xref: /openbmc/linux/arch/x86/xen/enlighten_hvm.c (revision aac28965)
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/apic.h>
13 #include <asm/cpu.h>
14 #include <asm/smp.h>
15 #include <asm/io_apic.h>
16 #include <asm/reboot.h>
17 #include <asm/setup.h>
18 #include <asm/idtentry.h>
19 #include <asm/hypervisor.h>
20 #include <asm/e820/api.h>
21 #include <asm/early_ioremap.h>
22 
23 #include <asm/xen/cpuid.h>
24 #include <asm/xen/hypervisor.h>
25 #include <asm/xen/page.h>
26 
27 #include "xen-ops.h"
28 #include "mmu.h"
29 #include "smp.h"
30 
31 static unsigned long shared_info_pfn;
32 
33 void xen_hvm_init_shared_info(void)
34 {
35 	struct xen_add_to_physmap xatp;
36 
37 	xatp.domid = DOMID_SELF;
38 	xatp.idx = 0;
39 	xatp.space = XENMAPSPACE_shared_info;
40 	xatp.gpfn = shared_info_pfn;
41 	if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
42 		BUG();
43 }
44 
45 static void __init reserve_shared_info(void)
46 {
47 	u64 pa;
48 
49 	/*
50 	 * Search for a free page starting at 4kB physical address.
51 	 * Low memory is preferred to avoid an EPT large page split up
52 	 * by the mapping.
53 	 * Starting below X86_RESERVE_LOW (usually 64kB) is fine as
54 	 * the BIOS used for HVM guests is well behaved and won't
55 	 * clobber memory other than the first 4kB.
56 	 */
57 	for (pa = PAGE_SIZE;
58 	     !e820__mapped_all(pa, pa + PAGE_SIZE, E820_TYPE_RAM) ||
59 	     memblock_is_reserved(pa);
60 	     pa += PAGE_SIZE)
61 		;
62 
63 	shared_info_pfn = PHYS_PFN(pa);
64 
65 	memblock_reserve(pa, PAGE_SIZE);
66 	HYPERVISOR_shared_info = early_memremap(pa, PAGE_SIZE);
67 }
68 
69 static void __init xen_hvm_init_mem_mapping(void)
70 {
71 	early_memunmap(HYPERVISOR_shared_info, PAGE_SIZE);
72 	HYPERVISOR_shared_info = __va(PFN_PHYS(shared_info_pfn));
73 
74 	/*
75 	 * The virtual address of the shared_info page has changed, so
76 	 * the vcpu_info pointer for VCPU 0 is now stale.
77 	 *
78 	 * The prepare_boot_cpu callback will re-initialize it via
79 	 * xen_vcpu_setup, but we can't rely on that to be called for
80 	 * old Xen versions (xen_have_vector_callback == 0).
81 	 *
82 	 * It is, in any case, bad to have a stale vcpu_info pointer
83 	 * so reset it now.
84 	 */
85 	xen_vcpu_info_reset(0);
86 }
87 
88 static void __init init_hvm_pv_info(void)
89 {
90 	int major, minor;
91 	uint32_t eax, ebx, ecx, edx, base;
92 
93 	base = xen_cpuid_base();
94 	eax = cpuid_eax(base + 1);
95 
96 	major = eax >> 16;
97 	minor = eax & 0xffff;
98 	printk(KERN_INFO "Xen version %d.%d.\n", major, minor);
99 
100 	xen_domain_type = XEN_HVM_DOMAIN;
101 
102 	/* PVH set up hypercall page in xen_prepare_pvh(). */
103 	if (xen_pvh_domain())
104 		pv_info.name = "Xen PVH";
105 	else {
106 		u64 pfn;
107 		uint32_t msr;
108 
109 		pv_info.name = "Xen HVM";
110 		msr = cpuid_ebx(base + 2);
111 		pfn = __pa(hypercall_page);
112 		wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
113 	}
114 
115 	xen_setup_features();
116 
117 	cpuid(base + 4, &eax, &ebx, &ecx, &edx);
118 	if (eax & XEN_HVM_CPUID_VCPU_ID_PRESENT)
119 		this_cpu_write(xen_vcpu_id, ebx);
120 	else
121 		this_cpu_write(xen_vcpu_id, smp_processor_id());
122 }
123 
124 DEFINE_IDTENTRY_SYSVEC(sysvec_xen_hvm_callback)
125 {
126 	struct pt_regs *old_regs = set_irq_regs(regs);
127 
128 	inc_irq_stat(irq_hv_callback_count);
129 
130 	xen_hvm_evtchn_do_upcall();
131 
132 	set_irq_regs(old_regs);
133 }
134 
135 #ifdef CONFIG_KEXEC_CORE
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 
143 static void xen_hvm_crash_shutdown(struct pt_regs *regs)
144 {
145 	native_machine_crash_shutdown(regs);
146 	xen_reboot(SHUTDOWN_soft_reset);
147 }
148 #endif
149 
150 static int xen_cpu_up_prepare_hvm(unsigned int cpu)
151 {
152 	int rc = 0;
153 
154 	/*
155 	 * This can happen if CPU was offlined earlier and
156 	 * offlining timed out in common_cpu_die().
157 	 */
158 	if (cpu_report_state(cpu) == CPU_DEAD_FROZEN) {
159 		xen_smp_intr_free(cpu);
160 		xen_uninit_lock_cpu(cpu);
161 	}
162 
163 	if (cpu_acpi_id(cpu) != U32_MAX)
164 		per_cpu(xen_vcpu_id, cpu) = cpu_acpi_id(cpu);
165 	else
166 		per_cpu(xen_vcpu_id, cpu) = cpu;
167 	xen_vcpu_setup(cpu);
168 	if (!xen_have_vector_callback)
169 		return 0;
170 
171 	if (xen_feature(XENFEAT_hvm_safe_pvclock))
172 		xen_setup_timer(cpu);
173 
174 	rc = xen_smp_intr_init(cpu);
175 	if (rc) {
176 		WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
177 		     cpu, rc);
178 	}
179 	return rc;
180 }
181 
182 static int xen_cpu_dead_hvm(unsigned int cpu)
183 {
184 	xen_smp_intr_free(cpu);
185 
186 	if (xen_have_vector_callback && xen_feature(XENFEAT_hvm_safe_pvclock))
187 		xen_teardown_timer(cpu);
188 	return 0;
189 }
190 
191 static bool no_vector_callback __initdata;
192 
193 static void __init xen_hvm_guest_init(void)
194 {
195 	if (xen_pv_domain())
196 		return;
197 
198 	xen_set_restricted_virtio_memory_access();
199 
200 	init_hvm_pv_info();
201 
202 	reserve_shared_info();
203 	xen_hvm_init_shared_info();
204 
205 	/*
206 	 * xen_vcpu is a pointer to the vcpu_info struct in the shared_info
207 	 * page, we use it in the event channel upcall and in some pvclock
208 	 * related functions.
209 	 */
210 	xen_vcpu_info_reset(0);
211 
212 	xen_panic_handler_init();
213 
214 	if (!no_vector_callback && xen_feature(XENFEAT_hvm_callback_vector))
215 		xen_have_vector_callback = 1;
216 
217 	xen_hvm_smp_init();
218 	WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_hvm, xen_cpu_dead_hvm));
219 	xen_unplug_emulated_devices();
220 	x86_init.irqs.intr_init = xen_init_IRQ;
221 	xen_hvm_init_time_ops();
222 	xen_hvm_init_mmu_ops();
223 
224 #ifdef CONFIG_KEXEC_CORE
225 	machine_ops.shutdown = xen_hvm_shutdown;
226 	machine_ops.crash_shutdown = xen_hvm_crash_shutdown;
227 #endif
228 }
229 
230 static __init int xen_parse_nopv(char *arg)
231 {
232 	pr_notice("\"xen_nopv\" is deprecated, please use \"nopv\" instead\n");
233 
234 	if (xen_cpuid_base())
235 		nopv = true;
236 	return 0;
237 }
238 early_param("xen_nopv", xen_parse_nopv);
239 
240 static __init int xen_parse_no_vector_callback(char *arg)
241 {
242 	no_vector_callback = true;
243 	return 0;
244 }
245 early_param("xen_no_vector_callback", xen_parse_no_vector_callback);
246 
247 static __init bool xen_x2apic_available(void)
248 {
249 	return x2apic_supported();
250 }
251 
252 static bool __init msi_ext_dest_id(void)
253 {
254        return cpuid_eax(xen_cpuid_base() + 4) & XEN_HVM_CPUID_EXT_DEST_ID;
255 }
256 
257 static __init void xen_hvm_guest_late_init(void)
258 {
259 #ifdef CONFIG_XEN_PVH
260 	/* Test for PVH domain (PVH boot path taken overrides ACPI flags). */
261 	if (!xen_pvh &&
262 	    (x86_platform.legacy.rtc || !x86_platform.legacy.no_vga))
263 		return;
264 
265 	/* PVH detected. */
266 	xen_pvh = true;
267 
268 	if (nopv)
269 		panic("\"nopv\" and \"xen_nopv\" parameters are unsupported in PVH guest.");
270 
271 	/* Make sure we don't fall back to (default) ACPI_IRQ_MODEL_PIC. */
272 	if (!nr_ioapics && acpi_irq_model == ACPI_IRQ_MODEL_PIC)
273 		acpi_irq_model = ACPI_IRQ_MODEL_PLATFORM;
274 
275 	machine_ops.emergency_restart = xen_emergency_restart;
276 	pv_info.name = "Xen PVH";
277 #endif
278 }
279 
280 static uint32_t __init xen_platform_hvm(void)
281 {
282 	uint32_t xen_domain = xen_cpuid_base();
283 	struct x86_hyper_init *h = &x86_hyper_xen_hvm.init;
284 
285 	if (xen_pv_domain())
286 		return 0;
287 
288 	if (xen_pvh_domain() && nopv) {
289 		/* Guest booting via the Xen-PVH boot entry goes here */
290 		pr_info("\"nopv\" parameter is ignored in PVH guest\n");
291 		nopv = false;
292 	} else if (nopv && xen_domain) {
293 		/*
294 		 * Guest booting via normal boot entry (like via grub2) goes
295 		 * here.
296 		 *
297 		 * Use interface functions for bare hardware if nopv,
298 		 * xen_hvm_guest_late_init is an exception as we need to
299 		 * detect PVH and panic there.
300 		 */
301 		h->init_platform = x86_init_noop;
302 		h->x2apic_available = bool_x86_init_noop;
303 		h->init_mem_mapping = x86_init_noop;
304 		h->init_after_bootmem = x86_init_noop;
305 		h->guest_late_init = xen_hvm_guest_late_init;
306 		x86_hyper_xen_hvm.runtime.pin_vcpu = x86_op_int_noop;
307 	}
308 	return xen_domain;
309 }
310 
311 struct hypervisor_x86 x86_hyper_xen_hvm __initdata = {
312 	.name                   = "Xen HVM",
313 	.detect                 = xen_platform_hvm,
314 	.type			= X86_HYPER_XEN_HVM,
315 	.init.init_platform     = xen_hvm_guest_init,
316 	.init.x2apic_available  = xen_x2apic_available,
317 	.init.init_mem_mapping	= xen_hvm_init_mem_mapping,
318 	.init.guest_late_init	= xen_hvm_guest_late_init,
319 	.init.msi_ext_dest_id   = msi_ext_dest_id,
320 	.runtime.pin_vcpu       = xen_pin_vcpu,
321 	.ignore_nopv            = true,
322 };
323