2 * X86-64 specific CPU setup.
3 * Copyright (C) 1995 Linus Torvalds
4 * Copyright 2001, 2002, 2003 SuSE Labs / Andi Kleen.
5 * See setup.c for older changelog.
7 #include <linux/init.h>
8 #include <linux/kernel.h>
9 #include <linux/sched.h>
10 #include <linux/string.h>
11 #include <linux/bootmem.h>
12 #include <linux/bitops.h>
13 #include <linux/module.h>
14 #include <linux/kgdb.h>
15 #include <linux/topology.h>
17 #include <asm/pgtable.h>
18 #include <asm/processor.h>
20 #include <asm/atomic.h>
21 #include <asm/mmu_context.h>
24 #include <asm/percpu.h>
25 #include <asm/proto.h>
26 #include <asm/sections.h>
27 #include <asm/setup.h>
28 #include <asm/genapic.h>
30 #ifndef CONFIG_DEBUG_BOOT_PARAMS
31 struct boot_params __initdata boot_params;
33 struct boot_params boot_params;
36 cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
38 struct x8664_pda **_cpu_pda __read_mostly;
39 EXPORT_SYMBOL(_cpu_pda);
41 struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
43 char boot_cpu_stack[IRQSTACKSIZE] __page_aligned_bss;
45 unsigned long __supported_pte_mask __read_mostly = ~0UL;
46 EXPORT_SYMBOL_GPL(__supported_pte_mask);
48 static int do_not_nx __cpuinitdata = 0;
51 Control non executable mappings for 64bit processes.
56 static int __init nonx_setup(char *str)
60 if (!strncmp(str, "on", 2)) {
61 __supported_pte_mask |= _PAGE_NX;
63 } else if (!strncmp(str, "off", 3)) {
65 __supported_pte_mask &= ~_PAGE_NX;
69 early_param("noexec", nonx_setup);
71 int force_personality32 = 0;
74 Control non executable heap for 32bit processes.
75 To control the stack too use noexec=off
77 on PROT_READ does not imply PROT_EXEC for 32bit processes (default)
78 off PROT_READ implies PROT_EXEC
80 static int __init nonx32_setup(char *str)
82 if (!strcmp(str, "on"))
83 force_personality32 &= ~READ_IMPLIES_EXEC;
84 else if (!strcmp(str, "off"))
85 force_personality32 |= READ_IMPLIES_EXEC;
88 __setup("noexec32=", nonx32_setup);
90 void pda_init(int cpu)
92 struct x8664_pda *pda = cpu_pda(cpu);
94 /* Setup up data that may be needed in __get_free_pages early */
95 asm volatile("movl %0,%%fs ; movl %0,%%gs" :: "r" (0));
96 /* Memory clobbers used to order PDA accessed */
98 wrmsrl(MSR_GS_BASE, pda);
101 pda->cpunumber = cpu;
104 (unsigned long)stack_thread_info() - PDA_STACKOFFSET + THREAD_SIZE;
105 pda->active_mm = &init_mm;
109 /* others are initialized in smpboot.c */
110 pda->pcurrent = &init_task;
111 pda->irqstackptr = boot_cpu_stack;
113 pda->irqstackptr = (char *)
114 __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
115 if (!pda->irqstackptr)
116 panic("cannot allocate irqstack for cpu %d", cpu);
118 if (pda->nodenumber == 0 && cpu_to_node(cpu) != NUMA_NO_NODE)
119 pda->nodenumber = cpu_to_node(cpu);
122 pda->irqstackptr += IRQSTACKSIZE-64;
125 char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]
126 __attribute__((section(".bss.page_aligned")));
128 extern asmlinkage void ignore_sysret(void);
130 /* May not be marked __init: used by software suspend */
131 void syscall_init(void)
134 * LSTAR and STAR live in a bit strange symbiosis.
135 * They both write to the same internal register. STAR allows to set CS/DS
136 * but only a 32bit target. LSTAR sets the 64bit rip.
138 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
139 wrmsrl(MSR_LSTAR, system_call);
140 wrmsrl(MSR_CSTAR, ignore_sysret);
142 #ifdef CONFIG_IA32_EMULATION
143 syscall32_cpu_init ();
146 /* Flags to clear on syscall */
147 wrmsrl(MSR_SYSCALL_MASK,
148 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
151 void __cpuinit check_efer(void)
155 rdmsrl(MSR_EFER, efer);
156 if (!(efer & EFER_NX) || do_not_nx) {
157 __supported_pte_mask &= ~_PAGE_NX;
161 unsigned long kernel_eflags;
164 * Copies of the original ist values from the tss are only accessed during
165 * debugging, no special alignment required.
167 DEFINE_PER_CPU(struct orig_ist, orig_ist);
170 * cpu_init() initializes state that is per-CPU. Some data is already
171 * initialized (naturally) in the bootstrap process, such as the GDT
172 * and IDT. We reload them nevertheless, this function acts as a
173 * 'CPU state barrier', nothing should get across.
174 * A lot of state is already set up in PDA init.
176 void __cpuinit cpu_init (void)
178 int cpu = stack_smp_processor_id();
179 struct tss_struct *t = &per_cpu(init_tss, cpu);
180 struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
182 char *estacks = NULL;
183 struct task_struct *me;
186 /* CPU 0 is initialised in head64.c */
190 estacks = boot_exception_stacks;
194 if (cpu_test_and_set(cpu, cpu_initialized))
195 panic("CPU#%d already initialized!\n", cpu);
197 printk("Initializing CPU#%d\n", cpu);
199 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
202 * Initialize the per-CPU GDT with the boot GDT,
203 * and set up the GDT descriptor:
206 memcpy(get_cpu_gdt_table(cpu), cpu_gdt_table, GDT_SIZE);
208 cpu_gdt_descr[cpu].size = GDT_SIZE;
209 load_gdt((const struct desc_ptr *)&cpu_gdt_descr[cpu]);
210 load_idt((const struct desc_ptr *)&idt_descr);
212 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
215 wrmsrl(MSR_FS_BASE, 0);
216 wrmsrl(MSR_KERNEL_GS_BASE, 0);
222 * set up and load the per-CPU TSS
224 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
225 static const unsigned int order[N_EXCEPTION_STACKS] = {
226 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
227 [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
230 estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
232 panic("Cannot allocate exception stack %ld %d\n",
235 estacks += PAGE_SIZE << order[v];
236 orig_ist->ist[v] = t->x86_tss.ist[v] = (unsigned long)estacks;
239 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
241 * <= is required because the CPU will access up to
242 * 8 bits beyond the end of the IO permission bitmap.
244 for (i = 0; i <= IO_BITMAP_LONGS; i++)
245 t->io_bitmap[i] = ~0UL;
247 atomic_inc(&init_mm.mm_count);
248 me->active_mm = &init_mm;
251 enter_lazy_tlb(&init_mm, me);
253 set_tss_desc(cpu, t);
255 load_LDT(&init_mm.context);
259 * If the kgdb is connected no debug regs should be altered. This
260 * is only applicable when KGDB and a KGDB I/O module are built
261 * into the kernel and you are using early debugging with
262 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
264 if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
265 arch_kgdb_ops.correct_hw_break();
269 * Clear all 6 debug registers:
272 set_debugreg(0UL, 0);
273 set_debugreg(0UL, 1);
274 set_debugreg(0UL, 2);
275 set_debugreg(0UL, 3);
276 set_debugreg(0UL, 6);
277 set_debugreg(0UL, 7);
279 /* If the kgdb is connected no debug regs should be altered. */
285 raw_local_save_flags(kernel_eflags);