Commit | Line | Data |
---|---|---|
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 | |
17 | */ | |
18 | #include <linux/errno.h> | |
19 | #include <linux/module.h> | |
20 | #include <linux/efi.h> | |
21 | #include <linux/bcd.h> | |
ce6234b5 | 22 | #include <linux/highmem.h> |
d3561b7f RR |
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> | |
13623d79 RR |
32 | #include <asm/fixmap.h> |
33 | #include <asm/apic.h> | |
da181a8b | 34 | #include <asm/tlbflush.h> |
6cb9a835 | 35 | #include <asm/timer.h> |
d3561b7f RR |
36 | |
37 | /* nop stub */ | |
45876233 | 38 | void _paravirt_nop(void) |
d3561b7f RR |
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 | ||
139ec7c4 RR |
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 ":") | |
63f70270 JF |
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"); | |
139ec7c4 | 62 | DEF_NATIVE(iret, "iret"); |
63f70270 JF |
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"); | |
139ec7c4 | 69 | |
63f70270 | 70 | DEF_NATIVE(ud2a, "ud2a"); |
139ec7c4 | 71 | |
ab144f5e AK |
72 | static unsigned native_patch(u8 type, u16 clobbers, void *ibuf, |
73 | unsigned long addr, unsigned len) | |
139ec7c4 | 74 | { |
63f70270 JF |
75 | const unsigned char *start, *end; |
76 | unsigned ret; | |
77 | ||
78 | switch(type) { | |
79 | #define SITE(x) case PARAVIRT_PATCH(x): start = start_##x; end = end_##x; goto patch_site | |
80 | SITE(irq_disable); | |
81 | SITE(irq_enable); | |
82 | SITE(restore_fl); | |
83 | SITE(save_fl); | |
84 | SITE(iret); | |
85 | SITE(irq_enable_sysexit); | |
86 | SITE(read_cr2); | |
87 | SITE(read_cr3); | |
88 | SITE(write_cr3); | |
89 | SITE(clts); | |
90 | SITE(read_tsc); | |
91 | #undef SITE | |
92 | ||
93 | patch_site: | |
ab144f5e | 94 | ret = paravirt_patch_insns(ibuf, len, start, end); |
63f70270 JF |
95 | break; |
96 | ||
97 | case PARAVIRT_PATCH(make_pgd): | |
98 | case PARAVIRT_PATCH(make_pte): | |
99 | case PARAVIRT_PATCH(pgd_val): | |
100 | case PARAVIRT_PATCH(pte_val): | |
101 | #ifdef CONFIG_X86_PAE | |
102 | case PARAVIRT_PATCH(make_pmd): | |
103 | case PARAVIRT_PATCH(pmd_val): | |
104 | #endif | |
105 | /* These functions end up returning exactly what | |
106 | they're passed, in the same registers. */ | |
107 | ret = paravirt_patch_nop(); | |
108 | break; | |
109 | ||
110 | default: | |
ab144f5e | 111 | ret = paravirt_patch_default(type, clobbers, ibuf, addr, len); |
63f70270 JF |
112 | break; |
113 | } | |
114 | ||
115 | return ret; | |
116 | } | |
117 | ||
118 | unsigned paravirt_patch_nop(void) | |
119 | { | |
120 | return 0; | |
121 | } | |
122 | ||
123 | unsigned paravirt_patch_ignore(unsigned len) | |
124 | { | |
125 | return len; | |
126 | } | |
127 | ||
19d36ccd AK |
128 | struct branch { |
129 | unsigned char opcode; | |
130 | u32 delta; | |
131 | } __attribute__((packed)); | |
132 | ||
ab144f5e AK |
133 | unsigned paravirt_patch_call(void *insnbuf, |
134 | const void *target, u16 tgt_clobbers, | |
135 | unsigned long addr, u16 site_clobbers, | |
63f70270 JF |
136 | unsigned len) |
137 | { | |
ab144f5e AK |
138 | struct branch *b = insnbuf; |
139 | unsigned long delta = (unsigned long)target - (addr+5); | |
63f70270 JF |
140 | |
141 | if (tgt_clobbers & ~site_clobbers) | |
142 | return len; /* target would clobber too much for this site */ | |
143 | if (len < 5) | |
144 | return len; /* call too long for patch site */ | |
139ec7c4 | 145 | |
ab144f5e AK |
146 | b->opcode = 0xe8; /* call */ |
147 | b->delta = delta; | |
148 | BUILD_BUG_ON(sizeof(*b) != 5); | |
139ec7c4 | 149 | |
63f70270 JF |
150 | return 5; |
151 | } | |
152 | ||
ab144f5e AK |
153 | unsigned paravirt_patch_jmp(const void *target, void *insnbuf, |
154 | unsigned long addr, unsigned len) | |
63f70270 | 155 | { |
ab144f5e AK |
156 | struct branch *b = insnbuf; |
157 | unsigned long delta = (unsigned long)target - (addr+5); | |
63f70270 JF |
158 | |
159 | if (len < 5) | |
160 | return len; /* call too long for patch site */ | |
161 | ||
ab144f5e AK |
162 | b->opcode = 0xe9; /* jmp */ |
163 | b->delta = delta; | |
63f70270 JF |
164 | |
165 | return 5; | |
166 | } | |
167 | ||
ab144f5e AK |
168 | unsigned paravirt_patch_default(u8 type, u16 clobbers, void *insnbuf, |
169 | unsigned long addr, unsigned len) | |
63f70270 JF |
170 | { |
171 | void *opfunc = *((void **)¶virt_ops + type); | |
172 | unsigned ret; | |
173 | ||
174 | if (opfunc == NULL) | |
175 | /* If there's no function, patch it with a ud2a (BUG) */ | |
ab144f5e | 176 | ret = paravirt_patch_insns(insnbuf, len, start_ud2a, end_ud2a); |
63f70270 JF |
177 | else if (opfunc == paravirt_nop) |
178 | /* If the operation is a nop, then nop the callsite */ | |
179 | ret = paravirt_patch_nop(); | |
180 | else if (type == PARAVIRT_PATCH(iret) || | |
181 | type == PARAVIRT_PATCH(irq_enable_sysexit)) | |
182 | /* If operation requires a jmp, then jmp */ | |
ab144f5e | 183 | ret = paravirt_patch_jmp(opfunc, insnbuf, addr, len); |
63f70270 JF |
184 | else |
185 | /* Otherwise call the function; assume target could | |
186 | clobber any caller-save reg */ | |
ab144f5e AK |
187 | ret = paravirt_patch_call(insnbuf, opfunc, CLBR_ANY, |
188 | addr, clobbers, len); | |
63f70270 JF |
189 | |
190 | return ret; | |
191 | } | |
192 | ||
ab144f5e | 193 | unsigned paravirt_patch_insns(void *insnbuf, unsigned len, |
63f70270 JF |
194 | const char *start, const char *end) |
195 | { | |
196 | unsigned insn_len = end - start; | |
139ec7c4 | 197 | |
63f70270 JF |
198 | if (insn_len > len || start == NULL) |
199 | insn_len = len; | |
200 | else | |
ab144f5e | 201 | memcpy(insnbuf, start, insn_len); |
139ec7c4 | 202 | |
139ec7c4 RR |
203 | return insn_len; |
204 | } | |
205 | ||
d3561b7f RR |
206 | void init_IRQ(void) |
207 | { | |
208 | paravirt_ops.init_IRQ(); | |
209 | } | |
210 | ||
1a1eecd1 | 211 | static void native_flush_tlb(void) |
da181a8b RR |
212 | { |
213 | __native_flush_tlb(); | |
214 | } | |
215 | ||
216 | /* | |
217 | * Global pages have to be flushed a bit differently. Not a real | |
218 | * performance problem because this does not happen often. | |
219 | */ | |
1a1eecd1 | 220 | static void native_flush_tlb_global(void) |
da181a8b RR |
221 | { |
222 | __native_flush_tlb_global(); | |
223 | } | |
224 | ||
63f70270 | 225 | static void native_flush_tlb_single(unsigned long addr) |
da181a8b RR |
226 | { |
227 | __native_flush_tlb_single(addr); | |
228 | } | |
229 | ||
d3561b7f | 230 | /* These are in entry.S */ |
1a1eecd1 AK |
231 | extern void native_iret(void); |
232 | extern void native_irq_enable_sysexit(void); | |
d3561b7f RR |
233 | |
234 | static int __init print_banner(void) | |
235 | { | |
236 | paravirt_ops.banner(); | |
237 | return 0; | |
238 | } | |
239 | core_initcall(print_banner); | |
240 | ||
d572929c JF |
241 | static struct resource reserve_ioports = { |
242 | .start = 0, | |
243 | .end = IO_SPACE_LIMIT, | |
244 | .name = "paravirt-ioport", | |
245 | .flags = IORESOURCE_IO | IORESOURCE_BUSY, | |
246 | }; | |
247 | ||
248 | static struct resource reserve_iomem = { | |
249 | .start = 0, | |
250 | .end = -1, | |
251 | .name = "paravirt-iomem", | |
252 | .flags = IORESOURCE_MEM | IORESOURCE_BUSY, | |
253 | }; | |
254 | ||
255 | /* | |
256 | * Reserve the whole legacy IO space to prevent any legacy drivers | |
257 | * from wasting time probing for their hardware. This is a fairly | |
258 | * brute-force approach to disabling all non-virtual drivers. | |
259 | * | |
260 | * Note that this must be called very early to have any effect. | |
261 | */ | |
262 | int paravirt_disable_iospace(void) | |
263 | { | |
264 | int ret; | |
265 | ||
266 | ret = request_resource(&ioport_resource, &reserve_ioports); | |
267 | if (ret == 0) { | |
268 | ret = request_resource(&iomem_resource, &reserve_iomem); | |
269 | if (ret) | |
270 | release_resource(&reserve_ioports); | |
271 | } | |
272 | ||
273 | return ret; | |
274 | } | |
275 | ||
d3561b7f RR |
276 | struct paravirt_ops paravirt_ops = { |
277 | .name = "bare hardware", | |
278 | .paravirt_enabled = 0, | |
279 | .kernel_rpl = 0, | |
5311ab62 | 280 | .shared_kernel_pmd = 1, /* Only used when CONFIG_X86_PAE is set */ |
d3561b7f | 281 | |
139ec7c4 | 282 | .patch = native_patch, |
d3561b7f | 283 | .banner = default_banner, |
45876233 | 284 | .arch_setup = paravirt_nop, |
d3561b7f RR |
285 | .memory_setup = machine_specific_memory_setup, |
286 | .get_wallclock = native_get_wallclock, | |
287 | .set_wallclock = native_set_wallclock, | |
e30fab3a | 288 | .time_init = hpet_time_init, |
d3561b7f RR |
289 | .init_IRQ = native_init_IRQ, |
290 | ||
291 | .cpuid = native_cpuid, | |
292 | .get_debugreg = native_get_debugreg, | |
293 | .set_debugreg = native_set_debugreg, | |
294 | .clts = native_clts, | |
295 | .read_cr0 = native_read_cr0, | |
296 | .write_cr0 = native_write_cr0, | |
297 | .read_cr2 = native_read_cr2, | |
298 | .write_cr2 = native_write_cr2, | |
299 | .read_cr3 = native_read_cr3, | |
300 | .write_cr3 = native_write_cr3, | |
301 | .read_cr4 = native_read_cr4, | |
302 | .read_cr4_safe = native_read_cr4_safe, | |
303 | .write_cr4 = native_write_cr4, | |
304 | .save_fl = native_save_fl, | |
305 | .restore_fl = native_restore_fl, | |
306 | .irq_disable = native_irq_disable, | |
307 | .irq_enable = native_irq_enable, | |
308 | .safe_halt = native_safe_halt, | |
309 | .halt = native_halt, | |
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, | |
688340ea | 315 | .sched_clock = native_sched_clock, |
1182d852 | 316 | .get_cpu_khz = native_calculate_cpu_khz, |
d3561b7f RR |
317 | .load_tr_desc = native_load_tr_desc, |
318 | .set_ldt = native_set_ldt, | |
319 | .load_gdt = native_load_gdt, | |
320 | .load_idt = native_load_idt, | |
321 | .store_gdt = native_store_gdt, | |
322 | .store_idt = native_store_idt, | |
323 | .store_tr = native_store_tr, | |
324 | .load_tls = native_load_tls, | |
90a0a06a RR |
325 | .write_ldt_entry = write_dt_entry, |
326 | .write_gdt_entry = write_dt_entry, | |
327 | .write_idt_entry = write_dt_entry, | |
d3561b7f RR |
328 | .load_esp0 = native_load_esp0, |
329 | ||
330 | .set_iopl_mask = native_set_iopl_mask, | |
331 | .io_delay = native_io_delay, | |
d3561b7f | 332 | |
13623d79 RR |
333 | #ifdef CONFIG_X86_LOCAL_APIC |
334 | .apic_write = native_apic_write, | |
335 | .apic_write_atomic = native_apic_write_atomic, | |
336 | .apic_read = native_apic_read, | |
bbab4f3b ZA |
337 | .setup_boot_clock = setup_boot_APIC_clock, |
338 | .setup_secondary_clock = setup_secondary_APIC_clock, | |
0260c196 | 339 | .startup_ipi_hook = paravirt_nop, |
13623d79 | 340 | #endif |
45876233 | 341 | .set_lazy_mode = paravirt_nop, |
13623d79 | 342 | |
b239fb25 JF |
343 | .pagetable_setup_start = native_pagetable_setup_start, |
344 | .pagetable_setup_done = native_pagetable_setup_done, | |
345 | ||
da181a8b RR |
346 | .flush_tlb_user = native_flush_tlb, |
347 | .flush_tlb_kernel = native_flush_tlb_global, | |
348 | .flush_tlb_single = native_flush_tlb_single, | |
d4c10477 | 349 | .flush_tlb_others = native_flush_tlb_others, |
da181a8b | 350 | |
45876233 JF |
351 | .alloc_pt = paravirt_nop, |
352 | .alloc_pd = paravirt_nop, | |
353 | .alloc_pd_clone = paravirt_nop, | |
354 | .release_pt = paravirt_nop, | |
355 | .release_pd = paravirt_nop, | |
c119ecce | 356 | |
da181a8b RR |
357 | .set_pte = native_set_pte, |
358 | .set_pte_at = native_set_pte_at, | |
359 | .set_pmd = native_set_pmd, | |
45876233 JF |
360 | .pte_update = paravirt_nop, |
361 | .pte_update_defer = paravirt_nop, | |
3dc494e8 | 362 | |
ce6234b5 JF |
363 | #ifdef CONFIG_HIGHPTE |
364 | .kmap_atomic_pte = kmap_atomic, | |
365 | #endif | |
366 | ||
da181a8b RR |
367 | #ifdef CONFIG_X86_PAE |
368 | .set_pte_atomic = native_set_pte_atomic, | |
369 | .set_pte_present = native_set_pte_present, | |
370 | .set_pud = native_set_pud, | |
371 | .pte_clear = native_pte_clear, | |
372 | .pmd_clear = native_pmd_clear, | |
3dc494e8 JF |
373 | |
374 | .pmd_val = native_pmd_val, | |
375 | .make_pmd = native_make_pmd, | |
da181a8b RR |
376 | #endif |
377 | ||
3dc494e8 JF |
378 | .pte_val = native_pte_val, |
379 | .pgd_val = native_pgd_val, | |
380 | ||
381 | .make_pte = native_make_pte, | |
382 | .make_pgd = native_make_pgd, | |
383 | ||
d3561b7f RR |
384 | .irq_enable_sysexit = native_irq_enable_sysexit, |
385 | .iret = native_iret, | |
ae5da273 | 386 | |
d6dd61c8 JF |
387 | .dup_mmap = paravirt_nop, |
388 | .exit_mmap = paravirt_nop, | |
389 | .activate_mm = paravirt_nop, | |
d3561b7f | 390 | }; |
0dbe5a11 | 391 | |
21564fd2 | 392 | EXPORT_SYMBOL(paravirt_ops); |