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
32 static unsigned long shared_info_pfn;
33
34 __ro_after_init bool xen_percpu_upcall;
35 EXPORT_SYMBOL_GPL(xen_percpu_upcall);
36
xen_hvm_init_shared_info(void)37 void xen_hvm_init_shared_info(void)
38 {
39 struct xen_add_to_physmap xatp;
40
41 xatp.domid = DOMID_SELF;
42 xatp.idx = 0;
43 xatp.space = XENMAPSPACE_shared_info;
44 xatp.gpfn = shared_info_pfn;
45 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
46 BUG();
47 }
48
reserve_shared_info(void)49 static void __init reserve_shared_info(void)
50 {
51 u64 pa;
52
53 /*
54 * Search for a free page starting at 4kB physical address.
55 * Low memory is preferred to avoid an EPT large page split up
56 * by the mapping.
57 * Starting below X86_RESERVE_LOW (usually 64kB) is fine as
58 * the BIOS used for HVM guests is well behaved and won't
59 * clobber memory other than the first 4kB.
60 */
61 for (pa = PAGE_SIZE;
62 !e820__mapped_all(pa, pa + PAGE_SIZE, E820_TYPE_RAM) ||
63 memblock_is_reserved(pa);
64 pa += PAGE_SIZE)
65 ;
66
67 shared_info_pfn = PHYS_PFN(pa);
68
69 memblock_reserve(pa, PAGE_SIZE);
70 HYPERVISOR_shared_info = early_memremap(pa, PAGE_SIZE);
71 }
72
xen_hvm_init_mem_mapping(void)73 static void __init xen_hvm_init_mem_mapping(void)
74 {
75 early_memunmap(HYPERVISOR_shared_info, PAGE_SIZE);
76 HYPERVISOR_shared_info = __va(PFN_PHYS(shared_info_pfn));
77
78 /*
79 * The virtual address of the shared_info page has changed, so
80 * the vcpu_info pointer for VCPU 0 is now stale.
81 *
82 * The prepare_boot_cpu callback will re-initialize it via
83 * xen_vcpu_setup, but we can't rely on that to be called for
84 * old Xen versions (xen_have_vector_callback == 0).
85 *
86 * It is, in any case, bad to have a stale vcpu_info pointer
87 * so reset it now.
88 */
89 xen_vcpu_info_reset(0);
90 }
91
init_hvm_pv_info(void)92 static void __init init_hvm_pv_info(void)
93 {
94 int major, minor;
95 uint32_t eax, ebx, ecx, edx, base;
96
97 base = xen_cpuid_base();
98 eax = cpuid_eax(base + 1);
99
100 major = eax >> 16;
101 minor = eax & 0xffff;
102 printk(KERN_INFO "Xen version %d.%d.\n", major, minor);
103
104 xen_domain_type = XEN_HVM_DOMAIN;
105
106 /* PVH set up hypercall page in xen_prepare_pvh(). */
107 if (xen_pvh_domain())
108 pv_info.name = "Xen PVH";
109 else
110 pv_info.name = "Xen HVM";
111
112 xen_setup_features();
113
114 cpuid(base + 4, &eax, &ebx, &ecx, &edx);
115 if (eax & XEN_HVM_CPUID_VCPU_ID_PRESENT)
116 this_cpu_write(xen_vcpu_id, ebx);
117 else
118 this_cpu_write(xen_vcpu_id, smp_processor_id());
119 }
120
DEFINE_IDTENTRY_SYSVEC(sysvec_xen_hvm_callback)121 DEFINE_IDTENTRY_SYSVEC(sysvec_xen_hvm_callback)
122 {
123 struct pt_regs *old_regs = set_irq_regs(regs);
124
125 if (xen_percpu_upcall)
126 apic_eoi();
127
128 inc_irq_stat(irq_hv_callback_count);
129
130 xen_evtchn_do_upcall();
131
132 set_irq_regs(old_regs);
133 }
134
135 #ifdef CONFIG_KEXEC_CORE
xen_hvm_shutdown(void)136 static void xen_hvm_shutdown(void)
137 {
138 native_machine_shutdown();
139 if (kexec_in_progress)
140 xen_reboot(SHUTDOWN_soft_reset);
141 }
142 #endif
143
144 #ifdef CONFIG_CRASH_DUMP
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) != CPU_ACPIID_INVALID)
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 #endif
233 #ifdef CONFIG_CRASH_DUMP
234 machine_ops.crash_shutdown = xen_hvm_crash_shutdown;
235 #endif
236 }
237
xen_parse_nopv(char * arg)238 static __init int xen_parse_nopv(char *arg)
239 {
240 pr_notice("\"xen_nopv\" is deprecated, please use \"nopv\" instead\n");
241
242 if (xen_cpuid_base())
243 nopv = true;
244 return 0;
245 }
246 early_param("xen_nopv", xen_parse_nopv);
247
xen_parse_no_vector_callback(char * arg)248 static __init int xen_parse_no_vector_callback(char *arg)
249 {
250 xen_have_vector_callback = false;
251 return 0;
252 }
253 early_param("xen_no_vector_callback", xen_parse_no_vector_callback);
254
xen_x2apic_available(void)255 static __init bool xen_x2apic_available(void)
256 {
257 return x2apic_supported();
258 }
259
msi_ext_dest_id(void)260 static bool __init msi_ext_dest_id(void)
261 {
262 return cpuid_eax(xen_cpuid_base() + 4) & XEN_HVM_CPUID_EXT_DEST_ID;
263 }
264
xen_hvm_guest_late_init(void)265 static __init void xen_hvm_guest_late_init(void)
266 {
267 #ifdef CONFIG_XEN_PVH
268 /* Test for PVH domain (PVH boot path taken overrides ACPI flags). */
269 if (!xen_pvh &&
270 (x86_platform.legacy.rtc || !x86_platform.legacy.no_vga))
271 return;
272
273 /* PVH detected. */
274 xen_pvh = true;
275
276 if (nopv)
277 panic("\"nopv\" and \"xen_nopv\" parameters are unsupported in PVH guest.");
278
279 /* Make sure we don't fall back to (default) ACPI_IRQ_MODEL_PIC. */
280 if (!nr_ioapics && acpi_irq_model == ACPI_IRQ_MODEL_PIC)
281 acpi_irq_model = ACPI_IRQ_MODEL_PLATFORM;
282
283 machine_ops.emergency_restart = xen_emergency_restart;
284 pv_info.name = "Xen PVH";
285 #endif
286 }
287
xen_platform_hvm(void)288 static uint32_t __init xen_platform_hvm(void)
289 {
290 uint32_t xen_domain = xen_cpuid_base();
291 struct x86_hyper_init *h = &x86_hyper_xen_hvm.init;
292
293 if (xen_pv_domain())
294 return 0;
295
296 /* Set correct hypercall function. */
297 if (xen_domain)
298 xen_hypercall_setfunc();
299
300 if (xen_pvh_domain() && nopv) {
301 /* Guest booting via the Xen-PVH boot entry goes here */
302 pr_info("\"nopv\" parameter is ignored in PVH guest\n");
303 nopv = false;
304 } else if (nopv && xen_domain) {
305 /*
306 * Guest booting via normal boot entry (like via grub2) goes
307 * here.
308 *
309 * Use interface functions for bare hardware if nopv,
310 * xen_hvm_guest_late_init is an exception as we need to
311 * detect PVH and panic there.
312 */
313 h->init_platform = x86_init_noop;
314 h->x2apic_available = bool_x86_init_noop;
315 h->init_mem_mapping = x86_init_noop;
316 h->init_after_bootmem = x86_init_noop;
317 h->guest_late_init = xen_hvm_guest_late_init;
318 x86_hyper_xen_hvm.runtime.pin_vcpu = x86_op_int_noop;
319 }
320 return xen_domain;
321 }
322
323 struct hypervisor_x86 x86_hyper_xen_hvm __initdata = {
324 .name = "Xen HVM",
325 .detect = xen_platform_hvm,
326 .type = X86_HYPER_XEN_HVM,
327 .init.init_platform = xen_hvm_guest_init,
328 .init.x2apic_available = xen_x2apic_available,
329 .init.init_mem_mapping = xen_hvm_init_mem_mapping,
330 .init.guest_late_init = xen_hvm_guest_late_init,
331 .init.msi_ext_dest_id = msi_ext_dest_id,
332 .runtime.pin_vcpu = xen_pin_vcpu,
333 .ignore_nopv = true,
334 };
335