xref: /openbmc/linux/arch/x86/xen/enlighten_hvm.c (revision 748008e1)
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/interface/memory.h>
12 
13 #include <asm/apic.h>
14 #include <asm/cpu.h>
15 #include <asm/smp.h>
16 #include <asm/io_apic.h>
17 #include <asm/reboot.h>
18 #include <asm/setup.h>
19 #include <asm/idtentry.h>
20 #include <asm/hypervisor.h>
21 #include <asm/e820/api.h>
22 #include <asm/early_ioremap.h>
23 
24 #include <asm/xen/cpuid.h>
25 #include <asm/xen/hypervisor.h>
26 #include <asm/xen/page.h>
27 
28 #include "xen-ops.h"
29 #include "mmu.h"
30 #include "smp.h"
31 
32 static unsigned long shared_info_pfn;
33 
34 void xen_hvm_init_shared_info(void)
35 {
36 	struct xen_add_to_physmap xatp;
37 
38 	xatp.domid = DOMID_SELF;
39 	xatp.idx = 0;
40 	xatp.space = XENMAPSPACE_shared_info;
41 	xatp.gpfn = shared_info_pfn;
42 	if (HYPERVISOR_memory_op(XENMEM_add_to_physmap, &xatp))
43 		BUG();
44 }
45 
46 static void __init reserve_shared_info(void)
47 {
48 	u64 pa;
49 
50 	/*
51 	 * Search for a free page starting at 4kB physical address.
52 	 * Low memory is preferred to avoid an EPT large page split up
53 	 * by the mapping.
54 	 * Starting below X86_RESERVE_LOW (usually 64kB) is fine as
55 	 * the BIOS used for HVM guests is well behaved and won't
56 	 * clobber memory other than the first 4kB.
57 	 */
58 	for (pa = PAGE_SIZE;
59 	     !e820__mapped_all(pa, pa + PAGE_SIZE, E820_TYPE_RAM) ||
60 	     memblock_is_reserved(pa);
61 	     pa += PAGE_SIZE)
62 		;
63 
64 	shared_info_pfn = PHYS_PFN(pa);
65 
66 	memblock_reserve(pa, PAGE_SIZE);
67 	HYPERVISOR_shared_info = early_memremap(pa, PAGE_SIZE);
68 }
69 
70 static void __init xen_hvm_init_mem_mapping(void)
71 {
72 	early_memunmap(HYPERVISOR_shared_info, PAGE_SIZE);
73 	HYPERVISOR_shared_info = __va(PFN_PHYS(shared_info_pfn));
74 
75 	/*
76 	 * The virtual address of the shared_info page has changed, so
77 	 * the vcpu_info pointer for VCPU 0 is now stale.
78 	 *
79 	 * The prepare_boot_cpu callback will re-initialize it via
80 	 * xen_vcpu_setup, but we can't rely on that to be called for
81 	 * old Xen versions (xen_have_vector_callback == 0).
82 	 *
83 	 * It is, in any case, bad to have a stale vcpu_info pointer
84 	 * so reset it now.
85 	 */
86 	xen_vcpu_info_reset(0);
87 }
88 
89 static void __init init_hvm_pv_info(void)
90 {
91 	int major, minor;
92 	uint32_t eax, ebx, ecx, edx, base;
93 
94 	base = xen_cpuid_base();
95 	eax = cpuid_eax(base + 1);
96 
97 	major = eax >> 16;
98 	minor = eax & 0xffff;
99 	printk(KERN_INFO "Xen version %d.%d.\n", major, minor);
100 
101 	xen_domain_type = XEN_HVM_DOMAIN;
102 
103 	/* PVH set up hypercall page in xen_prepare_pvh(). */
104 	if (xen_pvh_domain())
105 		pv_info.name = "Xen PVH";
106 	else {
107 		u64 pfn;
108 		uint32_t msr;
109 
110 		pv_info.name = "Xen HVM";
111 		msr = cpuid_ebx(base + 2);
112 		pfn = __pa(hypercall_page);
113 		wrmsr_safe(msr, (u32)pfn, (u32)(pfn >> 32));
114 	}
115 
116 	xen_setup_features();
117 
118 	cpuid(base + 4, &eax, &ebx, &ecx, &edx);
119 	if (eax & XEN_HVM_CPUID_VCPU_ID_PRESENT)
120 		this_cpu_write(xen_vcpu_id, ebx);
121 	else
122 		this_cpu_write(xen_vcpu_id, smp_processor_id());
123 }
124 
125 DEFINE_IDTENTRY_SYSVEC(sysvec_xen_hvm_callback)
126 {
127 	struct pt_regs *old_regs = set_irq_regs(regs);
128 
129 	inc_irq_stat(irq_hv_callback_count);
130 
131 	xen_hvm_evtchn_do_upcall();
132 
133 	set_irq_regs(old_regs);
134 }
135 
136 #ifdef CONFIG_KEXEC_CORE
137 static void xen_hvm_shutdown(void)
138 {
139 	native_machine_shutdown();
140 	if (kexec_in_progress)
141 		xen_reboot(SHUTDOWN_soft_reset);
142 }
143 
144 static void xen_hvm_crash_shutdown(struct pt_regs *regs)
145 {
146 	native_machine_crash_shutdown(regs);
147 	xen_reboot(SHUTDOWN_soft_reset);
148 }
149 #endif
150 
151 static int xen_cpu_up_prepare_hvm(unsigned int cpu)
152 {
153 	int rc = 0;
154 
155 	/*
156 	 * This can happen if CPU was offlined earlier and
157 	 * offlining timed out in common_cpu_die().
158 	 */
159 	if (cpu_report_state(cpu) == CPU_DEAD_FROZEN) {
160 		xen_smp_intr_free(cpu);
161 		xen_uninit_lock_cpu(cpu);
162 	}
163 
164 	if (cpu_acpi_id(cpu) != U32_MAX)
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_feature(XENFEAT_hvm_safe_pvclock))
173 		xen_setup_timer(cpu);
174 
175 	rc = xen_smp_intr_init(cpu);
176 	if (rc) {
177 		WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
178 		     cpu, rc);
179 	}
180 	return rc;
181 }
182 
183 static int xen_cpu_dead_hvm(unsigned int cpu)
184 {
185 	xen_smp_intr_free(cpu);
186 
187 	if (xen_have_vector_callback && xen_feature(XENFEAT_hvm_safe_pvclock))
188 		xen_teardown_timer(cpu);
189 	return 0;
190 }
191 
192 static bool no_vector_callback __initdata;
193 
194 static void __init xen_hvm_guest_init(void)
195 {
196 	if (xen_pv_domain())
197 		return;
198 
199 	if (IS_ENABLED(CONFIG_XEN_VIRTIO_FORCE_GRANT))
200 		virtio_set_mem_acc_cb(virtio_require_restricted_mem_acc);
201 
202 	init_hvm_pv_info();
203 
204 	reserve_shared_info();
205 	xen_hvm_init_shared_info();
206 
207 	/*
208 	 * xen_vcpu is a pointer to the vcpu_info struct in the shared_info
209 	 * page, we use it in the event channel upcall and in some pvclock
210 	 * related functions.
211 	 */
212 	xen_vcpu_info_reset(0);
213 
214 	xen_panic_handler_init();
215 
216 	if (!no_vector_callback && xen_feature(XENFEAT_hvm_callback_vector))
217 		xen_have_vector_callback = 1;
218 
219 	xen_hvm_smp_init();
220 	WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_hvm, xen_cpu_dead_hvm));
221 	xen_unplug_emulated_devices();
222 	x86_init.irqs.intr_init = xen_init_IRQ;
223 	xen_hvm_init_time_ops();
224 	xen_hvm_init_mmu_ops();
225 
226 #ifdef CONFIG_KEXEC_CORE
227 	machine_ops.shutdown = xen_hvm_shutdown;
228 	machine_ops.crash_shutdown = xen_hvm_crash_shutdown;
229 #endif
230 }
231 
232 static __init int xen_parse_nopv(char *arg)
233 {
234 	pr_notice("\"xen_nopv\" is deprecated, please use \"nopv\" instead\n");
235 
236 	if (xen_cpuid_base())
237 		nopv = true;
238 	return 0;
239 }
240 early_param("xen_nopv", xen_parse_nopv);
241 
242 static __init int xen_parse_no_vector_callback(char *arg)
243 {
244 	no_vector_callback = true;
245 	return 0;
246 }
247 early_param("xen_no_vector_callback", xen_parse_no_vector_callback);
248 
249 static __init bool xen_x2apic_available(void)
250 {
251 	return x2apic_supported();
252 }
253 
254 static bool __init msi_ext_dest_id(void)
255 {
256        return cpuid_eax(xen_cpuid_base() + 4) & XEN_HVM_CPUID_EXT_DEST_ID;
257 }
258 
259 static __init void xen_hvm_guest_late_init(void)
260 {
261 #ifdef CONFIG_XEN_PVH
262 	/* Test for PVH domain (PVH boot path taken overrides ACPI flags). */
263 	if (!xen_pvh &&
264 	    (x86_platform.legacy.rtc || !x86_platform.legacy.no_vga))
265 		return;
266 
267 	/* PVH detected. */
268 	xen_pvh = true;
269 
270 	if (nopv)
271 		panic("\"nopv\" and \"xen_nopv\" parameters are unsupported in PVH guest.");
272 
273 	/* Make sure we don't fall back to (default) ACPI_IRQ_MODEL_PIC. */
274 	if (!nr_ioapics && acpi_irq_model == ACPI_IRQ_MODEL_PIC)
275 		acpi_irq_model = ACPI_IRQ_MODEL_PLATFORM;
276 
277 	machine_ops.emergency_restart = xen_emergency_restart;
278 	pv_info.name = "Xen PVH";
279 #endif
280 }
281 
282 static uint32_t __init xen_platform_hvm(void)
283 {
284 	uint32_t xen_domain = xen_cpuid_base();
285 	struct x86_hyper_init *h = &x86_hyper_xen_hvm.init;
286 
287 	if (xen_pv_domain())
288 		return 0;
289 
290 	if (xen_pvh_domain() && nopv) {
291 		/* Guest booting via the Xen-PVH boot entry goes here */
292 		pr_info("\"nopv\" parameter is ignored in PVH guest\n");
293 		nopv = false;
294 	} else if (nopv && xen_domain) {
295 		/*
296 		 * Guest booting via normal boot entry (like via grub2) goes
297 		 * here.
298 		 *
299 		 * Use interface functions for bare hardware if nopv,
300 		 * xen_hvm_guest_late_init is an exception as we need to
301 		 * detect PVH and panic there.
302 		 */
303 		h->init_platform = x86_init_noop;
304 		h->x2apic_available = bool_x86_init_noop;
305 		h->init_mem_mapping = x86_init_noop;
306 		h->init_after_bootmem = x86_init_noop;
307 		h->guest_late_init = xen_hvm_guest_late_init;
308 		x86_hyper_xen_hvm.runtime.pin_vcpu = x86_op_int_noop;
309 	}
310 	return xen_domain;
311 }
312 
313 struct hypervisor_x86 x86_hyper_xen_hvm __initdata = {
314 	.name                   = "Xen HVM",
315 	.detect                 = xen_platform_hvm,
316 	.type			= X86_HYPER_XEN_HVM,
317 	.init.init_platform     = xen_hvm_guest_init,
318 	.init.x2apic_available  = xen_x2apic_available,
319 	.init.init_mem_mapping	= xen_hvm_init_mem_mapping,
320 	.init.guest_late_init	= xen_hvm_guest_late_init,
321 	.init.msi_ext_dest_id   = msi_ext_dest_id,
322 	.runtime.pin_vcpu       = xen_pin_vcpu,
323 	.ignore_nopv            = true,
324 };
325