/*
- * Copyright (c) 2003-2009 Apple Inc. All rights reserved.
+ * Copyright (c) 2003-2016 Apple Inc. All rights reserved.
*
* @APPLE_OSREFERENCE_LICENSE_HEADER_START@
*
* the rights to redistribute these changes.
*/
-#include <platforms.h>
-#include <mach_kdb.h>
#include <mach/i386/vm_param.h>
#include <kern/xpr.h>
#include <kern/cpu_data.h>
#include <kern/processor.h>
+#include <sys/kdebug.h>
#include <console/serial_protos.h>
#include <vm/vm_page.h>
#include <vm/pmap.h>
#include <i386/pmCPU.h>
#include <i386/tsc.h>
#include <i386/locks.h> /* LcksOpts */
-#ifdef __i386__
-#include <i386/cpu_capabilities.h>
-#if MACH_KDB
-#include <machine/db_machdep.h>
-#endif
-#endif
#if DEBUG
#include <machine/pal_routines.h>
#endif
+#if MONOTONIC
+#include <kern/monotonic.h>
+#endif /* MONOTONIC */
+
+#include <san/kasan.h>
+
#if DEBUG
#define DBG(x...) kprintf(x)
#else
#define DBG(x...)
#endif
-#if MACH_KDB
-#include <ddb/db_aout.h>
-#endif /* MACH_KDB */
int debug_task;
extern const char version_variant[];
extern int nx_enabled;
-#ifdef __x86_64__
-extern void *low_eintstack;
-#endif
+/*
+ * Set initial values so that ml_phys_* routines can use the booter's ID mapping
+ * to touch physical space before the kernel's physical aperture exists.
+ */
+uint64_t physmap_base = 0;
+uint64_t physmap_max = 4*GB;
-void *KPTphys;
+pd_entry_t *KPTphys;
pd_entry_t *IdlePTD;
-#ifdef __i386__
-pd_entry_t *IdlePDPT64;
-#endif
+pdpt_entry_t *IdlePDPT;
+pml4_entry_t *IdlePML4;
char *physfree;
uintptr_t tmp = (uintptr_t)physfree;
bzero(physfree, npages * PAGE_SIZE);
physfree += npages * PAGE_SIZE;
-#ifdef __x86_64__
tmp += VM_MIN_KERNEL_ADDRESS & ~LOW_4GB_MASK;
-#endif
return (void *)tmp;
}
extern pmap_paddr_t first_avail;
-#ifdef __x86_64__
int break_kprintf = 0;
uint64_t
set_cr3_raw((uint32_t) new_cr3);
}
-#endif
-#ifdef __i386__
-#define ID_MAP_VTOP(x) x
-#endif
-#ifdef __x86_64__
// Set up the physical mapping - NPHYSMAP GB of memory mapped at a high address
// NPHYSMAP is determined by the maximum supported RAM size plus 4GB to account
// the PCI hole (which is less 4GB but not more).
-// Compile-time guard:
-extern int maxphymapsupported[NPHYSMAP <= PTE_PER_PAGE ? 1 : -1];
+/* Compile-time guard: NPHYSMAP is capped to 256GiB, accounting for
+ * randomisation
+ */
+extern int maxphymapsupported[NPHYSMAP <= (PTE_PER_PAGE/2) ? 1 : -1];
+
static void
physmap_init(void)
{
pt_entry_t entries[PTE_PER_PAGE];
} * physmapL2 = ALLOCPAGES(NPHYSMAP);
- uintptr_t i;
- for(i=0;i<NPHYSMAP;i++) {
- physmapL3[i] = ((uintptr_t)ID_MAP_VTOP(&physmapL2[i]))
+ uint64_t i;
+ uint8_t phys_random_L3 = early_random() & 0xFF;
+
+ /* We assume NX support. Mark all levels of the PHYSMAP NX
+ * to avoid granting executability via a single bit flip.
+ */
+#if DEVELOPMENT || DEBUG
+ uint32_t reg[4];
+ do_cpuid(0x80000000, reg);
+ if (reg[eax] >= 0x80000001) {
+ do_cpuid(0x80000001, reg);
+ assert(reg[edx] & CPUID_EXTFEATURE_XD);
+ }
+#endif /* DEVELOPMENT || DEBUG */
+
+ for(i = 0; i < NPHYSMAP; i++) {
+ physmapL3[i + phys_random_L3] =
+ ((uintptr_t)ID_MAP_VTOP(&physmapL2[i]))
| INTEL_PTE_VALID
+ | INTEL_PTE_NX
| INTEL_PTE_WRITE;
- uintptr_t j;
- for(j=0;j<PTE_PER_PAGE;j++) {
- physmapL2[i].entries[j] = (((i*PTE_PER_PAGE+j)<<PDSHIFT)
+
+ uint64_t j;
+ for(j = 0; j < PTE_PER_PAGE; j++) {
+ physmapL2[i].entries[j] =
+ ((i * PTE_PER_PAGE + j) << PDSHIFT)
| INTEL_PTE_PS
| INTEL_PTE_VALID
- | INTEL_PTE_WRITE);
+ | INTEL_PTE_NX
+ | INTEL_PTE_WRITE;
}
}
- IdlePML4[KERNEL_PHYSMAP_INDEX] = ((uintptr_t)ID_MAP_VTOP(physmapL3))
- | INTEL_PTE_VALID
- | INTEL_PTE_WRITE;
- if (cpuid_extfeatures() & CPUID_EXTFEATURE_XD) {
- IdlePML4[KERNEL_PHYSMAP_INDEX] |= INTEL_PTE_NX;
- }
+ IdlePML4[KERNEL_PHYSMAP_PML4_INDEX] =
+ ((uintptr_t)ID_MAP_VTOP(physmapL3))
+ | INTEL_PTE_VALID
+ | INTEL_PTE_NX
+ | INTEL_PTE_WRITE;
+
+ physmap_base = KVADDR(KERNEL_PHYSMAP_PML4_INDEX, phys_random_L3, 0, 0);
+ physmap_max = physmap_base + NPHYSMAP * GB;
+ DBG("Physical address map base: 0x%qx\n", physmap_base);
+ DBG("Physical map idlepml4[%d]: 0x%llx\n",
+ KERNEL_PHYSMAP_PML4_INDEX, IdlePML4[KERNEL_PHYSMAP_PML4_INDEX]);
+}
- DBG("physical map idlepml4[%d]: 0x%llx\n",
- KERNEL_PHYSMAP_INDEX, IdlePML4[KERNEL_PHYSMAP_INDEX]);
+static void
+descriptor_alias_init()
+{
+ vm_offset_t master_gdt_phys;
+ vm_offset_t master_gdt_alias_phys;
+ vm_offset_t master_idt_phys;
+ vm_offset_t master_idt_alias_phys;
+
+ assert(((vm_offset_t)master_gdt & PAGE_MASK) == 0);
+ assert(((vm_offset_t)master_idt64 & PAGE_MASK) == 0);
+
+ master_gdt_phys = (vm_offset_t) ID_MAP_VTOP(master_gdt);
+ master_idt_phys = (vm_offset_t) ID_MAP_VTOP(master_idt64);
+ master_gdt_alias_phys = (vm_offset_t) ID_MAP_VTOP(MASTER_GDT_ALIAS);
+ master_idt_alias_phys = (vm_offset_t) ID_MAP_VTOP(MASTER_IDT_ALIAS);
+
+ DBG("master_gdt_phys: %p\n", (void *) master_gdt_phys);
+ DBG("master_idt_phys: %p\n", (void *) master_idt_phys);
+ DBG("master_gdt_alias_phys: %p\n", (void *) master_gdt_alias_phys);
+ DBG("master_idt_alias_phys: %p\n", (void *) master_idt_alias_phys);
+
+ KPTphys[atop_kernel(master_gdt_alias_phys)] = master_gdt_phys |
+ INTEL_PTE_VALID | INTEL_PTE_NX | INTEL_PTE_WRITE;
+ KPTphys[atop_kernel(master_idt_alias_phys)] = master_idt_phys |
+ INTEL_PTE_VALID | INTEL_PTE_NX; /* read-only */
}
-#endif
static void
Idle_PTs_init(void)
/* Allocate the "idle" kernel page tables: */
KPTphys = ALLOCPAGES(NKPT); /* level 1 */
IdlePTD = ALLOCPAGES(NPGPTD); /* level 2 */
+ IdlePDPT = ALLOCPAGES(1); /* level 3 */
+ IdlePML4 = ALLOCPAGES(1); /* level 4 */
-#ifdef __x86_64__
- physmap_init();
-#else
- IdlePDPT64 = ALLOCPAGES(1);
-
- // Recursive mapping of PTEs
- fillkpt(IdlePTD, INTEL_PTE_WRITE, (uintptr_t)IdlePTD, PTDPTDI, NPGPTD);
- // commpage
- fillkpt(IdlePTD, INTEL_PTE_WRITE|INTEL_PTE_USER, (uintptr_t)ALLOCPAGES(1), _COMM_PAGE32_BASE_ADDRESS >> PDESHIFT,1);
-#endif
// Fill the lowest level with everything up to physfree
fillkpt(KPTphys,
- INTEL_PTE_WRITE, 0, 0, (int)(((uintptr_t)physfree) >> PAGE_SHIFT));
+ INTEL_PTE_WRITE, 0, 0, (int)(((uintptr_t)physfree) >> PAGE_SHIFT));
- // Rewrite the 2nd-lowest level to point to pages of KPTphys.
- // This was previously filled statically by idle_pt.c, and thus
- // must be done after the KPTphys fill since IdlePTD is in use
+ /* IdlePTD */
fillkpt(IdlePTD,
- INTEL_PTE_WRITE, (uintptr_t)ID_MAP_VTOP(KPTphys), 0, NKPT);
+ INTEL_PTE_WRITE, (uintptr_t)ID_MAP_VTOP(KPTphys), 0, NKPT);
// IdlePDPT entries
-#ifdef __i386__
- fillkpt(IdlePDPT, 0, (uintptr_t)IdlePTD, 0, NPGPTD);
-#else
- fillkpt(IdlePDPT, INTEL_PTE_WRITE, (uintptr_t)ID_MAP_VTOP(IdlePTD), 0, NPGPTD);
-#endif
+ fillkpt(IdlePDPT,
+ INTEL_PTE_WRITE, (uintptr_t)ID_MAP_VTOP(IdlePTD), 0, NPGPTD);
+
+ // IdlePML4 single entry for kernel space.
+ fillkpt(IdlePML4 + KERNEL_PML4_INDEX,
+ INTEL_PTE_WRITE, (uintptr_t)ID_MAP_VTOP(IdlePDPT), 0, 1);
+
+ postcode(VSTART_PHYSMAP_INIT);
+
+ physmap_init();
+
+ postcode(VSTART_DESC_ALIAS_INIT);
+
+ descriptor_alias_init();
+
+ postcode(VSTART_SET_CR3);
+
+ // Switch to the page tables..
+ set_cr3_raw((uintptr_t)ID_MAP_VTOP(IdlePML4));
+
+}
+
+extern void vstart_trap_handler;
+
+#define BOOT_TRAP_VECTOR(t) \
+ [t] = { \
+ (uintptr_t) &vstart_trap_handler, \
+ KERNEL64_CS, \
+ 0, \
+ ACC_P|ACC_PL_K|ACC_INTR_GATE, \
+ 0 \
+ },
+
+/* Recursive macro to iterate 0..31 */
+#define L0(x,n) x(n)
+#define L1(x,n) L0(x,n-1) L0(x,n)
+#define L2(x,n) L1(x,n-2) L1(x,n)
+#define L3(x,n) L2(x,n-4) L2(x,n)
+#define L4(x,n) L3(x,n-8) L3(x,n)
+#define L5(x,n) L4(x,n-16) L4(x,n)
+#define FOR_0_TO_31(x) L5(x,31)
+
+/*
+ * Bootstrap IDT. Active only during early startup.
+ * Only the trap vectors are defined since interrupts are masked.
+ * All traps point to a common handler.
+ */
+struct fake_descriptor64 master_boot_idt64[IDTSZ]
+ __attribute__((section("__HIB,__desc")))
+ __attribute__((aligned(PAGE_SIZE))) = {
+ FOR_0_TO_31(BOOT_TRAP_VECTOR)
+};
- // Flush the TLB now we're done rewriting the page tables..
- set_cr3_raw(get_cr3_raw());
+static void
+vstart_idt_init(void)
+{
+ x86_64_desc_register_t vstart_idt = {
+ sizeof(master_boot_idt64),
+ master_boot_idt64 };
+
+ fix_desc64(master_boot_idt64, 32);
+ lidt((void *)&vstart_idt);
}
/*
* Non-bootstrap processors are called with argument boot_args_start NULL.
* These processors switch immediately to the existing kernel page tables.
*/
+__attribute__((noreturn))
void
vstart(vm_offset_t boot_args_start)
{
boolean_t is_boot_cpu = !(boot_args_start == 0);
- int cpu;
- uint32_t lphysfree;
+ int cpu = 0;
+ uint32_t lphysfree;
postcode(VSTART_ENTRY);
if (is_boot_cpu) {
+ /*
+ * Set-up temporary trap handlers during page-table set-up.
+ */
+ vstart_idt_init();
+ postcode(VSTART_IDT_INIT);
+
/*
* Get startup parameters.
*/
kernelBootArgs = (boot_args *)boot_args_start;
lphysfree = kernelBootArgs->kaddr + kernelBootArgs->ksize;
physfree = (void *)(uintptr_t)((lphysfree + PAGE_SIZE - 1) &~ (PAGE_SIZE - 1));
-#if DEBUG
+
+#if DEVELOPMENT || DEBUG
pal_serial_init();
#endif
DBG("revision 0x%x\n", kernelBootArgs->Revision);
kernelBootArgs,
&kernelBootArgs->ksize,
&kernelBootArgs->kaddr);
-#ifdef __x86_64__
- /* enable NX/XD, boot processor */
- if (cpuid_extfeatures() & CPUID_EXTFEATURE_XD) {
- wrmsr64(MSR_IA32_EFER, rdmsr64(MSR_IA32_EFER) | MSR_IA32_EFER_NXE);
- DBG("vstart() NX/XD enabled\n");
- }
-#endif
- postcode(PSTART_PAGE_TABLES);
-
- Idle_PTs_init();
+ DBG("SMBIOS mem sz 0x%llx\n", kernelBootArgs->PhysicalMemorySize);
- first_avail = (vm_offset_t)ID_MAP_VTOP(physfree);
-
- cpu = 0;
- cpu_data_alloc(TRUE);
-
-
/*
* Setup boot args given the physical start address.
+ * Note: PE_init_platform needs to be called before Idle_PTs_init
+ * because access to the DeviceTree is required to read the
+ * random seed before generating a random physical map slide.
*/
kernelBootArgs = (boot_args *)
ml_static_ptovirt(boot_args_start);
DBG("i386_init(0x%lx) kernelBootArgs=%p\n",
(unsigned long)boot_args_start, kernelBootArgs);
+#if KASAN
+ kasan_reserve_memory(kernelBootArgs);
+#endif
+
PE_init_platform(FALSE, kernelBootArgs);
postcode(PE_INIT_PLATFORM_D);
+
+ Idle_PTs_init();
+ postcode(VSTART_IDLE_PTS_INIT);
+
+#if KASAN
+ /* Init kasan and map whatever was stolen from physfree */
+ kasan_init();
+ kasan_notify_stolen((uintptr_t)ml_static_ptovirt((vm_offset_t)physfree));
+#endif
+
+#if MONOTONIC
+ mt_init();
+#endif /* MONOTONIC */
+
+ first_avail = (vm_offset_t)ID_MAP_VTOP(physfree);
+
+ cpu_data_alloc(TRUE);
+
+ cpu_desc_init(cpu_datap(0));
+ postcode(VSTART_CPU_DESC_INIT);
+ cpu_desc_load(cpu_datap(0));
+
+ postcode(VSTART_CPU_MODE_INIT);
+ cpu_syscall_init(cpu_datap(0)); /* cpu_syscall_init() will be
+ * invoked on the APs
+ * via i386_init_slave()
+ */
} else {
+ /* Switch to kernel's page tables (from the Boot PTs) */
+ set_cr3_raw((uintptr_t)ID_MAP_VTOP(IdlePML4));
/* Find our logical cpu number */
cpu = lapic_to_cpu[(LAPIC_READ(ID)>>LAPIC_ID_SHIFT) & LAPIC_ID_MASK];
DBG("CPU: %d, GSBASE initial value: 0x%llx\n", cpu, rdmsr64(MSR_IA32_GS_BASE));
-#ifdef __x86_64__
- if (cpuid_extfeatures() & CPUID_EXTFEATURE_XD) {
- wrmsr64(MSR_IA32_EFER, rdmsr64(MSR_IA32_EFER) | MSR_IA32_EFER_NXE);
- DBG("vstart() NX/XD enabled, non-boot\n");
- }
-#endif
+ cpu_desc_load(cpu_datap(cpu));
}
-#ifdef __x86_64__
- if(is_boot_cpu)
- cpu_desc_init64(cpu_datap(cpu));
- cpu_desc_load64(cpu_datap(cpu));
-#else
- if(is_boot_cpu)
- cpu_desc_init(cpu_datap(cpu));
- cpu_desc_load(cpu_datap(cpu));
-#endif
- if (is_boot_cpu)
- cpu_mode_init(current_cpu_datap()); /* cpu_mode_init() will be
- * invoked on the APs
- * via i386_init_slave()
- */
-#ifdef __x86_64__
- /* Done with identity mapping */
- IdlePML4[0] = 0;
-#endif
-
postcode(VSTART_EXIT);
-#ifdef __i386__
- if (cpuid_extfeatures() & CPUID_EXTFEATURE_XD) {
- wrmsr64(MSR_IA32_EFER, rdmsr64(MSR_IA32_EFER) | MSR_IA32_EFER_NXE);
- DBG("vstart() NX/XD enabled, i386\n");
- }
+ x86_init_wrapper(is_boot_cpu ? (uintptr_t) i386_init
+ : (uintptr_t) i386_init_slave,
+ cpu_datap(cpu)->cpu_int_stack_top);
+}
- if (is_boot_cpu)
- i386_init();
- else
- i386_init_slave();
- /*NOTREACHED*/
-#else
- /* We need to switch to a new per-cpu stack, but we must do this atomically with
- * the call to ensure the compiler doesn't assume anything about the stack before
- * e.g. tail-call optimisations
- */
- if (is_boot_cpu)
- {
- asm volatile(
- "mov %1, %%rdi;"
- "mov %0, %%rsp;"
- "call _i386_init;" : : "r"
- (cpu_datap(cpu)->cpu_int_stack_top), "r" (boot_args_start));
- }
- else
- {
- asm volatile(
- "mov %0, %%rsp;"
- "call _i386_init_slave;" : : "r"
- (cpu_datap(cpu)->cpu_int_stack_top));
- }
- /*NOTREACHED*/
-#endif
+void
+pstate_trace(void)
+{
}
/*
postcode(I386_INIT_ENTRY);
pal_i386_init();
+ tsc_init();
+ rtclock_early_init(); /* mach_absolute_time() now functionsl */
+
+ kernel_debug_string_early("i386_init");
+ pstate_trace();
#if CONFIG_MCA
/* Initialize machine-check handling */
mca_cpu_init();
#endif
-
- kernel_early_bootstrap();
-
master_cpu = 0;
cpu_init();
panic_init(); /* Init this in case we need debugger */
/* setup debugging output if one has been chosen */
+ kernel_debug_string_early("PE_init_kprintf");
PE_init_kprintf(FALSE);
+ kernel_debug_string_early("kernel_early_bootstrap");
+ kernel_early_bootstrap();
+
if (!PE_parse_boot_argn("diag", &dgWork.dgFlags, sizeof (dgWork.dgFlags)))
dgWork.dgFlags = 0;
serialmode = 0;
- if(PE_parse_boot_argn("serial", &serialmode, sizeof (serialmode))) {
+ if (PE_parse_boot_argn("serial", &serialmode, sizeof(serialmode))) {
/* We want a serial keyboard and/or console */
kprintf("Serial mode specified: %08X\n", serialmode);
+ int force_sync = serialmode & SERIALMODE_SYNCDRAIN;
+ if (force_sync || PE_parse_boot_argn("drain_uart_sync", &force_sync, sizeof(force_sync))) {
+ if (force_sync) {
+ serialmode |= SERIALMODE_SYNCDRAIN;
+ kprintf(
+ "WARNING: Forcing uart driver to output synchronously."
+ "printf()s/IOLogs will impact kernel performance.\n"
+ "You are advised to avoid using 'drain_uart_sync' boot-arg.\n");
+ }
+ }
}
- if(serialmode & 1) {
+ if (serialmode & SERIALMODE_OUTPUT) {
(void)switch_to_serial_console();
- disableConsoleOutput = FALSE; /* Allow printfs to happen */
+ disableConsoleOutput = FALSE; /* Allow printfs to happen */
}
/* setup console output */
+ kernel_debug_string_early("PE_init_printf");
PE_init_printf(FALSE);
kprintf("version_variant = %s\n", version_variant);
/*
* debug support for > 4G systems
*/
- if (!PE_parse_boot_argn("himemory_mode", &vm_himemory_mode, sizeof (vm_himemory_mode)))
- vm_himemory_mode = 0;
+ PE_parse_boot_argn("himemory_mode", &vm_himemory_mode, sizeof (vm_himemory_mode));
+ if (vm_himemory_mode != 0)
+ kprintf("himemory_mode: %d\n", vm_himemory_mode);
if (!PE_parse_boot_argn("immediate_NMI", &fidn, sizeof (fidn)))
force_immediate_debugger_NMI = FALSE;
&urgency_notification_assert_abstime_threshold,
sizeof(urgency_notification_assert_abstime_threshold));
-#if CONFIG_YONAH
- /*
- * At this point we check whether we are a 64-bit processor
- * and that we're not restricted to legacy mode, 32-bit operation.
- */
- if (cpuid_extfeatures() & CPUID_EXTFEATURE_EM64T) {
- boolean_t legacy_mode;
- kprintf("EM64T supported");
- if (PE_parse_boot_argn("-legacy", &legacy_mode, sizeof (legacy_mode))) {
- kprintf(" but legacy mode forced\n");
- IA32e = FALSE;
- } else {
- kprintf(" and will be enabled\n");
- }
- } else
- IA32e = FALSE;
-#endif
-
if (!(cpuid_extfeatures() & CPUID_EXTFEATURE_XD))
nx_enabled = 0;
/*
* VM initialization, after this we're using page tables...
- * The maximum number of cpus must be set beforehand.
+ * Thn maximum number of cpus must be set beforehand.
*/
+ kernel_debug_string_early("i386_vm_init");
i386_vm_init(maxmemtouse, IA32e, kernelBootArgs);
/* create the console for verbose or pretty mode */
PE_init_platform(TRUE, kernelBootArgs);
PE_create_console();
- tsc_init();
+ kernel_debug_string_early("power_management_init");
power_management_init();
-
processor_bootstrap();
thread_bootstrap();
+ pstate_trace();
+ kernel_debug_string_early("machine_startup");
machine_startup();
+ pstate_trace();
}
static void
assert(!ml_get_interrupts_enabled());
- cpu_mode_init(current_cpu_datap());
+ cpu_syscall_init(current_cpu_datap());
+ pmap_cpu_init();
#if CONFIG_MCA
mca_cpu_init();
#endif
+ LAPIC_INIT();
lapic_configure();
LAPIC_DUMP();
LAPIC_CPU_MAP_DUMP();
#if CONFIG_MTRR
mtrr_update_cpu();
#endif
+ /* update CPU microcode */
+ ucode_update_wake();
} else
init_param = FAST_SLAVE_INIT;
- /* update CPU microcode */
- ucode_update_wake();
-
#if CONFIG_VMX
/* resume VT operation */
- vmx_resume();
+ vmx_resume(FALSE);
#endif
#if CONFIG_MTRR
cpu_thread_init(); /* not strictly necessary */
-#ifdef __x86_64__
- /* Re-zero the identity-map for the idle PT's. This MUST be done before
- * cpu_running is set so that other slaves can set up their own
- * identity-map */
- if (!fast_restart)
- IdlePML4[0] = 0;
-#endif
-
- cpu_init(); /* Sets cpu_running which starter cpu waits for */
-
+ cpu_init(); /* Sets cpu_running which starter cpu waits for */
slave_main(init_param);
panic("do_init_slave() returned from slave_main()");
projects / apple / xnu.git / blobdiff