x86: Fix alternatives and kprobes to remap write-protected kernel text
[linux-2.6.git] / arch / i386 / 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 #include <linux/errno.h>
19 #include <linux/module.h>
20 #include <linux/efi.h>
21 #include <linux/bcd.h>
22 #include <linux/highmem.h>
23
24 #include <asm/bug.h>
25 #include <asm/paravirt.h>
26 #include <asm/desc.h>
27 #include <asm/setup.h>
28 #include <asm/arch_hooks.h>
29 #include <asm/time.h>
30 #include <asm/irq.h>
31 #include <asm/delay.h>
32 #include <asm/fixmap.h>
33 #include <asm/apic.h>
34 #include <asm/tlbflush.h>
35 #include <asm/timer.h>
36
37 /* nop stub */
38 void _paravirt_nop(void)
39 {
40 }
41
42 static void __init default_banner(void)
43 {
44         printk(KERN_INFO "Booting paravirtualized kernel on %s\n",
45                paravirt_ops.name);
46 }
47
48 char *memory_setup(void)
49 {
50         return paravirt_ops.memory_setup();
51 }
52
53 /* Simple instruction patching code. */
54 #define DEF_NATIVE(name, code)                                  \
55         extern const char start_##name[], end_##name[];         \
56         asm("start_" #name ": " code "; end_" #name ":")
57
58 DEF_NATIVE(irq_disable, "cli");
59 DEF_NATIVE(irq_enable, "sti");
60 DEF_NATIVE(restore_fl, "push %eax; popf");
61 DEF_NATIVE(save_fl, "pushf; pop %eax");
62 DEF_NATIVE(iret, "iret");
63 DEF_NATIVE(irq_enable_sysexit, "sti; sysexit");
64 DEF_NATIVE(read_cr2, "mov %cr2, %eax");
65 DEF_NATIVE(write_cr3, "mov %eax, %cr3");
66 DEF_NATIVE(read_cr3, "mov %cr3, %eax");
67 DEF_NATIVE(clts, "clts");
68 DEF_NATIVE(read_tsc, "rdtsc");
69
70 DEF_NATIVE(ud2a, "ud2a");
71
72 static unsigned native_patch(u8 type, u16 clobbers, void *insns, unsigned len)
73 {
74         const unsigned char *start, *end;
75         unsigned ret;
76
77         switch(type) {
78 #define SITE(x) case PARAVIRT_PATCH(x): start = start_##x; end = end_##x; goto patch_site
79                 SITE(irq_disable);
80                 SITE(irq_enable);
81                 SITE(restore_fl);
82                 SITE(save_fl);
83                 SITE(iret);
84                 SITE(irq_enable_sysexit);
85                 SITE(read_cr2);
86                 SITE(read_cr3);
87                 SITE(write_cr3);
88                 SITE(clts);
89                 SITE(read_tsc);
90 #undef SITE
91
92         patch_site:
93                 ret = paravirt_patch_insns(insns, len, start, end);
94                 break;
95
96         case PARAVIRT_PATCH(make_pgd):
97         case PARAVIRT_PATCH(make_pte):
98         case PARAVIRT_PATCH(pgd_val):
99         case PARAVIRT_PATCH(pte_val):
100 #ifdef CONFIG_X86_PAE
101         case PARAVIRT_PATCH(make_pmd):
102         case PARAVIRT_PATCH(pmd_val):
103 #endif
104                 /* These functions end up returning exactly what
105                    they're passed, in the same registers. */
106                 ret = paravirt_patch_nop();
107                 break;
108
109         default:
110                 ret = paravirt_patch_default(type, clobbers, insns, len);
111                 break;
112         }
113
114         return ret;
115 }
116
117 unsigned paravirt_patch_nop(void)
118 {
119         return 0;
120 }
121
122 unsigned paravirt_patch_ignore(unsigned len)
123 {
124         return len;
125 }
126
127 struct branch {
128         unsigned char opcode;
129         u32 delta;
130 } __attribute__((packed));
131
132 unsigned paravirt_patch_call(void *target, u16 tgt_clobbers,
133                              void *site, u16 site_clobbers,
134                              unsigned len)
135 {
136         unsigned char *call = site;
137         unsigned long delta = (unsigned long)target - (unsigned long)(call+5);
138         struct branch b;
139
140         if (tgt_clobbers & ~site_clobbers)
141                 return len;     /* target would clobber too much for this site */
142         if (len < 5)
143                 return len;     /* call too long for patch site */
144
145         b.opcode = 0xe8; /* call */
146         b.delta = delta;
147         BUILD_BUG_ON(sizeof(b) != 5);
148         text_poke(call, (unsigned char *)&b, 5);
149
150         return 5;
151 }
152
153 unsigned paravirt_patch_jmp(void *target, void *site, unsigned len)
154 {
155         unsigned char *jmp = site;
156         unsigned long delta = (unsigned long)target - (unsigned long)(jmp+5);
157
158         if (len < 5)
159                 return len;     /* call too long for patch site */
160
161         b.opcode = 0xe9;        /* jmp */
162         b.delta = delta;
163         text_poke(call, (unsigned char *)&b, 5);
164
165         return 5;
166 }
167
168 unsigned paravirt_patch_default(u8 type, u16 clobbers, void *site, unsigned len)
169 {
170         void *opfunc = *((void **)&paravirt_ops + type);
171         unsigned ret;
172
173         if (opfunc == NULL)
174                 /* If there's no function, patch it with a ud2a (BUG) */
175                 ret = paravirt_patch_insns(site, len, start_ud2a, end_ud2a);
176         else if (opfunc == paravirt_nop)
177                 /* If the operation is a nop, then nop the callsite */
178                 ret = paravirt_patch_nop();
179         else if (type == PARAVIRT_PATCH(iret) ||
180                  type == PARAVIRT_PATCH(irq_enable_sysexit))
181                 /* If operation requires a jmp, then jmp */
182                 ret = paravirt_patch_jmp(opfunc, site, len);
183         else
184                 /* Otherwise call the function; assume target could
185                    clobber any caller-save reg */
186                 ret = paravirt_patch_call(opfunc, CLBR_ANY,
187                                           site, clobbers, len);
188
189         return ret;
190 }
191
192 unsigned paravirt_patch_insns(void *site, unsigned len,
193                               const char *start, const char *end)
194 {
195         unsigned insn_len = end - start;
196
197         if (insn_len > len || start == NULL)
198                 insn_len = len;
199         else
200                 memcpy(site, start, insn_len);
201
202         return insn_len;
203 }
204
205 void init_IRQ(void)
206 {
207         paravirt_ops.init_IRQ();
208 }
209
210 static void native_flush_tlb(void)
211 {
212         __native_flush_tlb();
213 }
214
215 /*
216  * Global pages have to be flushed a bit differently. Not a real
217  * performance problem because this does not happen often.
218  */
219 static void native_flush_tlb_global(void)
220 {
221         __native_flush_tlb_global();
222 }
223
224 static void native_flush_tlb_single(unsigned long addr)
225 {
226         __native_flush_tlb_single(addr);
227 }
228
229 /* These are in entry.S */
230 extern void native_iret(void);
231 extern void native_irq_enable_sysexit(void);
232
233 static int __init print_banner(void)
234 {
235         paravirt_ops.banner();
236         return 0;
237 }
238 core_initcall(print_banner);
239
240 static struct resource reserve_ioports = {
241         .start = 0,
242         .end = IO_SPACE_LIMIT,
243         .name = "paravirt-ioport",
244         .flags = IORESOURCE_IO | IORESOURCE_BUSY,
245 };
246
247 static struct resource reserve_iomem = {
248         .start = 0,
249         .end = -1,
250         .name = "paravirt-iomem",
251         .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
252 };
253
254 /*
255  * Reserve the whole legacy IO space to prevent any legacy drivers
256  * from wasting time probing for their hardware.  This is a fairly
257  * brute-force approach to disabling all non-virtual drivers.
258  *
259  * Note that this must be called very early to have any effect.
260  */
261 int paravirt_disable_iospace(void)
262 {
263         int ret;
264
265         ret = request_resource(&ioport_resource, &reserve_ioports);
266         if (ret == 0) {
267                 ret = request_resource(&iomem_resource, &reserve_iomem);
268                 if (ret)
269                         release_resource(&reserve_ioports);
270         }
271
272         return ret;
273 }
274
275 struct paravirt_ops paravirt_ops = {
276         .name = "bare hardware",
277         .paravirt_enabled = 0,
278         .kernel_rpl = 0,
279         .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */
280
281         .patch = native_patch,
282         .banner = default_banner,
283         .arch_setup = paravirt_nop,
284         .memory_setup = machine_specific_memory_setup,
285         .get_wallclock = native_get_wallclock,
286         .set_wallclock = native_set_wallclock,
287         .time_init = hpet_time_init,
288         .init_IRQ = native_init_IRQ,
289
290         .cpuid = native_cpuid,
291         .get_debugreg = native_get_debugreg,
292         .set_debugreg = native_set_debugreg,
293         .clts = native_clts,
294         .read_cr0 = native_read_cr0,
295         .write_cr0 = native_write_cr0,
296         .read_cr2 = native_read_cr2,
297         .write_cr2 = native_write_cr2,
298         .read_cr3 = native_read_cr3,
299         .write_cr3 = native_write_cr3,
300         .read_cr4 = native_read_cr4,
301         .read_cr4_safe = native_read_cr4_safe,
302         .write_cr4 = native_write_cr4,
303         .save_fl = native_save_fl,
304         .restore_fl = native_restore_fl,
305         .irq_disable = native_irq_disable,
306         .irq_enable = native_irq_enable,
307         .safe_halt = native_safe_halt,
308         .halt = native_halt,
309         .wbinvd = native_wbinvd,
310         .read_msr = native_read_msr_safe,
311         .write_msr = native_write_msr_safe,
312         .read_tsc = native_read_tsc,
313         .read_pmc = native_read_pmc,
314         .sched_clock = native_sched_clock,
315         .get_cpu_khz = native_calculate_cpu_khz,
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,
324         .write_ldt_entry = write_dt_entry,
325         .write_gdt_entry = write_dt_entry,
326         .write_idt_entry = write_dt_entry,
327         .load_esp0 = native_load_esp0,
328
329         .set_iopl_mask = native_set_iopl_mask,
330         .io_delay = native_io_delay,
331
332 #ifdef CONFIG_X86_LOCAL_APIC
333         .apic_write = native_apic_write,
334         .apic_write_atomic = native_apic_write_atomic,
335         .apic_read = native_apic_read,
336         .setup_boot_clock = setup_boot_APIC_clock,
337         .setup_secondary_clock = setup_secondary_APIC_clock,
338         .startup_ipi_hook = paravirt_nop,
339 #endif
340         .set_lazy_mode = paravirt_nop,
341
342         .pagetable_setup_start = native_pagetable_setup_start,
343         .pagetable_setup_done = native_pagetable_setup_done,
344
345         .flush_tlb_user = native_flush_tlb,
346         .flush_tlb_kernel = native_flush_tlb_global,
347         .flush_tlb_single = native_flush_tlb_single,
348         .flush_tlb_others = native_flush_tlb_others,
349
350         .alloc_pt = paravirt_nop,
351         .alloc_pd = paravirt_nop,
352         .alloc_pd_clone = paravirt_nop,
353         .release_pt = paravirt_nop,
354         .release_pd = paravirt_nop,
355
356         .set_pte = native_set_pte,
357         .set_pte_at = native_set_pte_at,
358         .set_pmd = native_set_pmd,
359         .pte_update = paravirt_nop,
360         .pte_update_defer = paravirt_nop,
361
362 #ifdef CONFIG_HIGHPTE
363         .kmap_atomic_pte = kmap_atomic,
364 #endif
365
366 #ifdef CONFIG_X86_PAE
367         .set_pte_atomic = native_set_pte_atomic,
368         .set_pte_present = native_set_pte_present,
369         .set_pud = native_set_pud,
370         .pte_clear = native_pte_clear,
371         .pmd_clear = native_pmd_clear,
372
373         .pmd_val = native_pmd_val,
374         .make_pmd = native_make_pmd,
375 #endif
376
377         .pte_val = native_pte_val,
378         .pgd_val = native_pgd_val,
379
380         .make_pte = native_make_pte,
381         .make_pgd = native_make_pgd,
382
383         .irq_enable_sysexit = native_irq_enable_sysexit,
384         .iret = native_iret,
385
386         .dup_mmap = paravirt_nop,
387         .exit_mmap = paravirt_nop,
388         .activate_mm = paravirt_nop,
389 };
390
391 EXPORT_SYMBOL(paravirt_ops);