Merge branch 'master' into for-next
[pandora-kernel.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
27 #include <asm/bug.h>
28 #include <asm/paravirt.h>
29 #include <asm/desc.h>
30 #include <asm/setup.h>
31 #include <asm/pgtable.h>
32 #include <asm/time.h>
33 #include <asm/pgalloc.h>
34 #include <asm/irq.h>
35 #include <asm/delay.h>
36 #include <asm/fixmap.h>
37 #include <asm/apic.h>
38 #include <asm/tlbflush.h>
39 #include <asm/timer.h>
40
41 /* nop stub */
42 void _paravirt_nop(void)
43 {
44 }
45
46 /* identity function, which can be inlined */
47 u32 _paravirt_ident_32(u32 x)
48 {
49         return x;
50 }
51
52 u64 _paravirt_ident_64(u64 x)
53 {
54         return x;
55 }
56
57 void __init default_banner(void)
58 {
59         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
60                pv_info.name);
61 }
62
63 /* Simple instruction patching code. */
64 #define DEF_NATIVE(ops, name, code)                                     \
65         extern const char start_##ops##_##name[], end_##ops##_##name[]; \
66         asm("start_" #ops "_" #name ": " code "; end_" #ops "_" #name ":")
67
68 /* Undefined instruction for dealing with missing ops pointers. */
69 static const unsigned char ud2a[] = { 0x0f, 0x0b };
70
71 unsigned paravirt_patch_nop(void)
72 {
73         return 0;
74 }
75
76 unsigned paravirt_patch_ignore(unsigned len)
77 {
78         return len;
79 }
80
81 struct branch {
82         unsigned char opcode;
83         u32 delta;
84 } __attribute__((packed));
85
86 unsigned paravirt_patch_call(void *insnbuf,
87                              const void *target, u16 tgt_clobbers,
88                              unsigned long addr, u16 site_clobbers,
89                              unsigned len)
90 {
91         struct branch *b = insnbuf;
92         unsigned long delta = (unsigned long)target - (addr+5);
93
94         if (tgt_clobbers & ~site_clobbers)
95                 return len;     /* target would clobber too much for this site */
96         if (len < 5)
97                 return len;     /* call too long for patch site */
98
99         b->opcode = 0xe8; /* call */
100         b->delta = delta;
101         BUILD_BUG_ON(sizeof(*b) != 5);
102
103         return 5;
104 }
105
106 unsigned paravirt_patch_jmp(void *insnbuf, const void *target,
107                             unsigned long addr, unsigned len)
108 {
109         struct branch *b = insnbuf;
110         unsigned long delta = (unsigned long)target - (addr+5);
111
112         if (len < 5)
113                 return len;     /* call too long for patch site */
114
115         b->opcode = 0xe9;       /* jmp */
116         b->delta = delta;
117
118         return 5;
119 }
120
121 /* Neat trick to map patch type back to the call within the
122  * corresponding structure. */
123 static void *get_call_destination(u8 type)
124 {
125         struct paravirt_patch_template tmpl = {
126                 .pv_init_ops = pv_init_ops,
127                 .pv_time_ops = pv_time_ops,
128                 .pv_cpu_ops = pv_cpu_ops,
129                 .pv_irq_ops = pv_irq_ops,
130                 .pv_apic_ops = pv_apic_ops,
131                 .pv_mmu_ops = pv_mmu_ops,
132 #ifdef CONFIG_PARAVIRT_SPINLOCKS
133                 .pv_lock_ops = pv_lock_ops,
134 #endif
135         };
136         return *((void **)&tmpl + type);
137 }
138
139 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf,
140                                 unsigned long addr, unsigned len)
141 {
142         void *opfunc = get_call_destination(type);
143         unsigned ret;
144
145         if (opfunc == NULL)
146                 /* If there's no function, patch it with a ud2a (BUG) */
147                 ret = paravirt_patch_insns(insnbuf, len, ud2a, ud2a+sizeof(ud2a));
148         else if (opfunc == _paravirt_nop)
149                 /* If the operation is a nop, then nop the callsite */
150                 ret = paravirt_patch_nop();
151
152         /* identity functions just return their single argument */
153         else if (opfunc == _paravirt_ident_32)
154                 ret = paravirt_patch_ident_32(insnbuf, len);
155         else if (opfunc == _paravirt_ident_64)
156                 ret = paravirt_patch_ident_64(insnbuf, len);
157
158         else if (type == PARAVIRT_PATCH(pv_cpu_ops.iret) ||
159                  type == PARAVIRT_PATCH(pv_cpu_ops.irq_enable_sysexit) ||
160                  type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret32) ||
161                  type == PARAVIRT_PATCH(pv_cpu_ops.usergs_sysret64))
162                 /* If operation requires a jmp, then jmp */
163                 ret = paravirt_patch_jmp(insnbuf, opfunc, addr, len);
164         else
165                 /* Otherwise call the function; assume target could
166                    clobber any caller-save reg */
167                 ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY,
168                                           addr, clobbers, len);
169
170         return ret;
171 }
172
173 unsigned paravirt_patch_insns(void *insnbuf, unsigned len,
174                               const char *start, const char *end)
175 {
176         unsigned insn_len = end - start;
177
178         if (insn_len > len || start == NULL)
179                 insn_len = len;
180         else
181                 memcpy(insnbuf, start, insn_len);
182
183         return insn_len;
184 }
185
186 static void native_flush_tlb(void)
187 {
188         __native_flush_tlb();
189 }
190
191 /*
192  * Global pages have to be flushed a bit differently. Not a real
193  * performance problem because this does not happen often.
194  */
195 static void native_flush_tlb_global(void)
196 {
197         __native_flush_tlb_global();
198 }
199
200 static void native_flush_tlb_single(unsigned long addr)
201 {
202         __native_flush_tlb_single(addr);
203 }
204
205 /* These are in entry.S */
206 extern void native_iret(void);
207 extern void native_irq_enable_sysexit(void);
208 extern void native_usergs_sysret32(void);
209 extern void native_usergs_sysret64(void);
210
211 static struct resource reserve_ioports = {
212         .start = 0,
213         .end = IO_SPACE_LIMIT,
214         .name = "paravirt-ioport",
215         .flags = IORESOURCE_IO | IORESOURCE_BUSY,
216 };
217
218 /*
219  * Reserve the whole legacy IO space to prevent any legacy drivers
220  * from wasting time probing for their hardware.  This is a fairly
221  * brute-force approach to disabling all non-virtual drivers.
222  *
223  * Note that this must be called very early to have any effect.
224  */
225 int paravirt_disable_iospace(void)
226 {
227         return request_resource(&ioport_resource, &reserve_ioports);
228 }
229
230 static DEFINE_PER_CPU(enum paravirt_lazy_mode, paravirt_lazy_mode) = PARAVIRT_LAZY_NONE;
231
232 static inline void enter_lazy(enum paravirt_lazy_mode mode)
233 {
234         BUG_ON(percpu_read(paravirt_lazy_mode) != PARAVIRT_LAZY_NONE);
235
236         percpu_write(paravirt_lazy_mode, mode);
237 }
238
239 static void leave_lazy(enum paravirt_lazy_mode mode)
240 {
241         BUG_ON(percpu_read(paravirt_lazy_mode) != mode);
242
243         percpu_write(paravirt_lazy_mode, PARAVIRT_LAZY_NONE);
244 }
245
246 void paravirt_enter_lazy_mmu(void)
247 {
248         enter_lazy(PARAVIRT_LAZY_MMU);
249 }
250
251 void paravirt_leave_lazy_mmu(void)
252 {
253         leave_lazy(PARAVIRT_LAZY_MMU);
254 }
255
256 void paravirt_start_context_switch(struct task_struct *prev)
257 {
258         BUG_ON(preemptible());
259
260         if (percpu_read(paravirt_lazy_mode) == PARAVIRT_LAZY_MMU) {
261                 arch_leave_lazy_mmu_mode();
262                 set_ti_thread_flag(task_thread_info(prev), TIF_LAZY_MMU_UPDATES);
263         }
264         enter_lazy(PARAVIRT_LAZY_CPU);
265 }
266
267 void paravirt_end_context_switch(struct task_struct *next)
268 {
269         BUG_ON(preemptible());
270
271         leave_lazy(PARAVIRT_LAZY_CPU);
272
273         if (test_and_clear_ti_thread_flag(task_thread_info(next), TIF_LAZY_MMU_UPDATES))
274                 arch_enter_lazy_mmu_mode();
275 }
276
277 enum paravirt_lazy_mode paravirt_get_lazy_mode(void)
278 {
279         if (in_interrupt())
280                 return PARAVIRT_LAZY_NONE;
281
282         return percpu_read(paravirt_lazy_mode);
283 }
284
285 void arch_flush_lazy_mmu_mode(void)
286 {
287         preempt_disable();
288
289         if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_MMU) {
290                 arch_leave_lazy_mmu_mode();
291                 arch_enter_lazy_mmu_mode();
292         }
293
294         preempt_enable();
295 }
296
297 struct pv_info pv_info = {
298         .name = "bare hardware",
299         .paravirt_enabled = 0,
300         .kernel_rpl = 0,
301         .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
302 };
303
304 struct pv_init_ops pv_init_ops = {
305         .patch = native_patch,
306 };
307
308 struct pv_time_ops pv_time_ops = {
309         .sched_clock = native_sched_clock,
310 };
311
312 struct pv_irq_ops pv_irq_ops = {
313         .save_fl = __PV_IS_CALLEE_SAVE(native_save_fl),
314         .restore_fl = __PV_IS_CALLEE_SAVE(native_restore_fl),
315         .irq_disable = __PV_IS_CALLEE_SAVE(native_irq_disable),
316         .irq_enable = __PV_IS_CALLEE_SAVE(native_irq_enable),
317         .safe_halt = native_safe_halt,
318         .halt = native_halt,
319 #ifdef CONFIG_X86_64
320         .adjust_exception_frame = paravirt_nop,
321 #endif
322 };
323
324 struct pv_cpu_ops pv_cpu_ops = {
325         .cpuid = native_cpuid,
326         .get_debugreg = native_get_debugreg,
327         .set_debugreg = native_set_debugreg,
328         .clts = native_clts,
329         .read_cr0 = native_read_cr0,
330         .write_cr0 = native_write_cr0,
331         .read_cr4 = native_read_cr4,
332         .read_cr4_safe = native_read_cr4_safe,
333         .write_cr4 = native_write_cr4,
334 #ifdef CONFIG_X86_64
335         .read_cr8 = native_read_cr8,
336         .write_cr8 = native_write_cr8,
337 #endif
338         .wbinvd = native_wbinvd,
339         .read_msr = native_read_msr_safe,
340         .rdmsr_regs = native_rdmsr_safe_regs,
341         .write_msr = native_write_msr_safe,
342         .wrmsr_regs = native_wrmsr_safe_regs,
343         .read_tsc = native_read_tsc,
344         .read_pmc = native_read_pmc,
345         .read_tscp = native_read_tscp,
346         .load_tr_desc = native_load_tr_desc,
347         .set_ldt = native_set_ldt,
348         .load_gdt = native_load_gdt,
349         .load_idt = native_load_idt,
350         .store_gdt = native_store_gdt,
351         .store_idt = native_store_idt,
352         .store_tr = native_store_tr,
353         .load_tls = native_load_tls,
354 #ifdef CONFIG_X86_64
355         .load_gs_index = native_load_gs_index,
356 #endif
357         .write_ldt_entry = native_write_ldt_entry,
358         .write_gdt_entry = native_write_gdt_entry,
359         .write_idt_entry = native_write_idt_entry,
360
361         .alloc_ldt = paravirt_nop,
362         .free_ldt = paravirt_nop,
363
364         .load_sp0 = native_load_sp0,
365
366 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
367         .irq_enable_sysexit = native_irq_enable_sysexit,
368 #endif
369 #ifdef CONFIG_X86_64
370 #ifdef CONFIG_IA32_EMULATION
371         .usergs_sysret32 = native_usergs_sysret32,
372 #endif
373         .usergs_sysret64 = native_usergs_sysret64,
374 #endif
375         .iret = native_iret,
376         .swapgs = native_swapgs,
377
378         .set_iopl_mask = native_set_iopl_mask,
379         .io_delay = native_io_delay,
380
381         .start_context_switch = paravirt_nop,
382         .end_context_switch = paravirt_nop,
383 };
384
385 struct pv_apic_ops pv_apic_ops = {
386 #ifdef CONFIG_X86_LOCAL_APIC
387         .startup_ipi_hook = paravirt_nop,
388 #endif
389 };
390
391 #if defined(CONFIG_X86_32) && !defined(CONFIG_X86_PAE)
392 /* 32-bit pagetable entries */
393 #define PTE_IDENT       __PV_IS_CALLEE_SAVE(_paravirt_ident_32)
394 #else
395 /* 64-bit pagetable entries */
396 #define PTE_IDENT       __PV_IS_CALLEE_SAVE(_paravirt_ident_64)
397 #endif
398
399 struct pv_mmu_ops pv_mmu_ops = {
400
401         .read_cr2 = native_read_cr2,
402         .write_cr2 = native_write_cr2,
403         .read_cr3 = native_read_cr3,
404         .write_cr3 = native_write_cr3,
405
406         .flush_tlb_user = native_flush_tlb,
407         .flush_tlb_kernel = native_flush_tlb_global,
408         .flush_tlb_single = native_flush_tlb_single,
409         .flush_tlb_others = native_flush_tlb_others,
410
411         .pgd_alloc = __paravirt_pgd_alloc,
412         .pgd_free = paravirt_nop,
413
414         .alloc_pte = paravirt_nop,
415         .alloc_pmd = paravirt_nop,
416         .alloc_pud = paravirt_nop,
417         .release_pte = paravirt_nop,
418         .release_pmd = paravirt_nop,
419         .release_pud = paravirt_nop,
420
421         .set_pte = native_set_pte,
422         .set_pte_at = native_set_pte_at,
423         .set_pmd = native_set_pmd,
424         .set_pmd_at = native_set_pmd_at,
425         .pte_update = paravirt_nop,
426         .pte_update_defer = paravirt_nop,
427         .pmd_update = paravirt_nop,
428         .pmd_update_defer = paravirt_nop,
429
430         .ptep_modify_prot_start = __ptep_modify_prot_start,
431         .ptep_modify_prot_commit = __ptep_modify_prot_commit,
432
433 #if PAGETABLE_LEVELS >= 3
434 #ifdef CONFIG_X86_PAE
435         .set_pte_atomic = native_set_pte_atomic,
436         .pte_clear = native_pte_clear,
437         .pmd_clear = native_pmd_clear,
438 #endif
439         .set_pud = native_set_pud,
440
441         .pmd_val = PTE_IDENT,
442         .make_pmd = PTE_IDENT,
443
444 #if PAGETABLE_LEVELS == 4
445         .pud_val = PTE_IDENT,
446         .make_pud = PTE_IDENT,
447
448         .set_pgd = native_set_pgd,
449 #endif
450 #endif /* PAGETABLE_LEVELS >= 3 */
451
452         .pte_val = PTE_IDENT,
453         .pgd_val = PTE_IDENT,
454
455         .make_pte = PTE_IDENT,
456         .make_pgd = PTE_IDENT,
457
458         .dup_mmap = paravirt_nop,
459         .exit_mmap = paravirt_nop,
460         .activate_mm = paravirt_nop,
461
462         .lazy_mode = {
463                 .enter = paravirt_nop,
464                 .leave = paravirt_nop,
465         },
466
467         .set_fixmap = native_set_fixmap,
468 };
469
470 EXPORT_SYMBOL_GPL(pv_time_ops);
471 EXPORT_SYMBOL    (pv_cpu_ops);
472 EXPORT_SYMBOL    (pv_mmu_ops);
473 EXPORT_SYMBOL_GPL(pv_apic_ops);
474 EXPORT_SYMBOL_GPL(pv_info);
475 EXPORT_SYMBOL    (pv_irq_ops);