arch/x86/xen/enlighten_hvm.c

   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
  34 static unsigned long shared_info_pfn;
  35
  36 __ro_after_init bool xen_percpu_upcall;
  37 EXPORT_SYMBOL_GPL(xen_percpu_upcall);
  38
  39 void 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, &xatp))
  48                 BUG();
  49 }
  50
  51 static 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(pa, pa + PAGE_SIZE, E820_TYPE_RAM) ||
  65              memblock_is_reserved(pa);
  66              pa += PAGE_SIZE)
  67                 ;
  68
  69         shared_info_pfn = PHYS_PFN(pa);
  70
  71         memblock_reserve(pa, PAGE_SIZE);
  72         HYPERVISOR_shared_info = early_memremap(pa, PAGE_SIZE);
  73 }
  74
  75 static void __init xen_hvm_init_mem_mapping(void)
  76 {
  77         early_memunmap(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(0);
  92 }
  93
  94 static 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(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(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(base + 4, &eax, &ebx, &ecx, &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
 130 DEFINE_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_hvm_evtchn_do_upcall();
 140
 141         set_irq_regs(old_regs);
 142 }
 143
 144 #ifdef CONFIG_KEXEC_CORE
 145 static void xen_hvm_shutdown(void)
 146 {
 147         native_machine_shutdown();
 148         if (kexec_in_progress)
 149                 xen_reboot(SHUTDOWN_soft_reset);
 150 }
 151
 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) != U32_MAX)
 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         machine_ops.crash_shutdown = xen_hvm_crash_shutdown;
 240 #endif
 241 }
 242
 243 static __init int xen_parse_nopv(char *arg)
 244 {
 245         pr_notice("\"xen_nopv\" is deprecated, please use \"nopv\" instead\n");
 246
 247         if (xen_cpuid_base())
 248                 nopv = true;
 249         return 0;
 250 }
 251 early_param("xen_nopv", xen_parse_nopv);
 252
 253 static __init int xen_parse_no_vector_callback(char *arg)
 254 {
 255         xen_have_vector_callback = false;
 256         return 0;
 257 }
 258 early_param("xen_no_vector_callback", xen_parse_no_vector_callback);
 259
 260 static __init bool xen_x2apic_available(void)
 261 {
 262         return x2apic_supported();
 263 }
 264
 265 static bool __init msi_ext_dest_id(void)
 266 {
 267        return cpuid_eax(xen_cpuid_base() + 4) & XEN_HVM_CPUID_EXT_DEST_ID;
 268 }
 269
 270 static __init void xen_hvm_guest_late_init(void)
 271 {
 272 #ifdef CONFIG_XEN_PVH
 273         /* Test for PVH domain (PVH boot path taken overrides ACPI flags). */
 274         if (!xen_pvh &&
 275             (x86_platform.legacy.rtc || !x86_platform.legacy.no_vga))
 276                 return;
 277
 278         /* PVH detected. */
 279         xen_pvh = true;
 280
 281         if (nopv)
 282                 panic("\"nopv\" and \"xen_nopv\" parameters are unsupported in PVH guest.");
 283
 284         /* Make sure we don't fall back to (default) ACPI_IRQ_MODEL_PIC. */
 285         if (!nr_ioapics && acpi_irq_model == ACPI_IRQ_MODEL_PIC)
 286                 acpi_irq_model = ACPI_IRQ_MODEL_PLATFORM;
 287
 288         machine_ops.emergency_restart = xen_emergency_restart;
 289         pv_info.name = "Xen PVH";
 290 #endif
 291 }
 292
 293 static uint32_t __init xen_platform_hvm(void)
 294 {
 295         uint32_t xen_domain = xen_cpuid_base();
 296         struct x86_hyper_init *h = &x86_hyper_xen_hvm.init;
 297
 298         if (xen_pv_domain())
 299                 return 0;
 300
 301         if (xen_pvh_domain() && nopv) {
 302                 /* Guest booting via the Xen-PVH boot entry goes here */
 303                 pr_info("\"nopv\" parameter is ignored in PVH guest\n");
 304                 nopv = false;
 305         } else if (nopv && xen_domain) {
 306                 /*
 307                  * Guest booting via normal boot entry (like via grub2) goes
 308                  * here.
 309                  *
 310                  * Use interface functions for bare hardware if nopv,
 311                  * xen_hvm_guest_late_init is an exception as we need to
 312                  * detect PVH and panic there.
 313                  */
 314                 h->init_platform = x86_init_noop;
 315                 h->x2apic_available = bool_x86_init_noop;
 316                 h->init_mem_mapping = x86_init_noop;
 317                 h->init_after_bootmem = x86_init_noop;
 318                 h->guest_late_init = xen_hvm_guest_late_init;
 319                 x86_hyper_xen_hvm.runtime.pin_vcpu = x86_op_int_noop;
 320         }
 321         return xen_domain;
 322 }
 323
 324 struct hypervisor_x86 x86_hyper_xen_hvm __initdata = {
 325         .name                   = "Xen HVM",
 326         .detect                 = xen_platform_hvm,
 327         .type                   = X86_HYPER_XEN_HVM,
 328         .init.init_platform     = xen_hvm_guest_init,
 329         .init.x2apic_available  = xen_x2apic_available,
 330         .init.init_mem_mapping  = xen_hvm_init_mem_mapping,
 331         .init.guest_late_init   = xen_hvm_guest_late_init,
 332         .init.msi_ext_dest_id   = msi_ext_dest_id,
 333         .runtime.pin_vcpu       = xen_pin_vcpu,
 334         .ignore_nopv            = true,
 335 };