * Xen SMP support
3 *
4 * This file implements the Xen versions of smp_ops. SMP under Xen is
5 * very straightforward. Bringing a CPU up is simply a matter of
6 * loading its initial context and setting it running.
7 *
8 * IPIs are handled through the Xen event mechanism.
9 *
10 * Because virtual CPUs can be scheduled onto any real CPU, there's no
11 * useful topology information for the kernel to make use of. As a
12 * result, all CPUs are treated as if they're single-core and
13 * single-threaded.
14 */
15#include <linux/sched.h>
16#include <linux/err.h>
17#include <linux/slab.h>
18#include <linux/smp.h>
19#include <linux/irq_work.h>
20#include <linux/tick.h>
21
22#include <asm/paravirt.h>
23#include <asm/desc.h>
24#include <asm/pgtable.h>
25#include <asm/cpu.h>
26
27#include <xen/interface/xen.h>
28#include <xen/interface/vcpu.h>
29
30#include <asm/xen/interface.h>
31#include <asm/xen/hypercall.h>
32
33#include <xen/xen.h>
34#include <xen/page.h>
35#include <xen/events.h>
36
37#include <xen/hvc-console.h>
38#include "xen-ops.h"
39#include "mmu.h"
40
41cpumask_var_t xen_cpu_initialized_map;
42
43struct xen_common_irq {
44 int irq;
45 char *name;
46};
47static DEFINE_PER_CPU(struct xen_common_irq, xen_resched_irq) = { .irq = -1 };
48static DEFINE_PER_CPU(struct xen_common_irq, xen_callfunc_irq) = { .irq = -1 };
49static DEFINE_PER_CPU(struct xen_common_irq, xen_callfuncsingle_irq) = { .irq = -1 };
50static DEFINE_PER_CPU(struct xen_common_irq, xen_irq_work) = { .irq = -1 };
51static DEFINE_PER_CPU(struct xen_common_irq, xen_debug_irq) = { .irq = -1 };
52
53static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id);
54static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id);
55static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id);
56
57/*
58 * Reschedule call back.
59 */
60static irqreturn_t xen_reschedule_interrupt(int irq, void *dev_id)
61{
62 inc_irq_stat(irq_resched_count);
63 scheduler_ipi();
64
65 return IRQ_HANDLED;
66}
67
68static void cpu_bringup(void)
69{
70 int cpu;
71
72 cpu_init();
73 touch_softlockup_watchdog();
74 preempt_disable();
75
76 /* PVH runs in ring 0 and allows us to do native syscalls. Yay! */
77 if (!xen_feature(XENFEAT_supervisor_mode_kernel)) {
78 xen_enable_sysenter();
79 xen_enable_syscall();
80 }
81 cpu = smp_processor_id();
82 smp_store_cpu_info(cpu);
83 cpu_data(cpu).x86_max_cores = 1;
84 set_cpu_sibling_map(cpu);
85
86 xen_setup_cpu_clockevents();
87
88 notify_cpu_starting(cpu);
89
90 set_cpu_online(cpu, true);
91
92 this_cpu_write(cpu_state, CPU_ONLINE);
93
94 wmb();
95
96 /* We can take interrupts now: we're officially "up". */
97 local_irq_enable();
98
99 wmb(); /* make sure everything is out */
100}
101
102/* Note: cpu parameter is only relevant for PVH */
103static void cpu_bringup_and_idle(int cpu)
104{
105#ifdef CONFIG_X86_64
106 if (xen_feature(XENFEAT_auto_translated_physmap) &&
107 xen_feature(XENFEAT_supervisor_mode_kernel))
108 xen_pvh_secondary_vcpu_init(cpu);
109#endif
110 cpu_bringup();
111 cpu_startup_entry(CPUHP_ONLINE);
112}
113
114static void xen_smp_intr_free(unsigned int cpu)
115{
116 if (per_cpu(xen_resched_irq, cpu).irq >= 0) {
117 unbind_from_irqhandler(per_cpu(xen_resched_irq, cpu).irq, NULL);
118 per_cpu(xen_resched_irq, cpu).irq = -1;
119 kfree(per_cpu(xen_resched_irq, cpu).name);
120 per_cpu(xen_resched_irq, cpu).name = NULL;
121 }
122 if (per_cpu(xen_callfunc_irq, cpu).irq >= 0) {
123 unbind_from_irqhandler(per_cpu(xen_callfunc_irq, cpu).irq, NULL);
124 per_cpu(xen_callfunc_irq, cpu).irq = -1;
125 kfree(per_cpu(xen_callfunc_irq, cpu).name);
126 per_cpu(xen_callfunc_irq, cpu).name = NULL;
127 }
128 if (per_cpu(xen_debug_irq, cpu).irq >= 0) {
129 unbind_from_irqhandler(per_cpu(xen_debug_irq, cpu).irq, NULL);
130 per_cpu(xen_debug_irq, cpu).irq = -1;
131 kfree(per_cpu(xen_debug_irq, cpu).name);
132 per_cpu(xen_debug_irq, cpu).name = NULL;
133 }
134 if (per_cpu(xen_callfuncsingle_irq, cpu).irq >= 0) {
135 unbind_from_irqhandler(per_cpu(xen_callfuncsingle_irq, cpu).irq,
136 NULL);
137 per_cpu(xen_callfuncsingle_irq, cpu).irq = -1;
138 kfree(per_cpu(xen_callfuncsingle_irq, cpu).name);
139 per_cpu(xen_callfuncsingle_irq, cpu).name = NULL;
140 }
141 if (xen_hvm_domain())
142 return;
143
144 if (per_cpu(xen_irq_work, cpu).irq >= 0) {
145 unbind_from_irqhandler(per_cpu(xen_irq_work, cpu).irq, NULL);
146 per_cpu(xen_irq_work, cpu).irq = -1;
147 kfree(per_cpu(xen_irq_work, cpu).name);
148 per_cpu(xen_irq_work, cpu).name = NULL;
149 }
150};
151static int xen_smp_intr_init(unsigned int cpu)
152{
153 int rc;
154 char *resched_name, *callfunc_name, *debug_name;
155
156 resched_name = kasprintf(GFP_KERNEL, "resched%d", cpu);
157 rc = bind_ipi_to_irqhandler(XEN_RESCHEDULE_VECTOR,
158 cpu,
159 xen_reschedule_interrupt,
160 IRQF_PERCPU|IRQF_NOBALANCING,
161 resched_name,
162 NULL);
163 if (rc < 0)
164 goto fail;
165 per_cpu(xen_resched_irq, cpu).irq = rc;
166 per_cpu(xen_resched_irq, cpu).name = resched_name;
167
168 callfunc_name = kasprintf(GFP_KERNEL, "callfunc%d", cpu);
169 rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_VECTOR,
170 cpu,
171 xen_call_function_interrupt,
172 IRQF_PERCPU|IRQF_NOBALANCING,
173 callfunc_name,
174 NULL);
175 if (rc < 0)
176 goto fail;
177 per_cpu(xen_callfunc_irq, cpu).irq = rc;
178 per_cpu(xen_callfunc_irq, cpu).name = callfunc_name;
179
180 debug_name = kasprintf(GFP_KERNEL, "debug%d", cpu);
181 rc = bind_virq_to_irqhandler(VIRQ_DEBUG, cpu, xen_debug_interrupt,
182 IRQF_PERCPU | IRQF_NOBALANCING,
183 debug_name, NULL);
184 if (rc < 0)
185 goto fail;
186 per_cpu(xen_debug_irq, cpu).irq = rc;
187 per_cpu(xen_debug_irq, cpu).name = debug_name;
188
189 callfunc_name = kasprintf(GFP_KERNEL, "callfuncsingle%d", cpu);
190 rc = bind_ipi_to_irqhandler(XEN_CALL_FUNCTION_SINGLE_VECTOR,
191 cpu,
192 xen_call_function_single_interrupt,
193 IRQF_PERCPU|IRQF_NOBALANCING,
194 callfunc_name,
195 NULL);
196 if (rc < 0)
197 goto fail;
198 per_cpu(xen_callfuncsingle_irq, cpu).irq = rc;
199 per_cpu(xen_callfuncsingle_irq, cpu).name = callfunc_name;
200
201 /*
202 * The IRQ worker on PVHVM goes through the native path and uses the
203 * IPI mechanism.
204 */
205 if (xen_hvm_domain())
206 return 0;
207
208 callfunc_name = kasprintf(GFP_KERNEL, "irqwork%d", cpu);
209 rc = bind_ipi_to_irqhandler(XEN_IRQ_WORK_VECTOR,
210 cpu,
211 xen_irq_work_interrupt,
212 IRQF_PERCPU|IRQF_NOBALANCING,
213 callfunc_name,
214 NULL);
215 if (rc < 0)
216 goto fail;
217 per_cpu(xen_irq_work, cpu).irq = rc;
218 per_cpu(xen_irq_work, cpu).name = callfunc_name;
219
220 return 0;
221
222 fail:
223 xen_smp_intr_free(cpu);
224 return rc;
225}
226
227static void __init xen_fill_possible_map(void)
228{
229 int i, rc;
230
231 if (xen_initial_domain())
232 return;
233
234 for (i = 0; i < nr_cpu_ids; i++) {
235 rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
236 if (rc >= 0) {
237 num_processors++;
238 set_cpu_possible(i, true);
239 }
240 }
241}
242
243static void __init xen_filter_cpu_maps(void)
244{
245 int i, rc;
246 unsigned int subtract = 0;
247
248 if (!xen_initial_domain())
249 return;
250
251 num_processors = 0;
252 disabled_cpus = 0;
253 for (i = 0; i < nr_cpu_ids; i++) {
254 rc = HYPERVISOR_vcpu_op(VCPUOP_is_up, i, NULL);
255 if (rc >= 0) {
256 num_processors++;
257 set_cpu_possible(i, true);
258 } else {
259 set_cpu_possible(i, false);
260 set_cpu_present(i, false);
261 subtract++;
262 }
263 }
264#ifdef CONFIG_HOTPLUG_CPU
265 /* This is akin to using 'nr_cpus' on the Linux command line.
266 * Which is OK as when we use 'dom0_max_vcpus=X' we can only
267 * have up to X, while nr_cpu_ids is greater than X. This
268 * normally is not a problem, except when CPU hotplugging
269 * is involved and then there might be more than X CPUs
270 * in the guest - which will not work as there is no
271 * hypercall to expand the max number of VCPUs an already
272 * running guest has. So cap it up to X. */
273 if (subtract)
274 nr_cpu_ids = nr_cpu_ids - subtract;
275#endif
276
277}
278
279static void __init xen_smp_prepare_boot_cpu(void)
280{
281 BUG_ON(smp_processor_id() != 0);
282 native_smp_prepare_boot_cpu();
283
284 if (xen_pv_domain()) {
285 if (!xen_feature(XENFEAT_writable_page_tables))
286 /* We've switched to the "real" per-cpu gdt, so make
287 * sure the old memory can be recycled. */
288 make_lowmem_page_readwrite(xen_initial_gdt);
289
290#ifdef CONFIG_X86_32
291 /*
292 * Xen starts us with XEN_FLAT_RING1_DS, but linux code
293 * expects __USER_DS
294 */
295 loadsegment(ds, __USER_DS);
296 loadsegment(es, __USER_DS);
297#endif
298
299 xen_filter_cpu_maps();
300 xen_setup_vcpu_info_placement();
301 }
302 /*
303 * The alternative logic (which patches the unlock/lock) runs before
304 * the smp bootup up code is activated. Hence we need to set this up
305 * the core kernel is being patched. Otherwise we will have only
306 * modules patched but not core code.
307 */
308 xen_init_spinlocks();
309}
310
311static void __init xen_smp_prepare_cpus(unsigned int max_cpus)
312{
313 unsigned cpu;
314 unsigned int i;
315
316 if (skip_ioapic_setup) {
317 char *m = (max_cpus == 0) ?
318 "The nosmp parameter is incompatible with Xen; " \
319 "use Xen dom0_max_vcpus=1 parameter" :
320 "The noapic parameter is incompatible with Xen";
321
322 xen_raw_printk(m);
323 panic(m);
324 }
325 xen_init_lock_cpu(0);
326
327 smp_store_boot_cpu_info();
328 cpu_data(0).x86_max_cores = 1;
329
330 for_each_possible_cpu(i) {
331 zalloc_cpumask_var(&per_cpu(cpu_sibling_map, i), GFP_KERNEL);
332 zalloc_cpumask_var(&per_cpu(cpu_core_map, i), GFP_KERNEL);
333 zalloc_cpumask_var(&per_cpu(cpu_llc_shared_map, i), GFP_KERNEL);
334 }
335 set_cpu_sibling_map(0);
336
337 if (xen_smp_intr_init(0))
338 BUG();
339
340 if (!alloc_cpumask_var(&xen_cpu_initialized_map, GFP_KERNEL))
341 panic("could not allocate xen_cpu_initialized_map\n");
342
343 cpumask_copy(xen_cpu_initialized_map, cpumask_of(0));
344
345 /* Restrict the possible_map according to max_cpus. */
346 while ((num_possible_cpus() > 1) && (num_possible_cpus() > max_cpus)) {
347 for (cpu = nr_cpu_ids - 1; !cpu_possible(cpu); cpu--)
348 continue;
349 set_cpu_possible(cpu, false);
350 }
351
352 for_each_possible_cpu(cpu)
353 set_cpu_present(cpu, true);
354}
355
356static int
357cpu_initialize_context(unsigned int cpu, struct task_struct *idle)
358{
359 struct vcpu_guest_context *ctxt;
360 struct desc_struct *gdt;
361 unsigned long gdt_mfn;
362
363 if (cpumask_test_and_set_cpu(cpu, xen_cpu_initialized_map))
364 return 0;
365
366 ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
367 if (ctxt == NULL)
368 return -ENOMEM;
369
370 gdt = get_cpu_gdt_table(cpu);
371
372#ifdef CONFIG_X86_32
373 /* Note: PVH is not yet supported on x86_32. */
374 ctxt->user_regs.fs = __KERNEL_PERCPU;
375 ctxt->user_regs.gs = __KERNEL_STACK_CANARY;
376#endif
377 ctxt->user_regs.eip = (unsigned long)cpu_bringup_and_idle;
378
379 memset(&ctxt->fpu_ctxt, 0, sizeof(ctxt->fpu_ctxt));
380
381 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
382 ctxt->flags = VGCF_IN_KERNEL;
383 ctxt->user_regs.eflags = 0x1000; /* IOPL_RING1 */
384 ctxt->user_regs.ds = __USER_DS;
385 ctxt->user_regs.es = __USER_DS;
386 ctxt->user_regs.ss = __KERNEL_DS;
387
388 xen_copy_trap_info(ctxt->trap_ctxt);
389
390 ctxt->ldt_ents = 0;
391
392 BUG_ON((unsigned long)gdt & ~PAGE_MASK);
393
394 gdt_mfn = arbitrary_virt_to_mfn(gdt);
395 make_lowmem_page_readonly(gdt);
396 make_lowmem_page_readonly(mfn_to_virt(gdt_mfn));
397
398 ctxt->gdt_frames[0] = gdt_mfn;
399 ctxt->gdt_ents = GDT_ENTRIES;
400
401 ctxt->kernel_ss = __KERNEL_DS;
402 ctxt->kernel_sp = idle->thread.sp0;
403
404#ifdef CONFIG_X86_32
405 ctxt->event_callback_cs = __KERNEL_CS;
406 ctxt->failsafe_callback_cs = __KERNEL_CS;
407#else
408 ctxt->gs_base_kernel = per_cpu_offset(cpu);
409#endif
410 ctxt->event_callback_eip =
411 (unsigned long)xen_hypervisor_callback;
412 ctxt->failsafe_callback_eip =
413 (unsigned long)xen_failsafe_callback;
414 ctxt->user_regs.cs = __KERNEL_CS;
415 per_cpu(xen_cr3, cpu) = __pa(swapper_pg_dir);
416#ifdef CONFIG_X86_32
417 }
418#else
419 } else
420 /* N.B. The user_regs.eip (cpu_bringup_and_idle) is called with
421 * %rdi having the cpu number - which means are passing in
422 * as the first parameter the cpu. Subtle!
423 */
424 ctxt->user_regs.rdi = cpu;
425#endif
426 ctxt->user_regs.esp = idle->thread.sp0 - sizeof(struct pt_regs);
427 ctxt->ctrlreg[3] = xen_pfn_to_cr3(virt_to_mfn(swapper_pg_dir));
428 if (HYPERVISOR_vcpu_op(VCPUOP_initialise, cpu, ctxt))
429 BUG();
430
431 kfree(ctxt);
432 return 0;
433}
434
435static int xen_cpu_up(unsigned int cpu, struct task_struct *idle)
436{
437 int rc;
438
439 per_cpu(current_task, cpu) = idle;
440#ifdef CONFIG_X86_32
441 irq_ctx_init(cpu);
442#else
443 clear_tsk_thread_flag(idle, TIF_FORK);
444#endif
445 per_cpu(kernel_stack, cpu) =
446 (unsigned long)task_stack_page(idle) -
447 KERNEL_STACK_OFFSET + THREAD_SIZE;
448
449 xen_setup_runstate_info(cpu);
450 xen_setup_timer(cpu);
451 xen_init_lock_cpu(cpu);
452
453 per_cpu(cpu_state, cpu) = CPU_UP_PREPARE;
454
455 /* make sure interrupts start blocked */
456 per_cpu(xen_vcpu, cpu)->evtchn_upcall_mask = 1;
457
458 rc = cpu_initialize_context(cpu, idle);
459 if (rc)
460 return rc;
461
462 if (num_online_cpus() == 1)
463 /* Just in case we booted with a single CPU. */
464 alternatives_enable_smp();
465
466 rc = xen_smp_intr_init(cpu);
467 if (rc)
468 return rc;
469
470 rc = HYPERVISOR_vcpu_op(VCPUOP_up, cpu, NULL);
471 BUG_ON(rc);
472
473 while(per_cpu(cpu_state, cpu) != CPU_ONLINE) {
474 HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
475 barrier();
476 }
477
478 return 0;
479}
480
481static void xen_smp_cpus_done(unsigned int max_cpus)
482{
483}
484
485#ifdef CONFIG_HOTPLUG_CPU
486static int xen_cpu_disable(void)
487{
488 unsigned int cpu = smp_processor_id();
489 if (cpu == 0)
490 return -EBUSY;
491
492 cpu_disable_common();
493
494 load_cr3(swapper_pg_dir);
495 return 0;
496}
497
498static void xen_cpu_die(unsigned int cpu)
499{
500 while (xen_pv_domain() && HYPERVISOR_vcpu_op(VCPUOP_is_up, cpu, NULL)) {
501 current->state = TASK_UNINTERRUPTIBLE;
502 schedule_timeout(HZ/10);
503 }
504 xen_smp_intr_free(cpu);
505 xen_uninit_lock_cpu(cpu);
506 xen_teardown_timer(cpu);
507}
508
509static void xen_play_dead(void) /* used only with HOTPLUG_CPU */
510{
511 play_dead_common();
512 HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
513 cpu_bringup();
514 /*
515 * commit 4b0c0f294 (tick: Cleanup NOHZ per cpu data on cpu down)
516 * clears certain data that the cpu_idle loop (which called us
517 * and that we return from) expects. The only way to get that
518 * data back is to call:
519 */
520 tick_nohz_idle_enter();
521}
522
523#else /* !CONFIG_HOTPLUG_CPU */
524static int xen_cpu_disable(void)
525{
526 return -ENOSYS;
527}
528
529static void xen_cpu_die(unsigned int cpu)
530{
531 BUG();
532}
533
534static void xen_play_dead(void)
535{
536 BUG();
537}
538
539#endif
540static void stop_self(void *v)
541{
542 int cpu = smp_processor_id();
543
544 /* make sure we're not pinning something down */
545 load_cr3(swapper_pg_dir);
546 /* should set up a minimal gdt */
547
548 set_cpu_online(cpu, false);
549
550 HYPERVISOR_vcpu_op(VCPUOP_down, cpu, NULL);
551 BUG();
552}
553
554static void xen_stop_other_cpus(int wait)
555{
556 smp_call_function(stop_self, NULL, wait);
557}
558
559static void xen_smp_send_reschedule(int cpu)
560{
561 xen_send_IPI_one(cpu, XEN_RESCHEDULE_VECTOR);
562}
563
564static void __xen_send_IPI_mask(const struct cpumask *mask,
565 int vector)
566{
567 unsigned cpu;
568
569 for_each_cpu_and(cpu, mask, cpu_online_mask)
570 xen_send_IPI_one(cpu, vector);
571}
572
573static void xen_smp_send_call_function_ipi(const struct cpumask *mask)
574{
575 int cpu;
576
577 __xen_send_IPI_mask(mask, XEN_CALL_FUNCTION_VECTOR);
578
579 /* Make sure other vcpus get a chance to run if they need to. */
580 for_each_cpu(cpu, mask) {
581 if (xen_vcpu_stolen(cpu)) {
582 HYPERVISOR_sched_op(SCHEDOP_yield, NULL);
583 break;
584 }
585 }
586}
587
588static void xen_smp_send_call_function_single_ipi(int cpu)
589{
590 __xen_send_IPI_mask(cpumask_of(cpu),
591 XEN_CALL_FUNCTION_SINGLE_VECTOR);
592}
593
594static inline int xen_map_vector(int vector)
595{
596 int xen_vector;
597
598 switch (vector) {
599 case RESCHEDULE_VECTOR:
600 xen_vector = XEN_RESCHEDULE_VECTOR;
601 break;
602 case CALL_FUNCTION_VECTOR:
603 xen_vector = XEN_CALL_FUNCTION_VECTOR;
604 break;
605 case CALL_FUNCTION_SINGLE_VECTOR:
606 xen_vector = XEN_CALL_FUNCTION_SINGLE_VECTOR;
607 break;
608 case IRQ_WORK_VECTOR:
609 xen_vector = XEN_IRQ_WORK_VECTOR;
610 break;
611#ifdef CONFIG_X86_64
612 case NMI_VECTOR:
613 case APIC_DM_NMI: /* Some use that instead of NMI_VECTOR */
614 xen_vector = XEN_NMI_VECTOR;
615 break;
616#endif
617 default:
618 xen_vector = -1;
619 printk(KERN_ERR "xen: vector 0x%x is not implemented\n",
620 vector);
621 }
622
623 return xen_vector;
624}
625
626void xen_send_IPI_mask(const struct cpumask *mask,
627 int vector)
628{
629 int xen_vector = xen_map_vector(vector);
630
631 if (xen_vector >= 0)
632 __xen_send_IPI_mask(mask, xen_vector);
633}
634
635void xen_send_IPI_all(int vector)
636{
637 int xen_vector = xen_map_vector(vector);
638
639 if (xen_vector >= 0)
640 __xen_send_IPI_mask(cpu_online_mask, xen_vector);
641}
642
643void xen_send_IPI_self(int vector)
644{
645 int xen_vector = xen_map_vector(vector);
646
647 if (xen_vector >= 0)
648 xen_send_IPI_one(smp_processor_id(), xen_vector);
649}
650
651void xen_send_IPI_mask_allbutself(const struct cpumask *mask,
652 int vector)
653{
654 unsigned cpu;
655 unsigned int this_cpu = smp_processor_id();
656 int xen_vector = xen_map_vector(vector);
657
658 if (!(num_online_cpus() > 1) || (xen_vector < 0))
659 return;
660
661 for_each_cpu_and(cpu, mask, cpu_online_mask) {
662 if (this_cpu == cpu)
663 continue;
664
665 xen_send_IPI_one(cpu, xen_vector);
666 }
667}
668
669void xen_send_IPI_allbutself(int vector)
670{
671 xen_send_IPI_mask_allbutself(cpu_online_mask, vector);
672}
673
674static irqreturn_t xen_call_function_interrupt(int irq, void *dev_id)
675{
676 irq_enter();
677 generic_smp_call_function_interrupt();
678 inc_irq_stat(irq_call_count);
679 irq_exit();
680
681 return IRQ_HANDLED;
682}
683
684static irqreturn_t xen_call_function_single_interrupt(int irq, void *dev_id)
685{
686 irq_enter();
687 generic_smp_call_function_single_interrupt();
688 inc_irq_stat(irq_call_count);
689 irq_exit();
690
691 return IRQ_HANDLED;
692}
693
694static irqreturn_t xen_irq_work_interrupt(int irq, void *dev_id)
695{
696 irq_enter();
697 irq_work_run();
698 inc_irq_stat(apic_irq_work_irqs);
699 irq_exit();
700
701 return IRQ_HANDLED;
702}
703
704static const struct smp_ops xen_smp_ops __initconst = {
705 .smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu,
706 .smp_prepare_cpus = xen_smp_prepare_cpus,
707 .smp_cpus_done = xen_smp_cpus_done,
708
709 .cpu_up = xen_cpu_up,
710 .cpu_die = xen_cpu_die,
711 .cpu_disable = xen_cpu_disable,
712 .play_dead = xen_play_dead,
713
714 .stop_other_cpus = xen_stop_other_cpus,
715 .smp_send_reschedule = xen_smp_send_reschedule,
716
717 .send_call_func_ipi = xen_smp_send_call_function_ipi,
718 .send_call_func_single_ipi = xen_smp_send_call_function_single_ipi,
719};
720
721void __init xen_smp_init(void)
722{
723 smp_ops = xen_smp_ops;
724 xen_fill_possible_map();
725}
726
727static void __init xen_hvm_smp_prepare_cpus(unsigned int max_cpus)
728{
729 native_smp_prepare_cpus(max_cpus);
730 WARN_ON(xen_smp_intr_init(0));
731
732 xen_init_lock_cpu(0);
733}
734
735static int xen_hvm_cpu_up(unsigned int cpu, struct task_struct *tidle)
736{
737 int rc;
738 /*
739 * xen_smp_intr_init() needs to run before native_cpu_up()
740 * so that IPI vectors are set up on the booting CPU before
741 * it is marked online in native_cpu_up().
742 */
743 rc = xen_smp_intr_init(cpu);
744 WARN_ON(rc);
745 if (!rc)
746 rc = native_cpu_up(cpu, tidle);
747
748 /*
749 * We must initialize the slowpath CPU kicker _after_ the native
750 * path has executed. If we initialized it before none of the
751 * unlocker IPI kicks would reach the booting CPU as the booting
752 * CPU had not set itself 'online' in cpu_online_mask. That mask
753 * is checked when IPIs are sent (on HVM at least).
754 */
755 xen_init_lock_cpu(cpu);
756 return rc;
757}
758
759static void xen_hvm_cpu_die(unsigned int cpu)
760{
761 xen_cpu_die(cpu);
762 native_cpu_die(cpu);
763}
764
765void __init xen_hvm_smp_init(void)
766{
767 if (!xen_have_vector_callback)
768 return;
769 smp_ops.smp_prepare_cpus = xen_hvm_smp_prepare_cpus;
770 smp_ops.smp_send_reschedule = xen_smp_send_reschedule;
771 smp_ops.cpu_up = xen_hvm_cpu_up;
772 smp_ops.cpu_die = xen_hvm_cpu_die;
773 smp_ops.send_call_func_ipi = xen_smp_send_call_function_ipi;
774 smp_ops.send_call_func_single_ipi = xen_smp_send_call_function_single_ipi;
775 smp_ops.smp_prepare_boot_cpu = xen_smp_prepare_boot_cpu;
776}