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
8 #include <xen/features.h>
9 #include <xen/events.h>
10 #include <xen/interface/memory.h>
11
12 #include <asm/cpu.h>
13 #include <asm/smp.h>
14 #include <asm/io_apic.h>
15 #include <asm/reboot.h>
16 #include <asm/setup.h>
17 #include <asm/idtentry.h>
18 #include <asm/hypervisor.h>
19 #include <asm/e820/api.h>
20 #include <asm/early_ioremap.h>
21
22 #include <asm/xen/cpuid.h>
23 #include <asm/xen/hypervisor.h>
24 #include <asm/xen/page.h>
25
26 #include "xen-ops.h"
27 #include "mmu.h"
28 #include "smp.h"
29
30 static unsigned long shared_info_pfn;
31
xen_hvm_init_shared_info(void)32 void xen_hvm_init_shared_info(void)
33 {
34 struct xen_add_to_physmap xatp;
35
36 xatp.domid = DOMID_SELF;
37 xatp.idx = 0;
38 xatp.space = XENMAPSPACE_shared_info;
39 xatp.gpfn = shared_info_pfn;
40 if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
41 BUG();
42 }
43
reserve_shared_info(void)44 static void __init reserve_shared_info(void)
45 {
46 u64 pa;
47
48 /*
49 * Search for a free page starting at 4kB physical address.
50 * Low memory is preferred to avoid an EPT large page split up
51 * by the mapping.
52 * Starting below X86_RESERVE_LOW (usually 64kB) is fine as
53 * the BIOS used for HVM guests is well behaved and won't
54 * clobber memory other than the first 4kB.
55 */
56 for (pa = PAGE_SIZE;
57 !e820__mapped_all(pa, pa + PAGE_SIZE, E820_TYPE_RAM) ||
58 memblock_is_reserved(pa);
59 pa += PAGE_SIZE)
60 ;
61
62 shared_info_pfn = PHYS_PFN(pa);
63
64 memblock_reserve(pa, PAGE_SIZE);
65 HYPERVISOR_shared_info = early_memremap(pa, PAGE_SIZE);
66 }
67
xen_hvm_init_mem_mapping(void)68 static void __init xen_hvm_init_mem_mapping(void)
69 {
70 early_memunmap(HYPERVISOR_shared_info, PAGE_SIZE);
71 HYPERVISOR_shared_info = __va(PFN_PHYS(shared_info_pfn));
72
73 /*
74 * The virtual address of the shared_info page has changed, so
75 * the vcpu_info pointer for VCPU 0 is now stale.
76 *
77 * The prepare_boot_cpu callback will re-initialize it via
78 * xen_vcpu_setup, but we can't rely on that to be called for
79 * old Xen versions (xen_have_vector_callback == 0).
80 *
81 * It is, in any case, bad to have a stale vcpu_info pointer
82 * so reset it now.
83 */
84 xen_vcpu_info_reset(0);
85 }
86
init_hvm_pv_info(void)87 static void __init init_hvm_pv_info(void)
88 {
89 int major, minor;
90 uint32_t eax, ebx, ecx, edx, base;
91
92 base = xen_cpuid_base();
93 eax = cpuid_eax(base + 1);
94
95 major = eax >> 16;
96 minor = eax & 0xffff;
97 printk(KERN_INFO "Xen version %d.%d.\n", major, minor);
98
99 xen_domain_type = XEN_HVM_DOMAIN;
100
101 /* PVH set up hypercall page in xen_prepare_pvh(). */
102 if (xen_pvh_domain())
103 pv_info.name = "Xen PVH";
104 else {
105 u64 pfn;
106 uint32_t msr;
107
108 pv_info.name = "Xen HVM";
109 msr = cpuid_ebx(base + 2);
110 pfn = __pa(hypercall_page);
111 wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
112 }
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 inc_irq_stat(irq_hv_callback_count);
128
129 xen_hvm_evtchn_do_upcall();
130
131 set_irq_regs(old_regs);
132 }
133
134 #ifdef CONFIG_KEXEC_CORE
xen_hvm_shutdown(void)135 static void xen_hvm_shutdown(void)
136 {
137 native_machine_shutdown();
138 if (kexec_in_progress)
139 xen_reboot(SHUTDOWN_soft_reset);
140 }
141
xen_hvm_crash_shutdown(struct pt_regs * regs)142 static void xen_hvm_crash_shutdown(struct pt_regs *regs)
143 {
144 native_machine_crash_shutdown(regs);
145 xen_reboot(SHUTDOWN_soft_reset);
146 }
147 #endif
148
xen_cpu_up_prepare_hvm(unsigned int cpu)149 static int xen_cpu_up_prepare_hvm(unsigned int cpu)
150 {
151 int rc = 0;
152
153 /*
154 * This can happen if CPU was offlined earlier and
155 * offlining timed out in common_cpu_die().
156 */
157 if (cpu_report_state(cpu) == CPU_DEAD_FROZEN) {
158 xen_smp_intr_free(cpu);
159 xen_uninit_lock_cpu(cpu);
160 }
161
162 if (cpu_acpi_id(cpu) != U32_MAX)
163 per_cpu(xen_vcpu_id, cpu) = cpu_acpi_id(cpu);
164 else
165 per_cpu(xen_vcpu_id, cpu) = cpu;
166 rc = xen_vcpu_setup(cpu);
167 if (rc || !xen_have_vector_callback)
168 return rc;
169
170 if (xen_feature(XENFEAT_hvm_safe_pvclock))
171 xen_setup_timer(cpu);
172
173 rc = xen_smp_intr_init(cpu);
174 if (rc) {
175 WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
176 cpu, rc);
177 }
178 return rc;
179 }
180
xen_cpu_dead_hvm(unsigned int cpu)181 static int xen_cpu_dead_hvm(unsigned int cpu)
182 {
183 xen_smp_intr_free(cpu);
184
185 if (xen_have_vector_callback && xen_feature(XENFEAT_hvm_safe_pvclock))
186 xen_teardown_timer(cpu);
187
188 return 0;
189 }
190
191 static bool no_vector_callback __initdata;
192
xen_hvm_guest_init(void)193 static void __init xen_hvm_guest_init(void)
194 {
195 if (xen_pv_domain())
196 return;
197
198 init_hvm_pv_info();
199
200 reserve_shared_info();
201 xen_hvm_init_shared_info();
202
203 /*
204 * xen_vcpu is a pointer to the vcpu_info struct in the shared_info
205 * page, we use it in the event channel upcall and in some pvclock
206 * related functions.
207 */
208 xen_vcpu_info_reset(0);
209
210 xen_panic_handler_init();
211
212 if (!no_vector_callback && xen_feature(XENFEAT_hvm_callback_vector))
213 xen_have_vector_callback = 1;
214
215 xen_hvm_smp_init();
216 WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_hvm, xen_cpu_dead_hvm));
217 xen_unplug_emulated_devices();
218 x86_init.irqs.intr_init = xen_init_IRQ;
219 xen_hvm_init_time_ops();
220 xen_hvm_init_mmu_ops();
221
222 #ifdef CONFIG_KEXEC_CORE
223 machine_ops.shutdown = xen_hvm_shutdown;
224 machine_ops.crash_shutdown = xen_hvm_crash_shutdown;
225 #endif
226 }
227
xen_parse_nopv(char * arg)228 static __init int xen_parse_nopv(char *arg)
229 {
230 pr_notice("\"xen_nopv\" is deprecated, please use \"nopv\" instead\n");
231
232 if (xen_cpuid_base())
233 nopv = true;
234 return 0;
235 }
236 early_param("xen_nopv", xen_parse_nopv);
237
xen_parse_no_vector_callback(char * arg)238 static __init int xen_parse_no_vector_callback(char *arg)
239 {
240 no_vector_callback = true;
241 return 0;
242 }
243 early_param("xen_no_vector_callback", xen_parse_no_vector_callback);
244
xen_hvm_need_lapic(void)245 bool __init xen_hvm_need_lapic(void)
246 {
247 if (xen_pv_domain())
248 return false;
249 if (!xen_hvm_domain())
250 return false;
251 if (xen_feature(XENFEAT_hvm_pirqs) && xen_have_vector_callback)
252 return false;
253 return true;
254 }
255
xen_hvm_guest_late_init(void)256 static __init void xen_hvm_guest_late_init(void)
257 {
258 #ifdef CONFIG_XEN_PVH
259 /* Test for PVH domain (PVH boot path taken overrides ACPI flags). */
260 if (!xen_pvh &&
261 (x86_platform.legacy.rtc || !x86_platform.legacy.no_vga))
262 return;
263
264 /* PVH detected. */
265 xen_pvh = true;
266
267 if (nopv)
268 panic("\"nopv\" and \"xen_nopv\" parameters are unsupported in PVH guest.");
269
270 /* Make sure we don't fall back to (default) ACPI_IRQ_MODEL_PIC. */
271 if (!nr_ioapics && acpi_irq_model == ACPI_IRQ_MODEL_PIC)
272 acpi_irq_model = ACPI_IRQ_MODEL_PLATFORM;
273
274 machine_ops.emergency_restart = xen_emergency_restart;
275 pv_info.name = "Xen PVH";
276 #endif
277 }
278
xen_platform_hvm(void)279 static uint32_t __init xen_platform_hvm(void)
280 {
281 uint32_t xen_domain = xen_cpuid_base();
282 struct x86_hyper_init *h = &x86_hyper_xen_hvm.init;
283
284 if (xen_pv_domain())
285 return 0;
286
287 if (xen_pvh_domain() && nopv) {
288 /* Guest booting via the Xen-PVH boot entry goes here */
289 pr_info("\"nopv\" parameter is ignored in PVH guest\n");
290 nopv = false;
291 } else if (nopv && xen_domain) {
292 /*
293 * Guest booting via normal boot entry (like via grub2) goes
294 * here.
295 *
296 * Use interface functions for bare hardware if nopv,
297 * xen_hvm_guest_late_init is an exception as we need to
298 * detect PVH and panic there.
299 */
300 h->init_platform = x86_init_noop;
301 h->x2apic_available = bool_x86_init_noop;
302 h->init_mem_mapping = x86_init_noop;
303 h->init_after_bootmem = x86_init_noop;
304 h->guest_late_init = xen_hvm_guest_late_init;
305 x86_hyper_xen_hvm.runtime.pin_vcpu = x86_op_int_noop;
306 }
307 return xen_domain;
308 }
309
310 struct hypervisor_x86 x86_hyper_xen_hvm __initdata = {
311 .name = "Xen HVM",
312 .detect = xen_platform_hvm,
313 .type = X86_HYPER_XEN_HVM,
314 .init.init_platform = xen_hvm_guest_init,
315 .init.x2apic_available = xen_x2apic_para_available,
316 .init.init_mem_mapping = xen_hvm_init_mem_mapping,
317 .init.guest_late_init = xen_hvm_guest_late_init,
318 .runtime.pin_vcpu = xen_pin_vcpu,
319 .ignore_nopv = true,
320 };
321