f27962cca0c74e3ef76a16bf4d0da34050e8f369
[cascardo/linux.git] / arch / x86 / kernel / paravirt.c
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
17
18     2007 - x86_64 support added by Glauber de Oliveira Costa, Red Hat Inc
19 */
20
21 #include <linux/errno.h>
22 #include <linux/module.h>
23 #include <linux/efi.h>
24 #include <linux/bcd.h>
25 #include <linux/highmem.h>
26 #include <linux/kprobes.h>
27
28 #include <asm/bug.h>
29 #include <asm/paravirt.h>
30 #include <asm/debugreg.h>
31 #include <asm/desc.h>
32 #include <asm/setup.h>
33 #include <asm/pgtable.h>
34 #include <asm/time.h>
35 #include <asm/pgalloc.h>
36 #include <asm/irq.h>
37 #include <asm/delay.h>
38 #include <asm/fixmap.h>
39 #include <asm/apic.h>
40 #include <asm/tlbflush.h>
41 #include <asm/timer.h>
42 #include <asm/special_insns.h>
43
44 /*
45  * nop stub, which must not clobber anything *including the stack* to
46  * avoid confusing the entry prologues.
47  */
48 extern void _paravirt_nop(void);
49 asm (".pushsection .entry.text, \"ax\"\n"
50      ".global _paravirt_nop\n"
51      "_paravirt_nop:\n\t"
52      "ret\n\t"
53      ".size _paravirt_nop, . - _paravirt_nop\n\t"
54      ".type _paravirt_nop, @function\n\t"
55      ".popsection");
56
57 /* identity function, which can be inlined */
58 u32 _paravirt_ident_32(u32 x)
59 {
60         return x;
61 }
62
63 u64 _paravirt_ident_64(u64 x)
64 {
65         return x;
66 }
67
68 void __init default_banner(void)
69 {
70         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
71                pv_info.name);
72 }
73
74 /* Undefined instruction for dealing with missing ops pointers. */
75 static const unsigned char ud2a[] = { 0x0f, 0x0b };
76
77 struct branch {
78         unsigned char opcode;
79         u32 delta;
80 } __attribute__((packed));
81
82 unsigned paravirt_patch_call(void *insnbuf,
83                              const void *target, u16 tgt_clobbers,
84                              unsigned long addr, u16 site_clobbers,
85                              unsigned len)
86 {
87         struct branch *b = insnbuf;
88         unsigned long delta = (unsigned long)target - (addr+5);
89
90         if (tgt_clobbers & ~site_clobbers)
91                 return len;     /* target would clobber too much for this site */
92         if (len < 5)
93                 return len;     /* call too long for patch site */
94
95         b->opcode = 0xe8; /* call */
96         b->delta = delta;
97         BUILD_BUG_ON(sizeof(*b) != 5);
98
99         return 5;
100 }
101
102 unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
103                             unsigned long addr, unsigned len)
104 {
105         struct branch *b = insnbuf;
106         unsigned long delta = (unsigned long)target - (addr+5);
107
108         if (len < 5)
109                 return len;     /* call too long for patch site */
110
111         b->opcode = 0xe9;       /* jmp */
112         b->delta = delta;
113
114         return 5;
115 }
116
117 /* Neat trick to map patch type back to the call within the
118  * corresponding structure. */
119 static void *get_call_destination(u8 type)
120 {
121         struct paravirt_patch_template tmpl = {
122                 .pv_init_ops = pv_init_ops,
123                 .pv_time_ops = pv_time_ops,
124                 .pv_cpu_ops = pv_cpu_ops,
125                 .pv_irq_ops = pv_irq_ops,
126                 .pv_mmu_ops = pv_mmu_ops,
127 #ifdef CONFIG_PARAVIRT_SPINLOCKS
128                 .pv_lock_ops = pv_lock_ops,
129 #endif
130         };
131         return *((void **)&tmpl + type);
132 }
133
134 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
135                                 unsigned long addr, unsigned len)
136 {
137         void *opfunc = get_call_destination(type);
138         unsigned ret;
139
140         if (opfunc == NULL)
141                 /* If there's no function, patch it with a ud2a (BUG) */
142                 ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a));
143         else if (opfunc == _paravirt_nop)
144                 ret = 0;
145
146         /* identity functions just return their single argument */
147         else if (opfunc == _paravirt_ident_32)
148                 ret = paravirt_patch_ident_32(insnbuf, len);
149         else if (opfunc == _paravirt_ident_64)
150                 ret = paravirt_patch_ident_64(insnbuf, len);
151
152         else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) ||
153 #ifdef CONFIG_X86_32
154                  type == PARAVIRT_PATCH(pv_cpu_ops.irq_enable_sysexit) ||
155 #endif
156                  type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret32) ||
157                  type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret64))
158                 /* If operation requires a jmp, then jmp */
159                 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len);
160         else
161                 /* Otherwise call the function; assume target could
162                    clobber any caller-save reg */
163                 ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY,
164                                           addr, clobbers, len);
165
166         return ret;
167 }
168
169 unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
170                               const char *start, const char *end)
171 {
172         unsigned insn_len = end - start;
173
174         if (insn_len > len || start == NULL)
175                 insn_len = len;
176         else
177                 memcpy(insnbuf, start, insn_len);
178
179         return insn_len;
180 }
181
182 static void native_flush_tlb(void)
183 {
184         __native_flush_tlb();
185 }
186
187 /*
188  * Global pages have to be flushed a bit differently. Not a real
189  * performance problem because this does not happen often.
190  */
191 static void native_flush_tlb_global(void)
192 {
193         __native_flush_tlb_global();
194 }
195
196 static void native_flush_tlb_single(unsigned long addr)
197 {
198         __native_flush_tlb_single(addr);
199 }
200
201 struct static_key paravirt_steal_enabled;
202 struct static_key paravirt_steal_rq_enabled;
203
204 static u64 native_steal_clock(int cpu)
205 {
206         return 0;
207 }
208
209 /* These are in entry.S */
210 extern void native_iret(void);
211 extern void native_irq_enable_sysexit(void);
212 extern void native_usergs_sysret32(void);
213 extern void native_usergs_sysret64(void);
214
215 static struct resource reserve_ioports = {
216         .start = 0,
217         .end = IO_SPACE_LIMIT,
218         .name = "paravirt-ioport",
219         .flags = IORESOURCE_IO | IORESOURCE_BUSY,
220 };
221
222 /*
223  * Reserve the whole legacy IO space to prevent any legacy drivers
224  * from wasting time probing for their hardware.  This is a fairly
225  * brute-force approach to disabling all non-virtual drivers.
226  *
227  * Note that this must be called very early to have any effect.
228  */
229 int paravirt_disable_iospace(void)
230 {
231         return request_resource(&ioport_resource, &reserve_ioports);
232 }
233
234 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
235
236 static inline void enter_lazy(enum paravirt_lazy_mode mode)
237 {
238         BUG_ON(this_cpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
239
240         this_cpu_write(paravirt_lazy_mode, mode);
241 }
242
243 static void leave_lazy(enum paravirt_lazy_mode mode)
244 {
245         BUG_ON(this_cpu_read(paravirt_lazy_mode) != mode);
246
247         this_cpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE);
248 }
249
250 void paravirt_enter_lazy_mmu(void)
251 {
252         enter_lazy(PARAVIRT_LAZY_MMU);
253 }
254
255 void paravirt_leave_lazy_mmu(void)
256 {
257         leave_lazy(PARAVIRT_LAZY_MMU);
258 }
259
260 void paravirt_flush_lazy_mmu(void)
261 {
262         preempt_disable();
263
264         if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
265                 arch_leave_lazy_mmu_mode();
266                 arch_enter_lazy_mmu_mode();
267         }
268
269         preempt_enable();
270 }
271
272 void paravirt_start_context_switch(struct task_struct *prev)
273 {
274         BUG_ON(preemptible());
275
276         if (this_cpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
277                 arch_leave_lazy_mmu_mode();
278                 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
279         }
280         enter_lazy(PARAVIRT_LAZY_CPU);
281 }
282
283 void paravirt_end_context_switch(struct task_struct *next)
284 {
285         BUG_ON(preemptible());
286
287         leave_lazy(PARAVIRT_LAZY_CPU);
288
289         if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
290                 arch_enter_lazy_mmu_mode();
291 }
292
293 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
294 {
295         if (in_interrupt())
296                 return PARAVIRT_LAZY_NONE;
297
298         return this_cpu_read(paravirt_lazy_mode);
299 }
300
301 struct pv_info pv_info = {
302         .name = "bare hardware",
303         .paravirt_enabled = 0,
304         .kernel_rpl = 0,
305         .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
306
307 #ifdef CONFIG_X86_64
308         .extra_user_64bit_cs = __USER_CS,
309 #endif
310 };
311
312 struct pv_init_ops pv_init_ops = {
313         .patch = native_patch,
314 };
315
316 struct pv_time_ops pv_time_ops = {
317         .sched_clock = native_sched_clock,
318         .steal_clock = native_steal_clock,
319 };
320
321 __visible struct pv_irq_ops pv_irq_ops = {
322         .save_fl = __PV_IS_CALLEE_SAVE(native_save_fl),
323         .restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl),
324         .irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable),
325         .irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable),
326         .safe_halt = native_safe_halt,
327         .halt = native_halt,
328 #ifdef CONFIG_X86_64
329         .adjust_exception_frame = paravirt_nop,
330 #endif
331 };
332
333 __visible struct pv_cpu_ops pv_cpu_ops = {
334         .cpuid = native_cpuid,
335         .get_debugreg = native_get_debugreg,
336         .set_debugreg = native_set_debugreg,
337         .clts = native_clts,
338         .read_cr0 = native_read_cr0,
339         .write_cr0 = native_write_cr0,
340         .read_cr4 = native_read_cr4,
341         .read_cr4_safe = native_read_cr4_safe,
342         .write_cr4 = native_write_cr4,
343 #ifdef CONFIG_X86_64
344         .read_cr8 = native_read_cr8,
345         .write_cr8 = native_write_cr8,
346 #endif
347         .wbinvd = native_wbinvd,
348         .read_msr = native_read_msr_safe,
349         .write_msr = native_write_msr_safe,
350         .read_pmc = native_read_pmc,
351         .load_tr_desc = native_load_tr_desc,
352         .set_ldt = native_set_ldt,
353         .load_gdt = native_load_gdt,
354         .load_idt = native_load_idt,
355         .store_idt = native_store_idt,
356         .store_tr = native_store_tr,
357         .load_tls = native_load_tls,
358 #ifdef CONFIG_X86_64
359         .load_gs_index = native_load_gs_index,
360 #endif
361         .write_ldt_entry = native_write_ldt_entry,
362         .write_gdt_entry = native_write_gdt_entry,
363         .write_idt_entry = native_write_idt_entry,
364
365         .alloc_ldt = paravirt_nop,
366         .free_ldt = paravirt_nop,
367
368         .load_sp0 = native_load_sp0,
369
370 #if defined(CONFIG_X86_32)
371         .irq_enable_sysexit = native_irq_enable_sysexit,
372 #endif
373 #ifdef CONFIG_X86_64
374 #ifdef CONFIG_IA32_EMULATION
375         .usergs_sysret32 = native_usergs_sysret32,
376 #endif
377         .usergs_sysret64 = native_usergs_sysret64,
378 #endif
379         .iret = native_iret,
380         .swapgs = native_swapgs,
381
382         .set_iopl_mask = native_set_iopl_mask,
383         .io_delay = native_io_delay,
384
385         .start_context_switch = paravirt_nop,
386         .end_context_switch = paravirt_nop,
387 };
388
389 /* At this point, native_get/set_debugreg has real function entries */
390 NOKPROBE_SYMBOL(native_get_debugreg);
391 NOKPROBE_SYMBOL(native_set_debugreg);
392 NOKPROBE_SYMBOL(native_load_idt);
393
394 #if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE)
395 /* 32-bit pagetable entries */
396 #define PTE_IDENT       __PV_IS_CALLEE_SAVE(_paravirt_ident_32)
397 #else
398 /* 64-bit pagetable entries */
399 #define PTE_IDENT       __PV_IS_CALLEE_SAVE(_paravirt_ident_64)
400 #endif
401
402 struct pv_mmu_ops pv_mmu_ops = {
403
404         .read_cr2 = native_read_cr2,
405         .write_cr2 = native_write_cr2,
406         .read_cr3 = native_read_cr3,
407         .write_cr3 = native_write_cr3,
408
409         .flush_tlb_user = native_flush_tlb,
410         .flush_tlb_kernel = native_flush_tlb_global,
411         .flush_tlb_single = native_flush_tlb_single,
412         .flush_tlb_others = native_flush_tlb_others,
413
414         .pgd_alloc = __paravirt_pgd_alloc,
415         .pgd_free = paravirt_nop,
416
417         .alloc_pte = paravirt_nop,
418         .alloc_pmd = paravirt_nop,
419         .alloc_pud = paravirt_nop,
420         .release_pte = paravirt_nop,
421         .release_pmd = paravirt_nop,
422         .release_pud = paravirt_nop,
423
424         .set_pte = native_set_pte,
425         .set_pte_at = native_set_pte_at,
426         .set_pmd = native_set_pmd,
427         .set_pmd_at = native_set_pmd_at,
428         .pte_update = paravirt_nop,
429         .pte_update_defer = paravirt_nop,
430         .pmd_update = paravirt_nop,
431         .pmd_update_defer = paravirt_nop,
432
433         .ptep_modify_prot_start = __ptep_modify_prot_start,
434         .ptep_modify_prot_commit = __ptep_modify_prot_commit,
435
436 #if CONFIG_PGTABLE_LEVELS >= 3
437 #ifdef CONFIG_X86_PAE
438         .set_pte_atomic = native_set_pte_atomic,
439         .pte_clear = native_pte_clear,
440         .pmd_clear = native_pmd_clear,
441 #endif
442         .set_pud = native_set_pud,
443
444         .pmd_val = PTE_IDENT,
445         .make_pmd = PTE_IDENT,
446
447 #if CONFIG_PGTABLE_LEVELS == 4
448         .pud_val = PTE_IDENT,
449         .make_pud = PTE_IDENT,
450
451         .set_pgd = native_set_pgd,
452 #endif
453 #endif /* CONFIG_PGTABLE_LEVELS >= 3 */
454
455         .pte_val = PTE_IDENT,
456         .pgd_val = PTE_IDENT,
457
458         .make_pte = PTE_IDENT,
459         .make_pgd = PTE_IDENT,
460
461         .dup_mmap = paravirt_nop,
462         .exit_mmap = paravirt_nop,
463         .activate_mm = paravirt_nop,
464
465         .lazy_mode = {
466                 .enter = paravirt_nop,
467                 .leave = paravirt_nop,
468                 .flush = paravirt_nop,
469         },
470
471         .set_fixmap = native_set_fixmap,
472 };
473
474 EXPORT_SYMBOL_GPL(pv_time_ops);
475 EXPORT_SYMBOL    (pv_cpu_ops);
476 EXPORT_SYMBOL    (pv_mmu_ops);
477 EXPORT_SYMBOL_GPL(pv_info);
478 EXPORT_SYMBOL    (pv_irq_ops);