Commit | Line | Data |
---|---|---|
7ce0bcfd ZA |
1 | /* |
2 | * VMI specific paravirt-ops implementation | |
3 | * | |
4 | * Copyright (C) 2005, VMware, Inc. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License as published by | |
8 | * the Free Software Foundation; either version 2 of the License, or | |
9 | * (at your option) any later version. | |
10 | * | |
11 | * This program is distributed in the hope that it will be useful, but | |
12 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or | |
14 | * NON INFRINGEMENT. See the GNU General Public License for more | |
15 | * details. | |
16 | * | |
17 | * You should have received a copy of the GNU General Public License | |
18 | * along with this program; if not, write to the Free Software | |
19 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
20 | * | |
21 | * Send feedback to zach@vmware.com | |
22 | * | |
23 | */ | |
24 | ||
25 | #include <linux/module.h> | |
7ce0bcfd ZA |
26 | #include <linux/cpu.h> |
27 | #include <linux/bootmem.h> | |
28 | #include <linux/mm.h> | |
eeef9c68 | 29 | #include <linux/highmem.h> |
7ce0bcfd ZA |
30 | #include <asm/vmi.h> |
31 | #include <asm/io.h> | |
32 | #include <asm/fixmap.h> | |
33 | #include <asm/apicdef.h> | |
34 | #include <asm/apic.h> | |
35 | #include <asm/processor.h> | |
36 | #include <asm/timer.h> | |
bbab4f3b | 37 | #include <asm/vmi_time.h> |
8f485612 | 38 | #include <asm/kmap_types.h> |
7ce0bcfd ZA |
39 | |
40 | /* Convenient for calling VMI functions indirectly in the ROM */ | |
41 | typedef u32 __attribute__((regparm(1))) (VROMFUNC)(void); | |
42 | typedef u64 __attribute__((regparm(2))) (VROMLONGFUNC)(int); | |
43 | ||
44 | #define call_vrom_func(rom,func) \ | |
45 | (((VROMFUNC *)(rom->func))()) | |
46 | ||
47 | #define call_vrom_long_func(rom,func,arg) \ | |
48 | (((VROMLONGFUNC *)(rom->func)) (arg)) | |
49 | ||
50 | static struct vrom_header *vmi_rom; | |
7ce0bcfd ZA |
51 | static int disable_pge; |
52 | static int disable_pse; | |
53 | static int disable_sep; | |
54 | static int disable_tsc; | |
55 | static int disable_mtrr; | |
7507ba34 | 56 | static int disable_noidle; |
772205f6 | 57 | static int disable_vmi_timer; |
7ce0bcfd ZA |
58 | |
59 | /* Cached VMI operations */ | |
30a1528d | 60 | static struct { |
7ce0bcfd ZA |
61 | void (*cpuid)(void /* non-c */); |
62 | void (*_set_ldt)(u32 selector); | |
63 | void (*set_tr)(u32 selector); | |
64 | void (*set_kernel_stack)(u32 selector, u32 esp0); | |
65 | void (*allocate_page)(u32, u32, u32, u32, u32); | |
66 | void (*release_page)(u32, u32); | |
67 | void (*set_pte)(pte_t, pte_t *, unsigned); | |
68 | void (*update_pte)(pte_t *, unsigned); | |
eeef9c68 ZA |
69 | void (*set_linear_mapping)(int, void *, u32, u32); |
70 | void (*_flush_tlb)(int); | |
7ce0bcfd | 71 | void (*set_initial_ap_state)(int, int); |
bbab4f3b | 72 | void (*halt)(void); |
49f19710 | 73 | void (*set_lazy_mode)(int mode); |
7ce0bcfd ZA |
74 | } vmi_ops; |
75 | ||
76 | /* XXX move this to alternative.h */ | |
77 | extern struct paravirt_patch __start_parainstructions[], | |
78 | __stop_parainstructions[]; | |
79 | ||
80 | /* | |
81 | * VMI patching routines. | |
82 | */ | |
83 | #define MNEM_CALL 0xe8 | |
84 | #define MNEM_JMP 0xe9 | |
85 | #define MNEM_RET 0xc3 | |
86 | ||
7ce0bcfd ZA |
87 | #define IRQ_PATCH_INT_MASK 0 |
88 | #define IRQ_PATCH_DISABLE 5 | |
89 | ||
90 | static inline void patch_offset(unsigned char *eip, unsigned char *dest) | |
91 | { | |
92 | *(unsigned long *)(eip+1) = dest-eip-5; | |
93 | } | |
94 | ||
95 | static unsigned patch_internal(int call, unsigned len, void *insns) | |
96 | { | |
97 | u64 reloc; | |
98 | struct vmi_relocation_info *const rel = (struct vmi_relocation_info *)&reloc; | |
99 | reloc = call_vrom_long_func(vmi_rom, get_reloc, call); | |
100 | switch(rel->type) { | |
101 | case VMI_RELOCATION_CALL_REL: | |
102 | BUG_ON(len < 5); | |
103 | *(char *)insns = MNEM_CALL; | |
104 | patch_offset(insns, rel->eip); | |
105 | return 5; | |
106 | ||
107 | case VMI_RELOCATION_JUMP_REL: | |
108 | BUG_ON(len < 5); | |
109 | *(char *)insns = MNEM_JMP; | |
110 | patch_offset(insns, rel->eip); | |
111 | return 5; | |
112 | ||
113 | case VMI_RELOCATION_NOP: | |
114 | /* obliterate the whole thing */ | |
115 | return 0; | |
116 | ||
117 | case VMI_RELOCATION_NONE: | |
118 | /* leave native code in place */ | |
119 | break; | |
120 | ||
121 | default: | |
122 | BUG(); | |
123 | } | |
124 | return len; | |
125 | } | |
126 | ||
127 | /* | |
128 | * Apply patch if appropriate, return length of new instruction | |
129 | * sequence. The callee does nop padding for us. | |
130 | */ | |
131 | static unsigned vmi_patch(u8 type, u16 clobbers, void *insns, unsigned len) | |
132 | { | |
133 | switch (type) { | |
d5822035 | 134 | case PARAVIRT_PATCH(irq_disable): |
7ce0bcfd | 135 | return patch_internal(VMI_CALL_DisableInterrupts, len, insns); |
d5822035 | 136 | case PARAVIRT_PATCH(irq_enable): |
7ce0bcfd | 137 | return patch_internal(VMI_CALL_EnableInterrupts, len, insns); |
d5822035 | 138 | case PARAVIRT_PATCH(restore_fl): |
7ce0bcfd | 139 | return patch_internal(VMI_CALL_SetInterruptMask, len, insns); |
d5822035 | 140 | case PARAVIRT_PATCH(save_fl): |
7ce0bcfd | 141 | return patch_internal(VMI_CALL_GetInterruptMask, len, insns); |
d5822035 | 142 | case PARAVIRT_PATCH(iret): |
7ce0bcfd | 143 | return patch_internal(VMI_CALL_IRET, len, insns); |
d5822035 | 144 | case PARAVIRT_PATCH(irq_enable_sysexit): |
7ce0bcfd ZA |
145 | return patch_internal(VMI_CALL_SYSEXIT, len, insns); |
146 | default: | |
147 | break; | |
148 | } | |
149 | return len; | |
150 | } | |
151 | ||
152 | /* CPUID has non-C semantics, and paravirt-ops API doesn't match hardware ISA */ | |
153 | static void vmi_cpuid(unsigned int *eax, unsigned int *ebx, | |
154 | unsigned int *ecx, unsigned int *edx) | |
155 | { | |
156 | int override = 0; | |
157 | if (*eax == 1) | |
158 | override = 1; | |
159 | asm volatile ("call *%6" | |
160 | : "=a" (*eax), | |
161 | "=b" (*ebx), | |
162 | "=c" (*ecx), | |
163 | "=d" (*edx) | |
164 | : "0" (*eax), "2" (*ecx), "r" (vmi_ops.cpuid)); | |
165 | if (override) { | |
166 | if (disable_pse) | |
167 | *edx &= ~X86_FEATURE_PSE; | |
168 | if (disable_pge) | |
169 | *edx &= ~X86_FEATURE_PGE; | |
170 | if (disable_sep) | |
171 | *edx &= ~X86_FEATURE_SEP; | |
172 | if (disable_tsc) | |
173 | *edx &= ~X86_FEATURE_TSC; | |
174 | if (disable_mtrr) | |
175 | *edx &= ~X86_FEATURE_MTRR; | |
176 | } | |
177 | } | |
178 | ||
179 | static inline void vmi_maybe_load_tls(struct desc_struct *gdt, int nr, struct desc_struct *new) | |
180 | { | |
181 | if (gdt[nr].a != new->a || gdt[nr].b != new->b) | |
182 | write_gdt_entry(gdt, nr, new->a, new->b); | |
183 | } | |
184 | ||
185 | static void vmi_load_tls(struct thread_struct *t, unsigned int cpu) | |
186 | { | |
187 | struct desc_struct *gdt = get_cpu_gdt_table(cpu); | |
188 | vmi_maybe_load_tls(gdt, GDT_ENTRY_TLS_MIN + 0, &t->tls_array[0]); | |
189 | vmi_maybe_load_tls(gdt, GDT_ENTRY_TLS_MIN + 1, &t->tls_array[1]); | |
190 | vmi_maybe_load_tls(gdt, GDT_ENTRY_TLS_MIN + 2, &t->tls_array[2]); | |
191 | } | |
192 | ||
193 | static void vmi_set_ldt(const void *addr, unsigned entries) | |
194 | { | |
195 | unsigned cpu = smp_processor_id(); | |
196 | u32 low, high; | |
197 | ||
198 | pack_descriptor(&low, &high, (unsigned long)addr, | |
199 | entries * sizeof(struct desc_struct) - 1, | |
200 | DESCTYPE_LDT, 0); | |
201 | write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_LDT, low, high); | |
202 | vmi_ops._set_ldt(entries ? GDT_ENTRY_LDT*sizeof(struct desc_struct) : 0); | |
203 | } | |
204 | ||
205 | static void vmi_set_tr(void) | |
206 | { | |
207 | vmi_ops.set_tr(GDT_ENTRY_TSS*sizeof(struct desc_struct)); | |
208 | } | |
209 | ||
210 | static void vmi_load_esp0(struct tss_struct *tss, | |
211 | struct thread_struct *thread) | |
212 | { | |
a75c54f9 | 213 | tss->x86_tss.esp0 = thread->esp0; |
7ce0bcfd ZA |
214 | |
215 | /* This can only happen when SEP is enabled, no need to test "SEP"arately */ | |
a75c54f9 RR |
216 | if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) { |
217 | tss->x86_tss.ss1 = thread->sysenter_cs; | |
7ce0bcfd ZA |
218 | wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0); |
219 | } | |
a75c54f9 | 220 | vmi_ops.set_kernel_stack(__KERNEL_DS, tss->x86_tss.esp0); |
7ce0bcfd ZA |
221 | } |
222 | ||
223 | static void vmi_flush_tlb_user(void) | |
224 | { | |
eeef9c68 | 225 | vmi_ops._flush_tlb(VMI_FLUSH_TLB); |
7ce0bcfd ZA |
226 | } |
227 | ||
228 | static void vmi_flush_tlb_kernel(void) | |
229 | { | |
eeef9c68 | 230 | vmi_ops._flush_tlb(VMI_FLUSH_TLB | VMI_FLUSH_GLOBAL); |
7ce0bcfd ZA |
231 | } |
232 | ||
233 | /* Stub to do nothing at all; used for delays and unimplemented calls */ | |
234 | static void vmi_nop(void) | |
235 | { | |
236 | } | |
237 | ||
bbab4f3b | 238 | /* For NO_IDLE_HZ, we stop the clock when halting the kernel */ |
bbab4f3b ZA |
239 | static fastcall void vmi_safe_halt(void) |
240 | { | |
241 | int idle = vmi_stop_hz_timer(); | |
242 | vmi_ops.halt(); | |
243 | if (idle) { | |
244 | local_irq_disable(); | |
245 | vmi_account_time_restart_hz_timer(); | |
246 | local_irq_enable(); | |
247 | } | |
248 | } | |
7ce0bcfd ZA |
249 | |
250 | #ifdef CONFIG_DEBUG_PAGE_TYPE | |
251 | ||
252 | #ifdef CONFIG_X86_PAE | |
253 | #define MAX_BOOT_PTS (2048+4+1) | |
254 | #else | |
255 | #define MAX_BOOT_PTS (1024+1) | |
256 | #endif | |
257 | ||
258 | /* | |
259 | * During boot, mem_map is not yet available in paging_init, so stash | |
260 | * all the boot page allocations here. | |
261 | */ | |
262 | static struct { | |
263 | u32 pfn; | |
264 | int type; | |
265 | } boot_page_allocations[MAX_BOOT_PTS]; | |
266 | static int num_boot_page_allocations; | |
267 | static int boot_allocations_applied; | |
268 | ||
269 | void vmi_apply_boot_page_allocations(void) | |
270 | { | |
271 | int i; | |
272 | BUG_ON(!mem_map); | |
273 | for (i = 0; i < num_boot_page_allocations; i++) { | |
274 | struct page *page = pfn_to_page(boot_page_allocations[i].pfn); | |
275 | page->type = boot_page_allocations[i].type; | |
276 | page->type = boot_page_allocations[i].type & | |
277 | ~(VMI_PAGE_ZEROED | VMI_PAGE_CLONE); | |
278 | } | |
279 | boot_allocations_applied = 1; | |
280 | } | |
281 | ||
282 | static void record_page_type(u32 pfn, int type) | |
283 | { | |
284 | BUG_ON(num_boot_page_allocations >= MAX_BOOT_PTS); | |
285 | boot_page_allocations[num_boot_page_allocations].pfn = pfn; | |
286 | boot_page_allocations[num_boot_page_allocations].type = type; | |
287 | num_boot_page_allocations++; | |
288 | } | |
289 | ||
290 | static void check_zeroed_page(u32 pfn, int type, struct page *page) | |
291 | { | |
292 | u32 *ptr; | |
293 | int i; | |
294 | int limit = PAGE_SIZE / sizeof(int); | |
295 | ||
296 | if (page_address(page)) | |
297 | ptr = (u32 *)page_address(page); | |
298 | else | |
299 | ptr = (u32 *)__va(pfn << PAGE_SHIFT); | |
300 | /* | |
301 | * When cloning the root in non-PAE mode, only the userspace | |
302 | * pdes need to be zeroed. | |
303 | */ | |
304 | if (type & VMI_PAGE_CLONE) | |
305 | limit = USER_PTRS_PER_PGD; | |
306 | for (i = 0; i < limit; i++) | |
307 | BUG_ON(ptr[i]); | |
308 | } | |
309 | ||
310 | /* | |
311 | * We stash the page type into struct page so we can verify the page | |
312 | * types are used properly. | |
313 | */ | |
314 | static void vmi_set_page_type(u32 pfn, int type) | |
315 | { | |
316 | /* PAE can have multiple roots per page - don't track */ | |
317 | if (PTRS_PER_PMD > 1 && (type & VMI_PAGE_PDP)) | |
318 | return; | |
319 | ||
320 | if (boot_allocations_applied) { | |
321 | struct page *page = pfn_to_page(pfn); | |
322 | if (type != VMI_PAGE_NORMAL) | |
323 | BUG_ON(page->type); | |
324 | else | |
325 | BUG_ON(page->type == VMI_PAGE_NORMAL); | |
326 | page->type = type & ~(VMI_PAGE_ZEROED | VMI_PAGE_CLONE); | |
327 | if (type & VMI_PAGE_ZEROED) | |
328 | check_zeroed_page(pfn, type, page); | |
329 | } else { | |
330 | record_page_type(pfn, type); | |
331 | } | |
332 | } | |
333 | ||
334 | static void vmi_check_page_type(u32 pfn, int type) | |
335 | { | |
336 | /* PAE can have multiple roots per page - skip checks */ | |
337 | if (PTRS_PER_PMD > 1 && (type & VMI_PAGE_PDP)) | |
338 | return; | |
339 | ||
340 | type &= ~(VMI_PAGE_ZEROED | VMI_PAGE_CLONE); | |
341 | if (boot_allocations_applied) { | |
342 | struct page *page = pfn_to_page(pfn); | |
343 | BUG_ON((page->type ^ type) & VMI_PAGE_PAE); | |
344 | BUG_ON(type == VMI_PAGE_NORMAL && page->type); | |
345 | BUG_ON((type & page->type) == 0); | |
346 | } | |
347 | } | |
348 | #else | |
349 | #define vmi_set_page_type(p,t) do { } while (0) | |
350 | #define vmi_check_page_type(p,t) do { } while (0) | |
351 | #endif | |
352 | ||
eeef9c68 ZA |
353 | #ifdef CONFIG_HIGHPTE |
354 | static void *vmi_kmap_atomic_pte(struct page *page, enum km_type type) | |
9a1c13e9 | 355 | { |
eeef9c68 ZA |
356 | void *va = kmap_atomic(page, type); |
357 | ||
9a1c13e9 ZA |
358 | /* |
359 | * Internally, the VMI ROM must map virtual addresses to physical | |
360 | * addresses for processing MMU updates. By the time MMU updates | |
361 | * are issued, this information is typically already lost. | |
362 | * Fortunately, the VMI provides a cache of mapping slots for active | |
363 | * page tables. | |
364 | * | |
365 | * We use slot zero for the linear mapping of physical memory, and | |
366 | * in HIGHPTE kernels, slot 1 and 2 for KM_PTE0 and KM_PTE1. | |
367 | * | |
368 | * args: SLOT VA COUNT PFN | |
369 | */ | |
370 | BUG_ON(type != KM_PTE0 && type != KM_PTE1); | |
eeef9c68 ZA |
371 | vmi_ops.set_linear_mapping((type - KM_PTE0)+1, va, 1, page_to_pfn(page)); |
372 | ||
373 | return va; | |
9a1c13e9 | 374 | } |
eeef9c68 | 375 | #endif |
9a1c13e9 | 376 | |
7ce0bcfd ZA |
377 | static void vmi_allocate_pt(u32 pfn) |
378 | { | |
379 | vmi_set_page_type(pfn, VMI_PAGE_L1); | |
380 | vmi_ops.allocate_page(pfn, VMI_PAGE_L1, 0, 0, 0); | |
381 | } | |
382 | ||
383 | static void vmi_allocate_pd(u32 pfn) | |
384 | { | |
385 | /* | |
386 | * This call comes in very early, before mem_map is setup. | |
387 | * It is called only for swapper_pg_dir, which already has | |
388 | * data on it. | |
389 | */ | |
390 | vmi_set_page_type(pfn, VMI_PAGE_L2); | |
391 | vmi_ops.allocate_page(pfn, VMI_PAGE_L2, 0, 0, 0); | |
392 | } | |
393 | ||
394 | static void vmi_allocate_pd_clone(u32 pfn, u32 clonepfn, u32 start, u32 count) | |
395 | { | |
396 | vmi_set_page_type(pfn, VMI_PAGE_L2 | VMI_PAGE_CLONE); | |
397 | vmi_check_page_type(clonepfn, VMI_PAGE_L2); | |
398 | vmi_ops.allocate_page(pfn, VMI_PAGE_L2 | VMI_PAGE_CLONE, clonepfn, start, count); | |
399 | } | |
400 | ||
401 | static void vmi_release_pt(u32 pfn) | |
402 | { | |
403 | vmi_ops.release_page(pfn, VMI_PAGE_L1); | |
404 | vmi_set_page_type(pfn, VMI_PAGE_NORMAL); | |
405 | } | |
406 | ||
407 | static void vmi_release_pd(u32 pfn) | |
408 | { | |
409 | vmi_ops.release_page(pfn, VMI_PAGE_L2); | |
410 | vmi_set_page_type(pfn, VMI_PAGE_NORMAL); | |
411 | } | |
412 | ||
413 | /* | |
414 | * Helper macros for MMU update flags. We can defer updates until a flush | |
415 | * or page invalidation only if the update is to the current address space | |
416 | * (otherwise, there is no flush). We must check against init_mm, since | |
417 | * this could be a kernel update, which usually passes init_mm, although | |
418 | * sometimes this check can be skipped if we know the particular function | |
419 | * is only called on user mode PTEs. We could change the kernel to pass | |
420 | * current->active_mm here, but in particular, I was unsure if changing | |
421 | * mm/highmem.c to do this would still be correct on other architectures. | |
422 | */ | |
423 | #define is_current_as(mm, mustbeuser) ((mm) == current->active_mm || \ | |
424 | (!mustbeuser && (mm) == &init_mm)) | |
425 | #define vmi_flags_addr(mm, addr, level, user) \ | |
426 | ((level) | (is_current_as(mm, user) ? \ | |
427 | (VMI_PAGE_CURRENT_AS | ((addr) & VMI_PAGE_VA_MASK)) : 0)) | |
428 | #define vmi_flags_addr_defer(mm, addr, level, user) \ | |
429 | ((level) | (is_current_as(mm, user) ? \ | |
430 | (VMI_PAGE_DEFER | VMI_PAGE_CURRENT_AS | ((addr) & VMI_PAGE_VA_MASK)) : 0)) | |
431 | ||
3dc494e8 | 432 | static void vmi_update_pte(struct mm_struct *mm, unsigned long addr, pte_t *ptep) |
7ce0bcfd ZA |
433 | { |
434 | vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE); | |
435 | vmi_ops.update_pte(ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0)); | |
436 | } | |
437 | ||
3dc494e8 | 438 | static void vmi_update_pte_defer(struct mm_struct *mm, unsigned long addr, pte_t *ptep) |
7ce0bcfd ZA |
439 | { |
440 | vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE); | |
441 | vmi_ops.update_pte(ptep, vmi_flags_addr_defer(mm, addr, VMI_PAGE_PT, 0)); | |
442 | } | |
443 | ||
444 | static void vmi_set_pte(pte_t *ptep, pte_t pte) | |
445 | { | |
446 | /* XXX because of set_pmd_pte, this can be called on PT or PD layers */ | |
447 | vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE | VMI_PAGE_PD); | |
448 | vmi_ops.set_pte(pte, ptep, VMI_PAGE_PT); | |
449 | } | |
450 | ||
3dc494e8 | 451 | static void vmi_set_pte_at(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pte) |
7ce0bcfd ZA |
452 | { |
453 | vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE); | |
454 | vmi_ops.set_pte(pte, ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0)); | |
455 | } | |
456 | ||
457 | static void vmi_set_pmd(pmd_t *pmdp, pmd_t pmdval) | |
458 | { | |
459 | #ifdef CONFIG_X86_PAE | |
460 | const pte_t pte = { pmdval.pmd, pmdval.pmd >> 32 }; | |
461 | vmi_check_page_type(__pa(pmdp) >> PAGE_SHIFT, VMI_PAGE_PMD); | |
462 | #else | |
463 | const pte_t pte = { pmdval.pud.pgd.pgd }; | |
464 | vmi_check_page_type(__pa(pmdp) >> PAGE_SHIFT, VMI_PAGE_PGD); | |
465 | #endif | |
466 | vmi_ops.set_pte(pte, (pte_t *)pmdp, VMI_PAGE_PD); | |
467 | } | |
468 | ||
469 | #ifdef CONFIG_X86_PAE | |
470 | ||
471 | static void vmi_set_pte_atomic(pte_t *ptep, pte_t pteval) | |
472 | { | |
473 | /* | |
474 | * XXX This is called from set_pmd_pte, but at both PT | |
475 | * and PD layers so the VMI_PAGE_PT flag is wrong. But | |
476 | * it is only called for large page mapping changes, | |
477 | * the Xen backend, doesn't support large pages, and the | |
478 | * ESX backend doesn't depend on the flag. | |
479 | */ | |
480 | set_64bit((unsigned long long *)ptep,pte_val(pteval)); | |
481 | vmi_ops.update_pte(ptep, VMI_PAGE_PT); | |
482 | } | |
483 | ||
484 | static void vmi_set_pte_present(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pte) | |
485 | { | |
486 | vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE); | |
487 | vmi_ops.set_pte(pte, ptep, vmi_flags_addr_defer(mm, addr, VMI_PAGE_PT, 1)); | |
488 | } | |
489 | ||
490 | static void vmi_set_pud(pud_t *pudp, pud_t pudval) | |
491 | { | |
492 | /* Um, eww */ | |
493 | const pte_t pte = { pudval.pgd.pgd, pudval.pgd.pgd >> 32 }; | |
494 | vmi_check_page_type(__pa(pudp) >> PAGE_SHIFT, VMI_PAGE_PGD); | |
495 | vmi_ops.set_pte(pte, (pte_t *)pudp, VMI_PAGE_PDP); | |
496 | } | |
497 | ||
498 | static void vmi_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep) | |
499 | { | |
500 | const pte_t pte = { 0 }; | |
501 | vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE); | |
502 | vmi_ops.set_pte(pte, ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0)); | |
503 | } | |
504 | ||
8eb68fae | 505 | static void vmi_pmd_clear(pmd_t *pmd) |
7ce0bcfd ZA |
506 | { |
507 | const pte_t pte = { 0 }; | |
508 | vmi_check_page_type(__pa(pmd) >> PAGE_SHIFT, VMI_PAGE_PMD); | |
509 | vmi_ops.set_pte(pte, (pte_t *)pmd, VMI_PAGE_PD); | |
510 | } | |
511 | #endif | |
512 | ||
513 | #ifdef CONFIG_SMP | |
c6b36e9a | 514 | static void __devinit |
7ce0bcfd ZA |
515 | vmi_startup_ipi_hook(int phys_apicid, unsigned long start_eip, |
516 | unsigned long start_esp) | |
517 | { | |
c6b36e9a ZA |
518 | struct vmi_ap_state ap; |
519 | ||
7ce0bcfd ZA |
520 | /* Default everything to zero. This is fine for most GPRs. */ |
521 | memset(&ap, 0, sizeof(struct vmi_ap_state)); | |
522 | ||
523 | ap.gdtr_limit = GDT_SIZE - 1; | |
524 | ap.gdtr_base = (unsigned long) get_cpu_gdt_table(phys_apicid); | |
525 | ||
526 | ap.idtr_limit = IDT_ENTRIES * 8 - 1; | |
527 | ap.idtr_base = (unsigned long) idt_table; | |
528 | ||
529 | ap.ldtr = 0; | |
530 | ||
531 | ap.cs = __KERNEL_CS; | |
532 | ap.eip = (unsigned long) start_eip; | |
533 | ap.ss = __KERNEL_DS; | |
534 | ap.esp = (unsigned long) start_esp; | |
535 | ||
536 | ap.ds = __USER_DS; | |
537 | ap.es = __USER_DS; | |
7c3576d2 | 538 | ap.fs = __KERNEL_PERCPU; |
7ce0bcfd ZA |
539 | ap.gs = 0; |
540 | ||
541 | ap.eflags = 0; | |
542 | ||
7ce0bcfd ZA |
543 | #ifdef CONFIG_X86_PAE |
544 | /* efer should match BSP efer. */ | |
545 | if (cpu_has_nx) { | |
546 | unsigned l, h; | |
547 | rdmsr(MSR_EFER, l, h); | |
548 | ap.efer = (unsigned long long) h << 32 | l; | |
549 | } | |
550 | #endif | |
551 | ||
552 | ap.cr3 = __pa(swapper_pg_dir); | |
553 | /* Protected mode, paging, AM, WP, NE, MP. */ | |
554 | ap.cr0 = 0x80050023; | |
555 | ap.cr4 = mmu_cr4_features; | |
c6b36e9a | 556 | vmi_ops.set_initial_ap_state((u32)&ap, phys_apicid); |
7ce0bcfd ZA |
557 | } |
558 | #endif | |
559 | ||
49f19710 ZA |
560 | static void vmi_set_lazy_mode(int mode) |
561 | { | |
562 | static DEFINE_PER_CPU(int, lazy_mode); | |
563 | ||
564 | if (!vmi_ops.set_lazy_mode) | |
565 | return; | |
566 | ||
567 | /* Modes should never nest or overlap */ | |
568 | BUG_ON(__get_cpu_var(lazy_mode) && !(mode == PARAVIRT_LAZY_NONE || | |
569 | mode == PARAVIRT_LAZY_FLUSH)); | |
570 | ||
571 | if (mode == PARAVIRT_LAZY_FLUSH) { | |
572 | vmi_ops.set_lazy_mode(0); | |
573 | vmi_ops.set_lazy_mode(__get_cpu_var(lazy_mode)); | |
574 | } else { | |
575 | vmi_ops.set_lazy_mode(mode); | |
576 | __get_cpu_var(lazy_mode) = mode; | |
577 | } | |
578 | } | |
579 | ||
7ce0bcfd ZA |
580 | static inline int __init check_vmi_rom(struct vrom_header *rom) |
581 | { | |
582 | struct pci_header *pci; | |
583 | struct pnp_header *pnp; | |
584 | const char *manufacturer = "UNKNOWN"; | |
585 | const char *product = "UNKNOWN"; | |
586 | const char *license = "unspecified"; | |
587 | ||
588 | if (rom->rom_signature != 0xaa55) | |
589 | return 0; | |
590 | if (rom->vrom_signature != VMI_SIGNATURE) | |
591 | return 0; | |
592 | if (rom->api_version_maj != VMI_API_REV_MAJOR || | |
593 | rom->api_version_min+1 < VMI_API_REV_MINOR+1) { | |
594 | printk(KERN_WARNING "VMI: Found mismatched rom version %d.%d\n", | |
595 | rom->api_version_maj, | |
596 | rom->api_version_min); | |
597 | return 0; | |
598 | } | |
599 | ||
600 | /* | |
601 | * Relying on the VMI_SIGNATURE field is not 100% safe, so check | |
602 | * the PCI header and device type to make sure this is really a | |
603 | * VMI device. | |
604 | */ | |
605 | if (!rom->pci_header_offs) { | |
606 | printk(KERN_WARNING "VMI: ROM does not contain PCI header.\n"); | |
607 | return 0; | |
608 | } | |
609 | ||
610 | pci = (struct pci_header *)((char *)rom+rom->pci_header_offs); | |
611 | if (pci->vendorID != PCI_VENDOR_ID_VMWARE || | |
612 | pci->deviceID != PCI_DEVICE_ID_VMWARE_VMI) { | |
613 | /* Allow it to run... anyways, but warn */ | |
614 | printk(KERN_WARNING "VMI: ROM from unknown manufacturer\n"); | |
615 | } | |
616 | ||
617 | if (rom->pnp_header_offs) { | |
618 | pnp = (struct pnp_header *)((char *)rom+rom->pnp_header_offs); | |
619 | if (pnp->manufacturer_offset) | |
620 | manufacturer = (const char *)rom+pnp->manufacturer_offset; | |
621 | if (pnp->product_offset) | |
622 | product = (const char *)rom+pnp->product_offset; | |
623 | } | |
624 | ||
625 | if (rom->license_offs) | |
626 | license = (char *)rom+rom->license_offs; | |
627 | ||
628 | printk(KERN_INFO "VMI: Found %s %s, API version %d.%d, ROM version %d.%d\n", | |
629 | manufacturer, product, | |
630 | rom->api_version_maj, rom->api_version_min, | |
631 | pci->rom_version_maj, pci->rom_version_min); | |
632 | ||
302cf930 AK |
633 | /* Don't allow BSD/MIT here for now because we don't want to end up |
634 | with any binary only shim layers */ | |
635 | if (strcmp(license, "GPL") && strcmp(license, "GPL v2")) { | |
636 | printk(KERN_WARNING "VMI: Non GPL license `%s' found for ROM. Not used.\n", | |
637 | license); | |
638 | return 0; | |
639 | } | |
640 | ||
7ce0bcfd ZA |
641 | return 1; |
642 | } | |
643 | ||
644 | /* | |
645 | * Probe for the VMI option ROM | |
646 | */ | |
647 | static inline int __init probe_vmi_rom(void) | |
648 | { | |
649 | unsigned long base; | |
650 | ||
651 | /* VMI ROM is in option ROM area, check signature */ | |
652 | for (base = 0xC0000; base < 0xE0000; base += 2048) { | |
653 | struct vrom_header *romstart; | |
654 | romstart = (struct vrom_header *)isa_bus_to_virt(base); | |
655 | if (check_vmi_rom(romstart)) { | |
656 | vmi_rom = romstart; | |
657 | return 1; | |
658 | } | |
659 | } | |
660 | return 0; | |
661 | } | |
662 | ||
663 | /* | |
664 | * VMI setup common to all processors | |
665 | */ | |
666 | void vmi_bringup(void) | |
667 | { | |
668 | /* We must establish the lowmem mapping for MMU ops to work */ | |
772205f6 | 669 | if (vmi_ops.set_linear_mapping) |
eeef9c68 | 670 | vmi_ops.set_linear_mapping(0, (void *)__PAGE_OFFSET, max_low_pfn, 0); |
7ce0bcfd ZA |
671 | } |
672 | ||
673 | /* | |
772205f6 | 674 | * Return a pointer to a VMI function or NULL if unimplemented |
7ce0bcfd ZA |
675 | */ |
676 | static void *vmi_get_function(int vmicall) | |
677 | { | |
678 | u64 reloc; | |
679 | const struct vmi_relocation_info *rel = (struct vmi_relocation_info *)&reloc; | |
680 | reloc = call_vrom_long_func(vmi_rom, get_reloc, vmicall); | |
681 | BUG_ON(rel->type == VMI_RELOCATION_JUMP_REL); | |
682 | if (rel->type == VMI_RELOCATION_CALL_REL) | |
683 | return (void *)rel->eip; | |
684 | else | |
772205f6 | 685 | return NULL; |
7ce0bcfd ZA |
686 | } |
687 | ||
688 | /* | |
689 | * Helper macro for making the VMI paravirt-ops fill code readable. | |
772205f6 ZA |
690 | * For unimplemented operations, fall back to default, unless nop |
691 | * is returned by the ROM. | |
7ce0bcfd ZA |
692 | */ |
693 | #define para_fill(opname, vmicall) \ | |
694 | do { \ | |
695 | reloc = call_vrom_long_func(vmi_rom, get_reloc, \ | |
696 | VMI_CALL_##vmicall); \ | |
0492c371 | 697 | if (rel->type == VMI_RELOCATION_CALL_REL) \ |
7ce0bcfd | 698 | paravirt_ops.opname = (void *)rel->eip; \ |
0492c371 | 699 | else if (rel->type == VMI_RELOCATION_NOP) \ |
772205f6 | 700 | paravirt_ops.opname = (void *)vmi_nop; \ |
0492c371 ZA |
701 | else if (rel->type != VMI_RELOCATION_NONE) \ |
702 | printk(KERN_WARNING "VMI: Unknown relocation " \ | |
703 | "type %d for " #vmicall"\n",\ | |
704 | rel->type); \ | |
772205f6 ZA |
705 | } while (0) |
706 | ||
707 | /* | |
708 | * Helper macro for making the VMI paravirt-ops fill code readable. | |
709 | * For cached operations which do not match the VMI ROM ABI and must | |
710 | * go through a tranlation stub. Ignore NOPs, since it is not clear | |
711 | * a NOP * VMI function corresponds to a NOP paravirt-op when the | |
712 | * functions are not in 1-1 correspondence. | |
713 | */ | |
714 | #define para_wrap(opname, wrapper, cache, vmicall) \ | |
715 | do { \ | |
716 | reloc = call_vrom_long_func(vmi_rom, get_reloc, \ | |
717 | VMI_CALL_##vmicall); \ | |
718 | BUG_ON(rel->type == VMI_RELOCATION_JUMP_REL); \ | |
719 | if (rel->type == VMI_RELOCATION_CALL_REL) { \ | |
720 | paravirt_ops.opname = wrapper; \ | |
721 | vmi_ops.cache = (void *)rel->eip; \ | |
7ce0bcfd ZA |
722 | } \ |
723 | } while (0) | |
724 | ||
772205f6 | 725 | |
7ce0bcfd ZA |
726 | /* |
727 | * Activate the VMI interface and switch into paravirtualized mode | |
728 | */ | |
729 | static inline int __init activate_vmi(void) | |
730 | { | |
731 | short kernel_cs; | |
732 | u64 reloc; | |
733 | const struct vmi_relocation_info *rel = (struct vmi_relocation_info *)&reloc; | |
734 | ||
735 | if (call_vrom_func(vmi_rom, vmi_init) != 0) { | |
736 | printk(KERN_ERR "VMI ROM failed to initialize!"); | |
737 | return 0; | |
738 | } | |
739 | savesegment(cs, kernel_cs); | |
740 | ||
741 | paravirt_ops.paravirt_enabled = 1; | |
742 | paravirt_ops.kernel_rpl = kernel_cs & SEGMENT_RPL_MASK; | |
743 | ||
744 | paravirt_ops.patch = vmi_patch; | |
745 | paravirt_ops.name = "vmi"; | |
746 | ||
747 | /* | |
748 | * Many of these operations are ABI compatible with VMI. | |
749 | * This means we can fill in the paravirt-ops with direct | |
750 | * pointers into the VMI ROM. If the calling convention for | |
751 | * these operations changes, this code needs to be updated. | |
752 | * | |
753 | * Exceptions | |
754 | * CPUID paravirt-op uses pointers, not the native ISA | |
755 | * halt has no VMI equivalent; all VMI halts are "safe" | |
756 | * no MSR support yet - just trap and emulate. VMI uses the | |
757 | * same ABI as the native ISA, but Linux wants exceptions | |
758 | * from bogus MSR read / write handled | |
759 | * rdpmc is not yet used in Linux | |
760 | */ | |
761 | ||
772205f6 ZA |
762 | /* CPUID is special, so very special it gets wrapped like a present */ |
763 | para_wrap(cpuid, vmi_cpuid, cpuid, CPUID); | |
7ce0bcfd ZA |
764 | |
765 | para_fill(clts, CLTS); | |
766 | para_fill(get_debugreg, GetDR); | |
767 | para_fill(set_debugreg, SetDR); | |
768 | para_fill(read_cr0, GetCR0); | |
769 | para_fill(read_cr2, GetCR2); | |
770 | para_fill(read_cr3, GetCR3); | |
771 | para_fill(read_cr4, GetCR4); | |
772 | para_fill(write_cr0, SetCR0); | |
773 | para_fill(write_cr2, SetCR2); | |
774 | para_fill(write_cr3, SetCR3); | |
775 | para_fill(write_cr4, SetCR4); | |
776 | para_fill(save_fl, GetInterruptMask); | |
777 | para_fill(restore_fl, SetInterruptMask); | |
778 | para_fill(irq_disable, DisableInterrupts); | |
779 | para_fill(irq_enable, EnableInterrupts); | |
772205f6 | 780 | |
7ce0bcfd | 781 | para_fill(wbinvd, WBINVD); |
772205f6 ZA |
782 | para_fill(read_tsc, RDTSC); |
783 | ||
784 | /* The following we emulate with trap and emulate for now */ | |
7ce0bcfd ZA |
785 | /* paravirt_ops.read_msr = vmi_rdmsr */ |
786 | /* paravirt_ops.write_msr = vmi_wrmsr */ | |
7ce0bcfd ZA |
787 | /* paravirt_ops.rdpmc = vmi_rdpmc */ |
788 | ||
772205f6 ZA |
789 | /* TR interface doesn't pass TR value, wrap */ |
790 | para_wrap(load_tr_desc, vmi_set_tr, set_tr, SetTR); | |
7ce0bcfd ZA |
791 | |
792 | /* LDT is special, too */ | |
772205f6 | 793 | para_wrap(set_ldt, vmi_set_ldt, _set_ldt, SetLDT); |
7ce0bcfd ZA |
794 | |
795 | para_fill(load_gdt, SetGDT); | |
796 | para_fill(load_idt, SetIDT); | |
797 | para_fill(store_gdt, GetGDT); | |
798 | para_fill(store_idt, GetIDT); | |
799 | para_fill(store_tr, GetTR); | |
800 | paravirt_ops.load_tls = vmi_load_tls; | |
801 | para_fill(write_ldt_entry, WriteLDTEntry); | |
802 | para_fill(write_gdt_entry, WriteGDTEntry); | |
803 | para_fill(write_idt_entry, WriteIDTEntry); | |
772205f6 | 804 | para_wrap(load_esp0, vmi_load_esp0, set_kernel_stack, UpdateKernelStack); |
7ce0bcfd | 805 | para_fill(set_iopl_mask, SetIOPLMask); |
772205f6 | 806 | para_fill(io_delay, IODelay); |
49f19710 | 807 | para_wrap(set_lazy_mode, vmi_set_lazy_mode, set_lazy_mode, SetLazyMode); |
7ce0bcfd | 808 | |
772205f6 | 809 | /* user and kernel flush are just handled with different flags to FlushTLB */ |
eeef9c68 ZA |
810 | para_wrap(flush_tlb_user, vmi_flush_tlb_user, _flush_tlb, FlushTLB); |
811 | para_wrap(flush_tlb_kernel, vmi_flush_tlb_kernel, _flush_tlb, FlushTLB); | |
7ce0bcfd ZA |
812 | para_fill(flush_tlb_single, InvalPage); |
813 | ||
814 | /* | |
815 | * Until a standard flag format can be agreed on, we need to | |
816 | * implement these as wrappers in Linux. Get the VMI ROM | |
817 | * function pointers for the two backend calls. | |
818 | */ | |
819 | #ifdef CONFIG_X86_PAE | |
820 | vmi_ops.set_pte = vmi_get_function(VMI_CALL_SetPxELong); | |
821 | vmi_ops.update_pte = vmi_get_function(VMI_CALL_UpdatePxELong); | |
822 | #else | |
823 | vmi_ops.set_pte = vmi_get_function(VMI_CALL_SetPxE); | |
824 | vmi_ops.update_pte = vmi_get_function(VMI_CALL_UpdatePxE); | |
825 | #endif | |
7ce0bcfd | 826 | |
772205f6 ZA |
827 | if (vmi_ops.set_pte) { |
828 | paravirt_ops.set_pte = vmi_set_pte; | |
829 | paravirt_ops.set_pte_at = vmi_set_pte_at; | |
830 | paravirt_ops.set_pmd = vmi_set_pmd; | |
7ce0bcfd | 831 | #ifdef CONFIG_X86_PAE |
772205f6 ZA |
832 | paravirt_ops.set_pte_atomic = vmi_set_pte_atomic; |
833 | paravirt_ops.set_pte_present = vmi_set_pte_present; | |
834 | paravirt_ops.set_pud = vmi_set_pud; | |
835 | paravirt_ops.pte_clear = vmi_pte_clear; | |
836 | paravirt_ops.pmd_clear = vmi_pmd_clear; | |
7ce0bcfd | 837 | #endif |
772205f6 ZA |
838 | } |
839 | ||
840 | if (vmi_ops.update_pte) { | |
841 | paravirt_ops.pte_update = vmi_update_pte; | |
842 | paravirt_ops.pte_update_defer = vmi_update_pte_defer; | |
843 | } | |
844 | ||
845 | vmi_ops.allocate_page = vmi_get_function(VMI_CALL_AllocatePage); | |
846 | if (vmi_ops.allocate_page) { | |
847 | paravirt_ops.alloc_pt = vmi_allocate_pt; | |
848 | paravirt_ops.alloc_pd = vmi_allocate_pd; | |
849 | paravirt_ops.alloc_pd_clone = vmi_allocate_pd_clone; | |
850 | } | |
851 | ||
852 | vmi_ops.release_page = vmi_get_function(VMI_CALL_ReleasePage); | |
853 | if (vmi_ops.release_page) { | |
854 | paravirt_ops.release_pt = vmi_release_pt; | |
855 | paravirt_ops.release_pd = vmi_release_pd; | |
856 | } | |
eeef9c68 ZA |
857 | |
858 | /* Set linear is needed in all cases */ | |
859 | vmi_ops.set_linear_mapping = vmi_get_function(VMI_CALL_SetLinearMapping); | |
860 | #ifdef CONFIG_HIGHPTE | |
861 | if (vmi_ops.set_linear_mapping) | |
862 | paravirt_ops.kmap_atomic_pte = vmi_kmap_atomic_pte; | |
a27fe809 | 863 | #endif |
772205f6 | 864 | |
7ce0bcfd ZA |
865 | /* |
866 | * These MUST always be patched. Don't support indirect jumps | |
867 | * through these operations, as the VMI interface may use either | |
868 | * a jump or a call to get to these operations, depending on | |
869 | * the backend. They are performance critical anyway, so requiring | |
870 | * a patch is not a big problem. | |
871 | */ | |
872 | paravirt_ops.irq_enable_sysexit = (void *)0xfeedbab0; | |
873 | paravirt_ops.iret = (void *)0xbadbab0; | |
874 | ||
875 | #ifdef CONFIG_SMP | |
772205f6 | 876 | para_wrap(startup_ipi_hook, vmi_startup_ipi_hook, set_initial_ap_state, SetInitialAPState); |
7ce0bcfd ZA |
877 | #endif |
878 | ||
879 | #ifdef CONFIG_X86_LOCAL_APIC | |
772205f6 ZA |
880 | para_fill(apic_read, APICRead); |
881 | para_fill(apic_write, APICWrite); | |
882 | para_fill(apic_write_atomic, APICWrite); | |
7ce0bcfd ZA |
883 | #endif |
884 | ||
bbab4f3b ZA |
885 | /* |
886 | * Check for VMI timer functionality by probing for a cycle frequency method | |
887 | */ | |
888 | reloc = call_vrom_long_func(vmi_rom, get_reloc, VMI_CALL_GetCycleFrequency); | |
772205f6 | 889 | if (!disable_vmi_timer && rel->type != VMI_RELOCATION_NONE) { |
bbab4f3b ZA |
890 | vmi_timer_ops.get_cycle_frequency = (void *)rel->eip; |
891 | vmi_timer_ops.get_cycle_counter = | |
892 | vmi_get_function(VMI_CALL_GetCycleCounter); | |
893 | vmi_timer_ops.get_wallclock = | |
894 | vmi_get_function(VMI_CALL_GetWallclockTime); | |
895 | vmi_timer_ops.wallclock_updated = | |
896 | vmi_get_function(VMI_CALL_WallclockUpdated); | |
897 | vmi_timer_ops.set_alarm = vmi_get_function(VMI_CALL_SetAlarm); | |
898 | vmi_timer_ops.cancel_alarm = | |
899 | vmi_get_function(VMI_CALL_CancelAlarm); | |
900 | paravirt_ops.time_init = vmi_time_init; | |
901 | paravirt_ops.get_wallclock = vmi_get_wallclock; | |
902 | paravirt_ops.set_wallclock = vmi_set_wallclock; | |
903 | #ifdef CONFIG_X86_LOCAL_APIC | |
904 | paravirt_ops.setup_boot_clock = vmi_timer_setup_boot_alarm; | |
905 | paravirt_ops.setup_secondary_clock = vmi_timer_setup_secondary_alarm; | |
906 | #endif | |
6cb9a835 | 907 | paravirt_ops.get_scheduled_cycles = vmi_get_sched_cycles; |
1182d852 | 908 | paravirt_ops.get_cpu_khz = vmi_cpu_khz; |
772205f6 ZA |
909 | |
910 | /* We have true wallclock functions; disable CMOS clock sync */ | |
911 | no_sync_cmos_clock = 1; | |
912 | } else { | |
913 | disable_noidle = 1; | |
914 | disable_vmi_timer = 1; | |
bbab4f3b | 915 | } |
772205f6 ZA |
916 | |
917 | /* No idle HZ mode only works if VMI timer and no idle is enabled */ | |
918 | if (disable_noidle || disable_vmi_timer) | |
7507ba34 | 919 | para_fill(safe_halt, Halt); |
772205f6 ZA |
920 | else |
921 | para_wrap(safe_halt, vmi_safe_halt, halt, Halt); | |
bbab4f3b | 922 | |
7ce0bcfd ZA |
923 | /* |
924 | * Alternative instruction rewriting doesn't happen soon enough | |
925 | * to convert VMI_IRET to a call instead of a jump; so we have | |
926 | * to do this before IRQs get reenabled. Fortunately, it is | |
927 | * idempotent. | |
928 | */ | |
929 | apply_paravirt(__start_parainstructions, __stop_parainstructions); | |
930 | ||
931 | vmi_bringup(); | |
932 | ||
933 | return 1; | |
934 | } | |
935 | ||
936 | #undef para_fill | |
937 | ||
938 | void __init vmi_init(void) | |
939 | { | |
940 | unsigned long flags; | |
941 | ||
942 | if (!vmi_rom) | |
943 | probe_vmi_rom(); | |
944 | else | |
945 | check_vmi_rom(vmi_rom); | |
946 | ||
947 | /* In case probing for or validating the ROM failed, basil */ | |
948 | if (!vmi_rom) | |
949 | return; | |
950 | ||
951 | reserve_top_address(-vmi_rom->virtual_top); | |
952 | ||
953 | local_irq_save(flags); | |
954 | activate_vmi(); | |
7507ba34 ZA |
955 | |
956 | #ifdef CONFIG_X86_IO_APIC | |
772205f6 | 957 | /* This is virtual hardware; timer routing is wired correctly */ |
7ce0bcfd ZA |
958 | no_timer_check = 1; |
959 | #endif | |
960 | local_irq_restore(flags & X86_EFLAGS_IF); | |
961 | } | |
962 | ||
963 | static int __init parse_vmi(char *arg) | |
964 | { | |
965 | if (!arg) | |
966 | return -EINVAL; | |
967 | ||
eda08b1b | 968 | if (!strcmp(arg, "disable_pge")) { |
7ce0bcfd ZA |
969 | clear_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability); |
970 | disable_pge = 1; | |
971 | } else if (!strcmp(arg, "disable_pse")) { | |
972 | clear_bit(X86_FEATURE_PSE, boot_cpu_data.x86_capability); | |
973 | disable_pse = 1; | |
974 | } else if (!strcmp(arg, "disable_sep")) { | |
975 | clear_bit(X86_FEATURE_SEP, boot_cpu_data.x86_capability); | |
976 | disable_sep = 1; | |
977 | } else if (!strcmp(arg, "disable_tsc")) { | |
978 | clear_bit(X86_FEATURE_TSC, boot_cpu_data.x86_capability); | |
979 | disable_tsc = 1; | |
980 | } else if (!strcmp(arg, "disable_mtrr")) { | |
981 | clear_bit(X86_FEATURE_MTRR, boot_cpu_data.x86_capability); | |
982 | disable_mtrr = 1; | |
772205f6 ZA |
983 | } else if (!strcmp(arg, "disable_timer")) { |
984 | disable_vmi_timer = 1; | |
985 | disable_noidle = 1; | |
7507ba34 ZA |
986 | } else if (!strcmp(arg, "disable_noidle")) |
987 | disable_noidle = 1; | |
7ce0bcfd ZA |
988 | return 0; |
989 | } | |
990 | ||
991 | early_param("vmi", parse_vmi); |