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>
15 #include <asm/pgtable.h>
16 #include <asm/processor.h>
18 #include <asm/atomic.h>
19 #include <asm/mmu_context.h>
22 #include <asm/percpu.h>
23 #include <asm/proto.h>
24 #include <asm/sections.h>
25 #include <asm/setup.h>
27 #ifndef CONFIG_DEBUG_BOOT_PARAMS
28 struct boot_params __initdata boot_params
;
30 struct boot_params boot_params
;
33 cpumask_t cpu_initialized __cpuinitdata
= CPU_MASK_NONE
;
35 struct x8664_pda
*_cpu_pda
[NR_CPUS
] __read_mostly
;
36 EXPORT_SYMBOL(_cpu_pda
);
37 struct x8664_pda boot_cpu_pda
[NR_CPUS
] __cacheline_aligned
;
39 struct desc_ptr idt_descr
= { 256 * 16 - 1, (unsigned long) idt_table
};
41 char boot_cpu_stack
[IRQSTACKSIZE
] __attribute__((section(".bss.page_aligned")));
43 unsigned long __supported_pte_mask __read_mostly
= ~0UL;
44 static int do_not_nx __cpuinitdata
= 0;
47 Control non executable mappings for 64bit processes.
52 static int __init
nonx_setup(char *str
)
56 if (!strncmp(str
, "on", 2)) {
57 __supported_pte_mask
|= _PAGE_NX
;
59 } else if (!strncmp(str
, "off", 3)) {
61 __supported_pte_mask
&= ~_PAGE_NX
;
65 early_param("noexec", nonx_setup
);
67 int force_personality32
= 0;
70 Control non executable heap for 32bit processes.
71 To control the stack too use noexec=off
73 on PROT_READ does not imply PROT_EXEC for 32bit processes
74 off PROT_READ implies PROT_EXEC (default)
76 static int __init
nonx32_setup(char *str
)
78 if (!strcmp(str
, "on"))
79 force_personality32
&= ~READ_IMPLIES_EXEC
;
80 else if (!strcmp(str
, "off"))
81 force_personality32
|= READ_IMPLIES_EXEC
;
84 __setup("noexec32=", nonx32_setup
);
88 * Declare PDA itself and support (irqstack,tss,pgd) as per cpu data.
89 * Always point %gs to its beginning
91 void __init
setup_per_cpu_areas(void)
96 #ifdef CONFIG_HOTPLUG_CPU
97 prefill_possible_map();
100 /* Copy section for each CPU (we discard the original) */
101 size
= PERCPU_ENOUGH_ROOM
;
103 printk(KERN_INFO
"PERCPU: Allocating %lu bytes of per cpu data\n", size
);
104 for_each_cpu_mask (i
, cpu_possible_map
) {
107 if (!NODE_DATA(cpu_to_node(i
))) {
108 printk("cpu with no node %d, num_online_nodes %d\n",
109 i
, num_online_nodes());
110 ptr
= alloc_bootmem_pages(size
);
112 ptr
= alloc_bootmem_pages_node(NODE_DATA(cpu_to_node(i
)), size
);
115 panic("Cannot allocate cpu data for CPU %d\n", i
);
116 cpu_pda(i
)->data_offset
= ptr
- __per_cpu_start
;
117 memcpy(ptr
, __per_cpu_start
, __per_cpu_end
- __per_cpu_start
);
121 void pda_init(int cpu
)
123 struct x8664_pda
*pda
= cpu_pda(cpu
);
125 /* Setup up data that may be needed in __get_free_pages early */
126 asm volatile("movl %0,%%fs ; movl %0,%%gs" :: "r" (0));
127 /* Memory clobbers used to order PDA accessed */
129 wrmsrl(MSR_GS_BASE
, pda
);
132 pda
->cpunumber
= cpu
;
135 (unsigned long)stack_thread_info() - PDA_STACKOFFSET
+ THREAD_SIZE
;
136 pda
->active_mm
= &init_mm
;
140 /* others are initialized in smpboot.c */
141 pda
->pcurrent
= &init_task
;
142 pda
->irqstackptr
= boot_cpu_stack
;
144 pda
->irqstackptr
= (char *)
145 __get_free_pages(GFP_ATOMIC
, IRQSTACK_ORDER
);
146 if (!pda
->irqstackptr
)
147 panic("cannot allocate irqstack for cpu %d", cpu
);
151 pda
->irqstackptr
+= IRQSTACKSIZE
-64;
154 char boot_exception_stacks
[(N_EXCEPTION_STACKS
- 1) * EXCEPTION_STKSZ
+ DEBUG_STKSZ
]
155 __attribute__((section(".bss.page_aligned")));
157 extern asmlinkage
void ignore_sysret(void);
159 /* May not be marked __init: used by software suspend */
160 void syscall_init(void)
163 * LSTAR and STAR live in a bit strange symbiosis.
164 * They both write to the same internal register. STAR allows to set CS/DS
165 * but only a 32bit target. LSTAR sets the 64bit rip.
167 wrmsrl(MSR_STAR
, ((u64
)__USER32_CS
)<<48 | ((u64
)__KERNEL_CS
)<<32);
168 wrmsrl(MSR_LSTAR
, system_call
);
169 wrmsrl(MSR_CSTAR
, ignore_sysret
);
171 #ifdef CONFIG_IA32_EMULATION
172 syscall32_cpu_init ();
175 /* Flags to clear on syscall */
176 wrmsrl(MSR_SYSCALL_MASK
,
177 X86_EFLAGS_TF
|X86_EFLAGS_DF
|X86_EFLAGS_IF
|X86_EFLAGS_IOPL
);
180 void __cpuinit
check_efer(void)
184 rdmsrl(MSR_EFER
, efer
);
185 if (!(efer
& EFER_NX
) || do_not_nx
) {
186 __supported_pte_mask
&= ~_PAGE_NX
;
190 unsigned long kernel_eflags
;
193 * Copies of the original ist values from the tss are only accessed during
194 * debugging, no special alignment required.
196 DEFINE_PER_CPU(struct orig_ist
, orig_ist
);
199 * cpu_init() initializes state that is per-CPU. Some data is already
200 * initialized (naturally) in the bootstrap process, such as the GDT
201 * and IDT. We reload them nevertheless, this function acts as a
202 * 'CPU state barrier', nothing should get across.
203 * A lot of state is already set up in PDA init.
205 void __cpuinit
cpu_init (void)
207 int cpu
= stack_smp_processor_id();
208 struct tss_struct
*t
= &per_cpu(init_tss
, cpu
);
209 struct orig_ist
*orig_ist
= &per_cpu(orig_ist
, cpu
);
211 char *estacks
= NULL
;
212 struct task_struct
*me
;
215 /* CPU 0 is initialised in head64.c */
219 estacks
= boot_exception_stacks
;
223 if (cpu_test_and_set(cpu
, cpu_initialized
))
224 panic("CPU#%d already initialized!\n", cpu
);
226 printk("Initializing CPU#%d\n", cpu
);
228 clear_in_cr4(X86_CR4_VME
|X86_CR4_PVI
|X86_CR4_TSD
|X86_CR4_DE
);
231 * Initialize the per-CPU GDT with the boot GDT,
232 * and set up the GDT descriptor:
235 memcpy(get_cpu_gdt_table(cpu
), cpu_gdt_table
, GDT_SIZE
);
237 cpu_gdt_descr
[cpu
].size
= GDT_SIZE
;
238 load_gdt((const struct desc_ptr
*)&cpu_gdt_descr
[cpu
]);
239 load_idt((const struct desc_ptr
*)&idt_descr
);
241 memset(me
->thread
.tls_array
, 0, GDT_ENTRY_TLS_ENTRIES
* 8);
244 wrmsrl(MSR_FS_BASE
, 0);
245 wrmsrl(MSR_KERNEL_GS_BASE
, 0);
251 * set up and load the per-CPU TSS
253 for (v
= 0; v
< N_EXCEPTION_STACKS
; v
++) {
254 static const unsigned int order
[N_EXCEPTION_STACKS
] = {
255 [0 ... N_EXCEPTION_STACKS
- 1] = EXCEPTION_STACK_ORDER
,
256 [DEBUG_STACK
- 1] = DEBUG_STACK_ORDER
259 estacks
= (char *)__get_free_pages(GFP_ATOMIC
, order
[v
]);
261 panic("Cannot allocate exception stack %ld %d\n",
264 estacks
+= PAGE_SIZE
<< order
[v
];
265 orig_ist
->ist
[v
] = t
->x86_tss
.ist
[v
] = (unsigned long)estacks
;
268 t
->x86_tss
.io_bitmap_base
= offsetof(struct tss_struct
, io_bitmap
);
270 * <= is required because the CPU will access up to
271 * 8 bits beyond the end of the IO permission bitmap.
273 for (i
= 0; i
<= IO_BITMAP_LONGS
; i
++)
274 t
->io_bitmap
[i
] = ~0UL;
276 atomic_inc(&init_mm
.mm_count
);
277 me
->active_mm
= &init_mm
;
280 enter_lazy_tlb(&init_mm
, me
);
282 set_tss_desc(cpu
, t
);
284 load_LDT(&init_mm
.context
);
287 * Clear all 6 debug registers:
290 set_debugreg(0UL, 0);
291 set_debugreg(0UL, 1);
292 set_debugreg(0UL, 2);
293 set_debugreg(0UL, 3);
294 set_debugreg(0UL, 6);
295 set_debugreg(0UL, 7);
299 raw_local_save_flags(kernel_eflags
);