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
34static unsigned long shared_info_pfn;
35
36__ro_after_init bool xen_percpu_upcall;
37EXPORT_SYMBOL_GPL(xen_percpu_upcall);
38
39void 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, arg: &xatp))
48 BUG();
49}
50
51static 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(start: pa, end: pa + PAGE_SIZE, type: E820_TYPE_RAM) ||
65 memblock_is_reserved(addr: pa);
66 pa += PAGE_SIZE)
67 ;
68
69 shared_info_pfn = PHYS_PFN(pa);
70
71 memblock_reserve(base: pa, PAGE_SIZE);
72 HYPERVISOR_shared_info = early_memremap(phys_addr: pa, PAGE_SIZE);
73}
74
75static void __init xen_hvm_init_mem_mapping(void)
76{
77 early_memunmap(addr: 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(cpu: 0);
92}
93
94static 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(op: 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 u64 pfn;
113 uint32_t msr;
114
115 pv_info.name = "Xen HVM";
116 msr = cpuid_ebx(op: base + 2);
117 pfn = __pa(hypercall_page);
118 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
119 }
120
121 xen_setup_features();
122
123 cpuid(op: base + 4, eax: &eax, ebx: &ebx, ecx: &ecx, edx: &edx);
124 if (eax & XEN_HVM_CPUID_VCPU_ID_PRESENT)
125 this_cpu_write(xen_vcpu_id, ebx);
126 else
127 this_cpu_write(xen_vcpu_id, smp_processor_id());
128}
129
130DEFINE_IDTENTRY_SYSVEC(sysvec_xen_hvm_callback)
131{
132 struct pt_regs *old_regs = set_irq_regs(regs);
133
134 if (xen_percpu_upcall)
135 apic_eoi();
136
137 inc_irq_stat(irq_hv_callback_count);
138
139 xen_evtchn_do_upcall();
140
141 set_irq_regs(old_regs);
142}
143
144#ifdef CONFIG_KEXEC_CORE
145static void xen_hvm_shutdown(void)
146{
147 native_machine_shutdown();
148 if (kexec_in_progress)
149 xen_reboot(SHUTDOWN_soft_reset);
150}
151#endif
152
153#ifdef CONFIG_CRASH_DUMP
154static void xen_hvm_crash_shutdown(struct pt_regs *regs)
155{
156 native_machine_crash_shutdown(regs);
157 xen_reboot(SHUTDOWN_soft_reset);
158}
159#endif
160
161static int xen_cpu_up_prepare_hvm(unsigned int cpu)
162{
163 int rc = 0;
164
165 /*
166 * If a CPU was offlined earlier and offlining timed out then the
167 * lock mechanism is still initialized. Uninit it unconditionally
168 * as it's safe to call even if already uninited. Interrupts and
169 * timer have already been handled in xen_cpu_dead_hvm().
170 */
171 xen_uninit_lock_cpu(cpu);
172
173 if (cpu_acpi_id(cpu) != CPU_ACPIID_INVALID)
174 per_cpu(xen_vcpu_id, cpu) = cpu_acpi_id(cpu);
175 else
176 per_cpu(xen_vcpu_id, cpu) = cpu;
177 xen_vcpu_setup(cpu);
178 if (!xen_have_vector_callback)
179 return 0;
180
181 if (xen_percpu_upcall) {
182 rc = xen_set_upcall_vector(cpu);
183 if (rc) {
184 WARN(1, "HVMOP_set_evtchn_upcall_vector"
185 " for CPU %d failed: %d\n", cpu, rc);
186 return rc;
187 }
188 }
189
190 if (xen_feature(XENFEAT_hvm_safe_pvclock))
191 xen_setup_timer(cpu);
192
193 rc = xen_smp_intr_init(cpu);
194 if (rc) {
195 WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
196 cpu, rc);
197 }
198 return rc;
199}
200
201static int xen_cpu_dead_hvm(unsigned int cpu)
202{
203 xen_smp_intr_free(cpu);
204
205 if (xen_have_vector_callback && xen_feature(XENFEAT_hvm_safe_pvclock))
206 xen_teardown_timer(cpu);
207 return 0;
208}
209
210static void __init xen_hvm_guest_init(void)
211{
212 if (xen_pv_domain())
213 return;
214
215 if (IS_ENABLED(CONFIG_XEN_VIRTIO_FORCE_GRANT))
216 virtio_set_mem_acc_cb(func: xen_virtio_restricted_mem_acc);
217
218 init_hvm_pv_info();
219
220 reserve_shared_info();
221 xen_hvm_init_shared_info();
222
223 /*
224 * xen_vcpu is a pointer to the vcpu_info struct in the shared_info
225 * page, we use it in the event channel upcall and in some pvclock
226 * related functions.
227 */
228 xen_vcpu_info_reset(cpu: 0);
229
230 xen_panic_handler_init();
231
232 xen_hvm_smp_init();
233 WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_hvm, xen_cpu_dead_hvm));
234 xen_unplug_emulated_devices();
235 x86_init.irqs.intr_init = xen_init_IRQ;
236 xen_hvm_init_time_ops();
237 xen_hvm_init_mmu_ops();
238
239#ifdef CONFIG_KEXEC_CORE
240 machine_ops.shutdown = xen_hvm_shutdown;
241#endif
242#ifdef CONFIG_CRASH_DUMP
243 machine_ops.crash_shutdown = xen_hvm_crash_shutdown;
244#endif
245}
246
247static __init int xen_parse_nopv(char *arg)
248{
249 pr_notice("\"xen_nopv\" is deprecated, please use \"nopv\" instead\n");
250
251 if (xen_cpuid_base())
252 nopv = true;
253 return 0;
254}
255early_param("xen_nopv", xen_parse_nopv);
256
257static __init int xen_parse_no_vector_callback(char *arg)
258{
259 xen_have_vector_callback = false;
260 return 0;
261}
262early_param("xen_no_vector_callback", xen_parse_no_vector_callback);
263
264static __init bool xen_x2apic_available(void)
265{
266 return x2apic_supported();
267}
268
269static bool __init msi_ext_dest_id(void)
270{
271 return cpuid_eax(op: xen_cpuid_base() + 4) & XEN_HVM_CPUID_EXT_DEST_ID;
272}
273
274static __init void xen_hvm_guest_late_init(void)
275{
276#ifdef CONFIG_XEN_PVH
277 /* Test for PVH domain (PVH boot path taken overrides ACPI flags). */
278 if (!xen_pvh &&
279 (x86_platform.legacy.rtc || !x86_platform.legacy.no_vga))
280 return;
281
282 /* PVH detected. */
283 xen_pvh = true;
284
285 if (nopv)
286 panic(fmt: "\"nopv\" and \"xen_nopv\" parameters are unsupported in PVH guest.");
287
288 /* Make sure we don't fall back to (default) ACPI_IRQ_MODEL_PIC. */
289 if (!nr_ioapics && acpi_irq_model == ACPI_IRQ_MODEL_PIC)
290 acpi_irq_model = ACPI_IRQ_MODEL_PLATFORM;
291
292 machine_ops.emergency_restart = xen_emergency_restart;
293 pv_info.name = "Xen PVH";
294#endif
295}
296
297static uint32_t __init xen_platform_hvm(void)
298{
299 uint32_t xen_domain = xen_cpuid_base();
300 struct x86_hyper_init *h = &x86_hyper_xen_hvm.init;
301
302 if (xen_pv_domain())
303 return 0;
304
305 if (xen_pvh_domain() && nopv) {
306 /* Guest booting via the Xen-PVH boot entry goes here */
307 pr_info("\"nopv\" parameter is ignored in PVH guest\n");
308 nopv = false;
309 } else if (nopv && xen_domain) {
310 /*
311 * Guest booting via normal boot entry (like via grub2) goes
312 * here.
313 *
314 * Use interface functions for bare hardware if nopv,
315 * xen_hvm_guest_late_init is an exception as we need to
316 * detect PVH and panic there.
317 */
318 h->init_platform = x86_init_noop;
319 h->x2apic_available = bool_x86_init_noop;
320 h->init_mem_mapping = x86_init_noop;
321 h->init_after_bootmem = x86_init_noop;
322 h->guest_late_init = xen_hvm_guest_late_init;
323 x86_hyper_xen_hvm.runtime.pin_vcpu = x86_op_int_noop;
324 }
325 return xen_domain;
326}
327
328struct hypervisor_x86 x86_hyper_xen_hvm __initdata = {
329 .name = "Xen HVM",
330 .detect = xen_platform_hvm,
331 .type = X86_HYPER_XEN_HVM,
332 .init.init_platform = xen_hvm_guest_init,
333 .init.x2apic_available = xen_x2apic_available,
334 .init.init_mem_mapping = xen_hvm_init_mem_mapping,
335 .init.guest_late_init = xen_hvm_guest_late_init,
336 .init.msi_ext_dest_id = msi_ext_dest_id,
337 .runtime.pin_vcpu = xen_pin_vcpu,
338 .ignore_nopv = true,
339};
340

source code of linux/arch/x86/xen/enlighten_hvm.c