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