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
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
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
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
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 u64 pfn;
111 uint32_t msr;
112
113 pv_info.name = "Xen HVM";
114 msr = cpuid_ebx(base + 2);
115 pfn = __pa(hypercall_page);
116 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
117 }
118
119 xen_setup_features();
120
121 cpuid(base + 4, &eax, &ebx, &ecx, &edx);
122 if (eax & XEN_HVM_CPUID_VCPU_ID_PRESENT)
123 this_cpu_write(xen_vcpu_id, ebx);
124 else
125 this_cpu_write(xen_vcpu_id, smp_processor_id());
126 }
127
128 DEFINE_IDTENTRY_SYSVEC(sysvec_xen_hvm_callback)
129 {
130 struct pt_regs *old_regs = set_irq_regs(regs);
131
132 if (xen_percpu_upcall)
133 apic_eoi();
134
135 inc_irq_stat(irq_hv_callback_count);
136
137 xen_evtchn_do_upcall();
138
139 set_irq_regs(old_regs);
140 }
141
142 #ifdef CONFIG_KEXEC_CORE
143 static void xen_hvm_shutdown(void)
144 {
145 native_machine_shutdown();
146 if (kexec_in_progress)
147 xen_reboot(SHUTDOWN_soft_reset);
148 }
149 #endif
150
151 #ifdef CONFIG_CRASH_DUMP
152 static void xen_hvm_crash_shutdown(struct pt_regs *regs)
153 {
154 native_machine_crash_shutdown(regs);
155 xen_reboot(SHUTDOWN_soft_reset);
156 }
157 #endif
158
159 static int xen_cpu_up_prepare_hvm(unsigned int cpu)
160 {
161 int rc = 0;
162
163 /*
164 * If a CPU was offlined earlier and offlining timed out then the
165 * lock mechanism is still initialized. Uninit it unconditionally
166 * as it's safe to call even if already uninited. Interrupts and
167 * timer have already been handled in xen_cpu_dead_hvm().
168 */
169 xen_uninit_lock_cpu(cpu);
170
171 if (cpu_acpi_id(cpu) != CPU_ACPIID_INVALID)
172 per_cpu(xen_vcpu_id, cpu) = cpu_acpi_id(cpu);
173 else
174 per_cpu(xen_vcpu_id, cpu) = cpu;
175 xen_vcpu_setup(cpu);
176 if (!xen_have_vector_callback)
177 return 0;
178
179 if (xen_percpu_upcall) {
180 rc = xen_set_upcall_vector(cpu);
181 if (rc) {
182 WARN(1, "HVMOP_set_evtchn_upcall_vector"
183 " for CPU %d failed: %d\n", cpu, rc);
184 return rc;
185 }
186 }
187
188 if (xen_feature(XENFEAT_hvm_safe_pvclock))
189 xen_setup_timer(cpu);
190
191 rc = xen_smp_intr_init(cpu);
192 if (rc) {
193 WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
194 cpu, rc);
195 }
196 return rc;
197 }
198
199 static int xen_cpu_dead_hvm(unsigned int cpu)
200 {
201 xen_smp_intr_free(cpu);
202
203 if (xen_have_vector_callback && xen_feature(XENFEAT_hvm_safe_pvclock))
204 xen_teardown_timer(cpu);
205 return 0;
206 }
207
208 static void __init xen_hvm_guest_init(void)
209 {
210 if (xen_pv_domain())
211 return;
212
213 if (IS_ENABLED(CONFIG_XEN_VIRTIO_FORCE_GRANT))
214 virtio_set_mem_acc_cb(xen_virtio_restricted_mem_acc);
215
216 init_hvm_pv_info();
217
218 reserve_shared_info();
219 xen_hvm_init_shared_info();
220
221 /*
222 * xen_vcpu is a pointer to the vcpu_info struct in the shared_info
223 * page, we use it in the event channel upcall and in some pvclock
224 * related functions.
225 */
226 xen_vcpu_info_reset(0);
227
228 xen_panic_handler_init();
229
230 xen_hvm_smp_init();
231 WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_hvm, xen_cpu_dead_hvm));
232 xen_unplug_emulated_devices();
233 x86_init.irqs.intr_init = xen_init_IRQ;
234 xen_hvm_init_time_ops();
235 xen_hvm_init_mmu_ops();
236
237 #ifdef CONFIG_KEXEC_CORE
238 machine_ops.shutdown = xen_hvm_shutdown;
239 #endif
240 #ifdef CONFIG_CRASH_DUMP
241 machine_ops.crash_shutdown = xen_hvm_crash_shutdown;
242 #endif
243 }
244
245 static __init int xen_parse_nopv(char *arg)
246 {
247 pr_notice("\"xen_nopv\" is deprecated, please use \"nopv\" instead\n");
248
249 if (xen_cpuid_base())
250 nopv = true;
251 return 0;
252 }
253 early_param("xen_nopv", xen_parse_nopv);
254
255 static __init int xen_parse_no_vector_callback(char *arg)
256 {
257 xen_have_vector_callback = false;
258 return 0;
259 }
260 early_param("xen_no_vector_callback", xen_parse_no_vector_callback);
261
262 static __init bool xen_x2apic_available(void)
263 {
264 return x2apic_supported();
265 }
266
267 static bool __init msi_ext_dest_id(void)
268 {
269 return cpuid_eax(xen_cpuid_base() + 4) & XEN_HVM_CPUID_EXT_DEST_ID;
270 }
271
272 static __init void xen_hvm_guest_late_init(void)
273 {
274 #ifdef CONFIG_XEN_PVH
275 /* Test for PVH domain (PVH boot path taken overrides ACPI flags). */
276 if (!xen_pvh &&
277 (x86_platform.legacy.rtc || !x86_platform.legacy.no_vga))
278 return;
279
280 /* PVH detected. */
281 xen_pvh = true;
282
283 if (nopv)
284 panic("\"nopv\" and \"xen_nopv\" parameters are unsupported in PVH guest.");
285
286 /* Make sure we don't fall back to (default) ACPI_IRQ_MODEL_PIC. */
287 if (!nr_ioapics && acpi_irq_model == ACPI_IRQ_MODEL_PIC)
288 acpi_irq_model = ACPI_IRQ_MODEL_PLATFORM;
289
290 machine_ops.emergency_restart = xen_emergency_restart;
291 pv_info.name = "Xen PVH";
292 #endif
293 }
294
295 static uint32_t __init xen_platform_hvm(void)
296 {
297 uint32_t xen_domain = xen_cpuid_base();
298 struct x86_hyper_init *h = &x86_hyper_xen_hvm.init;
299
300 if (xen_pv_domain())
301 return 0;
302
303 if (xen_pvh_domain() && nopv) {
304 /* Guest booting via the Xen-PVH boot entry goes here */
305 pr_info("\"nopv\" parameter is ignored in PVH guest\n");
306 nopv = false;
307 } else if (nopv && xen_domain) {
308 /*
309 * Guest booting via normal boot entry (like via grub2) goes
310 * here.
311 *
312 * Use interface functions for bare hardware if nopv,
313 * xen_hvm_guest_late_init is an exception as we need to
314 * detect PVH and panic there.
315 */
316 h->init_platform = x86_init_noop;
317 h->x2apic_available = bool_x86_init_noop;
318 h->init_mem_mapping = x86_init_noop;
319 h->init_after_bootmem = x86_init_noop;
320 h->guest_late_init = xen_hvm_guest_late_init;
321 x86_hyper_xen_hvm.runtime.pin_vcpu = x86_op_int_noop;
322 }
323 return xen_domain;
324 }
325
326 struct hypervisor_x86 x86_hyper_xen_hvm __initdata = {
327 .name = "Xen HVM",
328 .detect = xen_platform_hvm,
329 .type = X86_HYPER_XEN_HVM,
330 .init.init_platform = xen_hvm_guest_init,
331 .init.x2apic_available = xen_x2apic_available,
332 .init.init_mem_mapping = xen_hvm_init_mem_mapping,
333 .init.guest_late_init = xen_hvm_guest_late_init,
334 .init.msi_ext_dest_id = msi_ext_dest_id,
335 .runtime.pin_vcpu = xen_pin_vcpu,
336 .ignore_nopv = true,
337 };
338
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.