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