1 // SPDX-License-Identifier: GPL-2.0
2
3 #include <linux/acpi.h>
4 #include <linux/cpu.h>
5 #include <linux/kexec.h>
6 #include <linux/memblock.h>
7 #include <linux/virtio_anchor.h>
8
9 #include <xen/features.h>
10 #include <xen/events.h>
11 #include <xen/hvm.h>
12 #include <xen/interface/hvm/hvm_op.h>
13 #include <xen/interface/memory.h>
14
15 #include <asm/apic.h>
16 #include <asm/cpu.h>
17 #include <asm/smp.h>
18 #include <asm/io_apic.h>
19 #include <asm/reboot.h>
20 #include <asm/setup.h>
21 #include <asm/idtentry.h>
22 #include <asm/hypervisor.h>
23 #include <asm/e820/api.h>
24 #include <asm/early_ioremap.h>
25
26 #include <asm/xen/cpuid.h>
27 #include <asm/xen/hypervisor.h>
28 #include <asm/xen/page.h>
29
30 #include "xen-ops.h"
31 #include "mmu.h"
32 #include "smp.h"
33
34 static unsigned long shared_info_pfn;
35
36 __ro_after_init bool xen_percpu_upcall;
37 EXPORT_SYMBOL_GPL(xen_percpu_upcall);
38
xen_hvm_init_shared_info(void)39 void xen_hvm_init_shared_info(void)
40 {
41 struct xen_add_to_physmap xatp;
42
43 xatp.domid = DOMID_SELF;
44 xatp.idx = 0;
45 xatp.space = XENMAPSPACE_shared_info;
46 xatp.gpfn = shared_info_pfn;
47 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
48 BUG();
49 }
50
reserve_shared_info(void)51 static void __init reserve_shared_info(void)
52 {
53 u64 pa;
54
55 /*
56 * Search for a free page starting at 4kB physical address.
57 * Low memory is preferred to avoid an EPT large page split up
58 * by the mapping.
59 * Starting below X86_RESERVE_LOW (usually 64kB) is fine as
60 * the BIOS used for HVM guests is well behaved and won't
61 * clobber memory other than the first 4kB.
62 */
63 for (pa = PAGE_SIZE;
64 !e820__mapped_all(pa, pa + PAGE_SIZE, E820_TYPE_RAM) ||
65 memblock_is_reserved(pa);
66 pa += PAGE_SIZE)
67 ;
68
69 shared_info_pfn = PHYS_PFN(pa);
70
71 memblock_reserve(pa, PAGE_SIZE);
72 HYPERVISOR_shared_info = early_memremap(pa, PAGE_SIZE);
73 }
74
xen_hvm_init_mem_mapping(void)75 static void __init xen_hvm_init_mem_mapping(void)
76 {
77 early_memunmap(HYPERVISOR_shared_info, PAGE_SIZE);
78 HYPERVISOR_shared_info = __va(PFN_PHYS(shared_info_pfn));
79
80 /*
81 * The virtual address of the shared_info page has changed, so
82 * the vcpu_info pointer for VCPU 0 is now stale.
83 *
84 * The prepare_boot_cpu callback will re-initialize it via
85 * xen_vcpu_setup, but we can't rely on that to be called for
86 * old Xen versions (xen_have_vector_callback == 0).
87 *
88 * It is, in any case, bad to have a stale vcpu_info pointer
89 * so reset it now.
90 */
91 xen_vcpu_info_reset(0);
92 }
93
init_hvm_pv_info(void)94 static void __init init_hvm_pv_info(void)
95 {
96 int major, minor;
97 uint32_t eax, ebx, ecx, edx, base;
98
99 base = xen_cpuid_base();
100 eax = cpuid_eax(base + 1);
101
102 major = eax >> 16;
103 minor = eax & 0xffff;
104 printk(KERN_INFO "Xen version %d.%d.\n", major, minor);
105
106 xen_domain_type = XEN_HVM_DOMAIN;
107
108 /* PVH set up hypercall page in xen_prepare_pvh(). */
109 if (xen_pvh_domain())
110 pv_info.name = "Xen PVH";
111 else
112 pv_info.name = "Xen HVM";
113
114 xen_setup_features();
115
116 cpuid(base + 4, &eax, &ebx, &ecx, &edx);
117 if (eax & XEN_HVM_CPUID_VCPU_ID_PRESENT)
118 this_cpu_write(xen_vcpu_id, ebx);
119 else
120 this_cpu_write(xen_vcpu_id, smp_processor_id());
121 }
122
DEFINE_IDTENTRY_SYSVEC(sysvec_xen_hvm_callback)123 DEFINE_IDTENTRY_SYSVEC(sysvec_xen_hvm_callback)
124 {
125 struct pt_regs *old_regs = set_irq_regs(regs);
126
127 if (xen_percpu_upcall)
128 apic_eoi();
129
130 inc_irq_stat(irq_hv_callback_count);
131
132 xen_evtchn_do_upcall();
133
134 set_irq_regs(old_regs);
135 }
136
137 #ifdef CONFIG_KEXEC_CORE
xen_hvm_shutdown(void)138 static void xen_hvm_shutdown(void)
139 {
140 native_machine_shutdown();
141 if (kexec_in_progress)
142 xen_reboot(SHUTDOWN_soft_reset);
143 }
144
xen_hvm_crash_shutdown(struct pt_regs * regs)145 static void xen_hvm_crash_shutdown(struct pt_regs *regs)
146 {
147 native_machine_crash_shutdown(regs);
148 xen_reboot(SHUTDOWN_soft_reset);
149 }
150 #endif
151
xen_cpu_up_prepare_hvm(unsigned int cpu)152 static int xen_cpu_up_prepare_hvm(unsigned int cpu)
153 {
154 int rc = 0;
155
156 /*
157 * If a CPU was offlined earlier and offlining timed out then the
158 * lock mechanism is still initialized. Uninit it unconditionally
159 * as it's safe to call even if already uninited. Interrupts and
160 * timer have already been handled in xen_cpu_dead_hvm().
161 */
162 xen_uninit_lock_cpu(cpu);
163
164 if (cpu_acpi_id(cpu) != U32_MAX)
165 per_cpu(xen_vcpu_id, cpu) = cpu_acpi_id(cpu);
166 else
167 per_cpu(xen_vcpu_id, cpu) = cpu;
168 xen_vcpu_setup(cpu);
169 if (!xen_have_vector_callback)
170 return 0;
171
172 if (xen_percpu_upcall) {
173 rc = xen_set_upcall_vector(cpu);
174 if (rc) {
175 WARN(1, "HVMOP_set_evtchn_upcall_vector"
176 " for CPU %d failed: %d\n", cpu, rc);
177 return rc;
178 }
179 }
180
181 if (xen_feature(XENFEAT_hvm_safe_pvclock))
182 xen_setup_timer(cpu);
183
184 rc = xen_smp_intr_init(cpu);
185 if (rc) {
186 WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
187 cpu, rc);
188 }
189 return rc;
190 }
191
xen_cpu_dead_hvm(unsigned int cpu)192 static int xen_cpu_dead_hvm(unsigned int cpu)
193 {
194 xen_smp_intr_free(cpu);
195
196 if (xen_have_vector_callback && xen_feature(XENFEAT_hvm_safe_pvclock))
197 xen_teardown_timer(cpu);
198 return 0;
199 }
200
xen_hvm_guest_init(void)201 static void __init xen_hvm_guest_init(void)
202 {
203 if (xen_pv_domain())
204 return;
205
206 if (IS_ENABLED(CONFIG_XEN_VIRTIO_FORCE_GRANT))
207 virtio_set_mem_acc_cb(xen_virtio_restricted_mem_acc);
208
209 init_hvm_pv_info();
210
211 reserve_shared_info();
212 xen_hvm_init_shared_info();
213
214 /*
215 * xen_vcpu is a pointer to the vcpu_info struct in the shared_info
216 * page, we use it in the event channel upcall and in some pvclock
217 * related functions.
218 */
219 xen_vcpu_info_reset(0);
220
221 xen_panic_handler_init();
222
223 xen_hvm_smp_init();
224 WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_hvm, xen_cpu_dead_hvm));
225 xen_unplug_emulated_devices();
226 x86_init.irqs.intr_init = xen_init_IRQ;
227 xen_hvm_init_time_ops();
228 xen_hvm_init_mmu_ops();
229
230 #ifdef CONFIG_KEXEC_CORE
231 machine_ops.shutdown = xen_hvm_shutdown;
232 machine_ops.crash_shutdown = xen_hvm_crash_shutdown;
233 #endif
234 }
235
xen_parse_nopv(char * arg)236 static __init int xen_parse_nopv(char *arg)
237 {
238 pr_notice("\"xen_nopv\" is deprecated, please use \"nopv\" instead\n");
239
240 if (xen_cpuid_base())
241 nopv = true;
242 return 0;
243 }
244 early_param("xen_nopv", xen_parse_nopv);
245
xen_parse_no_vector_callback(char * arg)246 static __init int xen_parse_no_vector_callback(char *arg)
247 {
248 xen_have_vector_callback = false;
249 return 0;
250 }
251 early_param("xen_no_vector_callback", xen_parse_no_vector_callback);
252
xen_x2apic_available(void)253 static __init bool xen_x2apic_available(void)
254 {
255 return x2apic_supported();
256 }
257
msi_ext_dest_id(void)258 static bool __init msi_ext_dest_id(void)
259 {
260 return cpuid_eax(xen_cpuid_base() + 4) & XEN_HVM_CPUID_EXT_DEST_ID;
261 }
262
xen_hvm_guest_late_init(void)263 static __init void xen_hvm_guest_late_init(void)
264 {
265 #ifdef CONFIG_XEN_PVH
266 /* Test for PVH domain (PVH boot path taken overrides ACPI flags). */
267 if (!xen_pvh &&
268 (x86_platform.legacy.rtc || !x86_platform.legacy.no_vga))
269 return;
270
271 /* PVH detected. */
272 xen_pvh = true;
273
274 if (nopv)
275 panic("\"nopv\" and \"xen_nopv\" parameters are unsupported in PVH guest.");
276
277 /* Make sure we don't fall back to (default) ACPI_IRQ_MODEL_PIC. */
278 if (!nr_ioapics && acpi_irq_model == ACPI_IRQ_MODEL_PIC)
279 acpi_irq_model = ACPI_IRQ_MODEL_PLATFORM;
280
281 machine_ops.emergency_restart = xen_emergency_restart;
282 pv_info.name = "Xen PVH";
283 #endif
284 }
285
xen_platform_hvm(void)286 static uint32_t __init xen_platform_hvm(void)
287 {
288 uint32_t xen_domain = xen_cpuid_base();
289 struct x86_hyper_init *h = &x86_hyper_xen_hvm.init;
290
291 if (xen_pv_domain())
292 return 0;
293
294 /* Set correct hypercall function. */
295 if (xen_domain)
296 xen_hypercall_setfunc();
297
298 if (xen_pvh_domain() && nopv) {
299 /* Guest booting via the Xen-PVH boot entry goes here */
300 pr_info("\"nopv\" parameter is ignored in PVH guest\n");
301 nopv = false;
302 } else if (nopv && xen_domain) {
303 /*
304 * Guest booting via normal boot entry (like via grub2) goes
305 * here.
306 *
307 * Use interface functions for bare hardware if nopv,
308 * xen_hvm_guest_late_init is an exception as we need to
309 * detect PVH and panic there.
310 */
311 h->init_platform = x86_init_noop;
312 h->x2apic_available = bool_x86_init_noop;
313 h->init_mem_mapping = x86_init_noop;
314 h->init_after_bootmem = x86_init_noop;
315 h->guest_late_init = xen_hvm_guest_late_init;
316 x86_hyper_xen_hvm.runtime.pin_vcpu = x86_op_int_noop;
317 }
318 return xen_domain;
319 }
320
321 struct hypervisor_x86 x86_hyper_xen_hvm __initdata = {
322 .name = "Xen HVM",
323 .detect = xen_platform_hvm,
324 .type = X86_HYPER_XEN_HVM,
325 .init.init_platform = xen_hvm_guest_init,
326 .init.x2apic_available = xen_x2apic_available,
327 .init.init_mem_mapping = xen_hvm_init_mem_mapping,
328 .init.guest_late_init = xen_hvm_guest_late_init,
329 .init.msi_ext_dest_id = msi_ext_dest_id,
330 .runtime.pin_vcpu = xen_pin_vcpu,
331 .ignore_nopv = true,
332 };
333