Merge branches 'x86/numa-fixes', 'x86/apic', 'x86/apm', 'x86/bitops', 'x86/build...
[linux-2.6-block.git] / arch / x86 / kernel / paravirt.c
CommitLineData
d3561b7f
RR
1/* Paravirtualization interfaces
2 Copyright (C) 2006 Rusty Russell IBM Corporation
3
4 This program is free software; you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation; either version 2 of the License, or
7 (at your option) any later version.
8
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
13
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
b1df07bd
GOC
17
18 2007 - x86_64 support added by Glauber de Oliveira Costa, Red Hat Inc
d3561b7f 19*/
b1df07bd 20
d3561b7f
RR
21#include <linux/errno.h>
22#include <linux/module.h>
23#include <linux/efi.h>
24#include <linux/bcd.h>
ce6234b5 25#include <linux/highmem.h>
d3561b7f
RR
26
27#include <asm/bug.h>
28#include <asm/paravirt.h>
29#include <asm/desc.h>
30#include <asm/setup.h>
31#include <asm/arch_hooks.h>
32#include <asm/time.h>
33#include <asm/irq.h>
34#include <asm/delay.h>
13623d79
RR
35#include <asm/fixmap.h>
36#include <asm/apic.h>
da181a8b 37#include <asm/tlbflush.h>
6cb9a835 38#include <asm/timer.h>
d3561b7f
RR
39
40/* nop stub */
45876233 41void _paravirt_nop(void)
d3561b7f
RR
42{
43}
44
45static void __init default_banner(void)
46{
47 printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
93b1eab3 48 pv_info.name);
d3561b7f
RR
49}
50
51char *memory_setup(void)
52{
93b1eab3 53 return pv_init_ops.memory_setup();
d3561b7f
RR
54}
55
139ec7c4 56/* Simple instruction patching code. */
93b1eab3
JF
57#define DEF_NATIVE(ops, name, code) \
58 extern const char start_##ops##_##name[], end_##ops##_##name[]; \
59 asm("start_" #ops "_" #name ": " code "; end_" #ops "_" #name ":")
60
93b1eab3
JF
61/* Undefined instruction for dealing with missing ops pointers. */
62static const unsigned char ud2a[] = { 0x0f, 0x0b };
139ec7c4 63
63f70270
JF
64unsigned paravirt_patch_nop(void)
65{
66 return 0;
67}
68
69unsigned paravirt_patch_ignore(unsigned len)
70{
71 return len;
72}
73
19d36ccd
AK
74struct branch {
75 unsigned char opcode;
76 u32 delta;
77} __attribute__((packed));
78
ab144f5e
AK
79unsigned paravirt_patch_call(void *insnbuf,
80 const void *target, u16 tgt_clobbers,
81 unsigned long addr, u16 site_clobbers,
63f70270
JF
82 unsigned len)
83{
ab144f5e
AK
84 struct branch *b = insnbuf;
85 unsigned long delta = (unsigned long)target - (addr+5);
63f70270
JF
86
87 if (tgt_clobbers & ~site_clobbers)
88 return len; /* target would clobber too much for this site */
89 if (len < 5)
90 return len; /* call too long for patch site */
139ec7c4 91
ab144f5e
AK
92 b->opcode = 0xe8; /* call */
93 b->delta = delta;
94 BUILD_BUG_ON(sizeof(*b) != 5);
139ec7c4 95
63f70270
JF
96 return 5;
97}
98
93b1eab3 99unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
ab144f5e 100 unsigned long addr, unsigned len)
63f70270 101{
ab144f5e
AK
102 struct branch *b = insnbuf;
103 unsigned long delta = (unsigned long)target - (addr+5);
63f70270
JF
104
105 if (len < 5)
106 return len; /* call too long for patch site */
107
ab144f5e
AK
108 b->opcode = 0xe9; /* jmp */
109 b->delta = delta;
63f70270
JF
110
111 return 5;
112}
113
93b1eab3
JF
114/* Neat trick to map patch type back to the call within the
115 * corresponding structure. */
116static void *get_call_destination(u8 type)
117{
118 struct paravirt_patch_template tmpl = {
119 .pv_init_ops = pv_init_ops,
93b1eab3
JF
120 .pv_time_ops = pv_time_ops,
121 .pv_cpu_ops = pv_cpu_ops,
122 .pv_irq_ops = pv_irq_ops,
123 .pv_apic_ops = pv_apic_ops,
124 .pv_mmu_ops = pv_mmu_ops,
125 };
126 return *((void **)&tmpl + type);
127}
128
ab144f5e
AK
129unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
130 unsigned long addr, unsigned len)
63f70270 131{
93b1eab3 132 void *opfunc = get_call_destination(type);
63f70270
JF
133 unsigned ret;
134
135 if (opfunc == NULL)
136 /* If there's no function, patch it with a ud2a (BUG) */
93b1eab3 137 ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a));
63f70270
JF
138 else if (opfunc == paravirt_nop)
139 /* If the operation is a nop, then nop the callsite */
140 ret = paravirt_patch_nop();
93b1eab3 141 else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) ||
6abcd98f 142 type == PARAVIRT_PATCH(pv_cpu_ops.irq_enable_syscall_ret))
63f70270 143 /* If operation requires a jmp, then jmp */
93b1eab3 144 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len);
63f70270
JF
145 else
146 /* Otherwise call the function; assume target could
147 clobber any caller-save reg */
ab144f5e
AK
148 ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY,
149 addr, clobbers, len);
63f70270
JF
150
151 return ret;
152}
153
ab144f5e 154unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
63f70270
JF
155 const char *start, const char *end)
156{
157 unsigned insn_len = end - start;
139ec7c4 158
63f70270
JF
159 if (insn_len > len || start == NULL)
160 insn_len = len;
161 else
ab144f5e 162 memcpy(insnbuf, start, insn_len);
139ec7c4 163
139ec7c4
RR
164 return insn_len;
165}
166
d3561b7f
RR
167void init_IRQ(void)
168{
93b1eab3 169 pv_irq_ops.init_IRQ();
d3561b7f
RR
170}
171
1a1eecd1 172static void native_flush_tlb(void)
da181a8b
RR
173{
174 __native_flush_tlb();
175}
176
177/*
178 * Global pages have to be flushed a bit differently. Not a real
179 * performance problem because this does not happen often.
180 */
1a1eecd1 181static void native_flush_tlb_global(void)
da181a8b
RR
182{
183 __native_flush_tlb_global();
184}
185
63f70270 186static void native_flush_tlb_single(unsigned long addr)
da181a8b
RR
187{
188 __native_flush_tlb_single(addr);
189}
190
d3561b7f 191/* These are in entry.S */
1a1eecd1 192extern void native_iret(void);
6abcd98f 193extern void native_irq_enable_syscall_ret(void);
d3561b7f
RR
194
195static int __init print_banner(void)
196{
93b1eab3 197 pv_init_ops.banner();
d3561b7f
RR
198 return 0;
199}
200core_initcall(print_banner);
201
d572929c
JF
202static struct resource reserve_ioports = {
203 .start = 0,
204 .end = IO_SPACE_LIMIT,
205 .name = "paravirt-ioport",
206 .flags = IORESOURCE_IO | IORESOURCE_BUSY,
207};
208
d572929c
JF
209/*
210 * Reserve the whole legacy IO space to prevent any legacy drivers
211 * from wasting time probing for their hardware. This is a fairly
212 * brute-force approach to disabling all non-virtual drivers.
213 *
214 * Note that this must be called very early to have any effect.
215 */
216int paravirt_disable_iospace(void)
217{
f7743fe6 218 return request_resource(&ioport_resource, &reserve_ioports);
d572929c
JF
219}
220
8965c1c0
JF
221static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
222
223static inline void enter_lazy(enum paravirt_lazy_mode mode)
224{
bfd074e0 225 BUG_ON(__get_cpu_var(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
8965c1c0
JF
226 BUG_ON(preemptible());
227
bfd074e0 228 __get_cpu_var(paravirt_lazy_mode) = mode;
8965c1c0
JF
229}
230
231void paravirt_leave_lazy(enum paravirt_lazy_mode mode)
232{
bfd074e0 233 BUG_ON(__get_cpu_var(paravirt_lazy_mode) != mode);
8965c1c0
JF
234 BUG_ON(preemptible());
235
bfd074e0 236 __get_cpu_var(paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
8965c1c0
JF
237}
238
239void paravirt_enter_lazy_mmu(void)
240{
241 enter_lazy(PARAVIRT_LAZY_MMU);
242}
243
244void paravirt_leave_lazy_mmu(void)
245{
246 paravirt_leave_lazy(PARAVIRT_LAZY_MMU);
247}
248
249void paravirt_enter_lazy_cpu(void)
250{
251 enter_lazy(PARAVIRT_LAZY_CPU);
252}
253
254void paravirt_leave_lazy_cpu(void)
255{
256 paravirt_leave_lazy(PARAVIRT_LAZY_CPU);
257}
258
259enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
260{
bfd074e0 261 return __get_cpu_var(paravirt_lazy_mode);
8965c1c0
JF
262}
263
93b1eab3 264struct pv_info pv_info = {
d3561b7f
RR
265 .name = "bare hardware",
266 .paravirt_enabled = 0,
267 .kernel_rpl = 0,
5311ab62 268 .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
93b1eab3 269};
d3561b7f 270
93b1eab3
JF
271struct pv_init_ops pv_init_ops = {
272 .patch = native_patch,
d3561b7f 273 .banner = default_banner,
45876233 274 .arch_setup = paravirt_nop,
d3561b7f 275 .memory_setup = machine_specific_memory_setup,
93b1eab3
JF
276};
277
278struct pv_time_ops pv_time_ops = {
279 .time_init = hpet_time_init,
d3561b7f
RR
280 .get_wallclock = native_get_wallclock,
281 .set_wallclock = native_set_wallclock,
93b1eab3
JF
282 .sched_clock = native_sched_clock,
283 .get_cpu_khz = native_calculate_cpu_khz,
284};
285
286struct pv_irq_ops pv_irq_ops = {
d3561b7f 287 .init_IRQ = native_init_IRQ,
93b1eab3
JF
288 .save_fl = native_save_fl,
289 .restore_fl = native_restore_fl,
290 .irq_disable = native_irq_disable,
291 .irq_enable = native_irq_enable,
292 .safe_halt = native_safe_halt,
293 .halt = native_halt,
294};
d3561b7f 295
93b1eab3 296struct pv_cpu_ops pv_cpu_ops = {
d3561b7f
RR
297 .cpuid = native_cpuid,
298 .get_debugreg = native_get_debugreg,
299 .set_debugreg = native_set_debugreg,
300 .clts = native_clts,
301 .read_cr0 = native_read_cr0,
302 .write_cr0 = native_write_cr0,
d3561b7f
RR
303 .read_cr4 = native_read_cr4,
304 .read_cr4_safe = native_read_cr4_safe,
305 .write_cr4 = native_write_cr4,
88b4755f
GOC
306#ifdef CONFIG_X86_64
307 .read_cr8 = native_read_cr8,
308 .write_cr8 = native_write_cr8,
309#endif
d3561b7f 310 .wbinvd = native_wbinvd,
90a0a06a
RR
311 .read_msr = native_read_msr_safe,
312 .write_msr = native_write_msr_safe,
d3561b7f
RR
313 .read_tsc = native_read_tsc,
314 .read_pmc = native_read_pmc,
e5aaac44 315 .read_tscp = native_read_tscp,
d3561b7f
RR
316 .load_tr_desc = native_load_tr_desc,
317 .set_ldt = native_set_ldt,
318 .load_gdt = native_load_gdt,
319 .load_idt = native_load_idt,
320 .store_gdt = native_store_gdt,
321 .store_idt = native_store_idt,
322 .store_tr = native_store_tr,
323 .load_tls = native_load_tls,
75b8bb3e 324 .write_ldt_entry = native_write_ldt_entry,
014b15be 325 .write_gdt_entry = native_write_gdt_entry,
8d947344 326 .write_idt_entry = native_write_idt_entry,
faca6227 327 .load_sp0 = native_load_sp0,
d3561b7f 328
6abcd98f 329 .irq_enable_syscall_ret = native_irq_enable_syscall_ret,
93b1eab3 330 .iret = native_iret,
e801f864 331 .swapgs = native_swapgs,
93b1eab3 332
d3561b7f
RR
333 .set_iopl_mask = native_set_iopl_mask,
334 .io_delay = native_io_delay,
8965c1c0
JF
335
336 .lazy_mode = {
337 .enter = paravirt_nop,
338 .leave = paravirt_nop,
339 },
93b1eab3 340};
d3561b7f 341
93b1eab3 342struct pv_apic_ops pv_apic_ops = {
13623d79
RR
343#ifdef CONFIG_X86_LOCAL_APIC
344 .apic_write = native_apic_write,
345 .apic_write_atomic = native_apic_write_atomic,
346 .apic_read = native_apic_read,
bbab4f3b
ZA
347 .setup_boot_clock = setup_boot_APIC_clock,
348 .setup_secondary_clock = setup_secondary_APIC_clock,
0260c196 349 .startup_ipi_hook = paravirt_nop,
13623d79 350#endif
93b1eab3
JF
351};
352
93b1eab3 353struct pv_mmu_ops pv_mmu_ops = {
d8dd8eec 354#ifndef CONFIG_X86_64
b239fb25
JF
355 .pagetable_setup_start = native_pagetable_setup_start,
356 .pagetable_setup_done = native_pagetable_setup_done,
d8dd8eec 357#endif
b239fb25 358
93b1eab3
JF
359 .read_cr2 = native_read_cr2,
360 .write_cr2 = native_write_cr2,
361 .read_cr3 = native_read_cr3,
362 .write_cr3 = native_write_cr3,
363
da181a8b
RR
364 .flush_tlb_user = native_flush_tlb,
365 .flush_tlb_kernel = native_flush_tlb_global,
366 .flush_tlb_single = native_flush_tlb_single,
d4c10477 367 .flush_tlb_others = native_flush_tlb_others,
da181a8b 368
6944a9c8
JF
369 .alloc_pte = paravirt_nop,
370 .alloc_pmd = paravirt_nop,
371 .alloc_pmd_clone = paravirt_nop,
2761fa09 372 .alloc_pud = paravirt_nop,
6944a9c8
JF
373 .release_pte = paravirt_nop,
374 .release_pmd = paravirt_nop,
2761fa09 375 .release_pud = paravirt_nop,
c119ecce 376
da181a8b
RR
377 .set_pte = native_set_pte,
378 .set_pte_at = native_set_pte_at,
379 .set_pmd = native_set_pmd,
45876233
JF
380 .pte_update = paravirt_nop,
381 .pte_update_defer = paravirt_nop,
3dc494e8 382
08b882c6
JF
383 .ptep_modify_prot_start = __ptep_modify_prot_start,
384 .ptep_modify_prot_commit = __ptep_modify_prot_commit,
385
ce6234b5
JF
386#ifdef CONFIG_HIGHPTE
387 .kmap_atomic_pte = kmap_atomic,
388#endif
389
f95f2f7b 390#if PAGETABLE_LEVELS >= 3
da181a8b
RR
391#ifdef CONFIG_X86_PAE
392 .set_pte_atomic = native_set_pte_atomic,
393 .set_pte_present = native_set_pte_present,
da181a8b
RR
394 .pte_clear = native_pte_clear,
395 .pmd_clear = native_pmd_clear,
f95f2f7b
EH
396#endif
397 .set_pud = native_set_pud,
3dc494e8
JF
398 .pmd_val = native_pmd_val,
399 .make_pmd = native_make_pmd,
f95f2f7b
EH
400
401#if PAGETABLE_LEVELS == 4
402 .pud_val = native_pud_val,
403 .make_pud = native_make_pud,
404 .set_pgd = native_set_pgd,
da181a8b 405#endif
f95f2f7b 406#endif /* PAGETABLE_LEVELS >= 3 */
da181a8b 407
3dc494e8 408 .pte_val = native_pte_val,
a15af1c9 409 .pte_flags = native_pte_val,
3dc494e8
JF
410 .pgd_val = native_pgd_val,
411
412 .make_pte = native_make_pte,
413 .make_pgd = native_make_pgd,
414
d6dd61c8
JF
415 .dup_mmap = paravirt_nop,
416 .exit_mmap = paravirt_nop,
417 .activate_mm = paravirt_nop,
8965c1c0
JF
418
419 .lazy_mode = {
420 .enter = paravirt_nop,
421 .leave = paravirt_nop,
422 },
d3561b7f 423};
0dbe5a11 424
93b1eab3 425EXPORT_SYMBOL_GPL(pv_time_ops);
f97b8954
JF
426EXPORT_SYMBOL (pv_cpu_ops);
427EXPORT_SYMBOL (pv_mmu_ops);
93b1eab3
JF
428EXPORT_SYMBOL_GPL(pv_apic_ops);
429EXPORT_SYMBOL_GPL(pv_info);
430EXPORT_SYMBOL (pv_irq_ops);