aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/kernel/vmi_32.c
diff options
context:
space:
mode:
Diffstat (limited to 'arch/x86/kernel/vmi_32.c')
-rw-r--r--arch/x86/kernel/vmi_32.c981
1 files changed, 981 insertions, 0 deletions
diff --git a/arch/x86/kernel/vmi_32.c b/arch/x86/kernel/vmi_32.c
new file mode 100644
index 000000000000..18673e0f193b
--- /dev/null
+++ b/arch/x86/kernel/vmi_32.c
@@ -0,0 +1,981 @@
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/cpu.h>
27#include <linux/bootmem.h>
28#include <linux/mm.h>
29#include <linux/highmem.h>
30#include <linux/sched.h>
31#include <asm/vmi.h>
32#include <asm/io.h>
33#include <asm/fixmap.h>
34#include <asm/apicdef.h>
35#include <asm/apic.h>
36#include <asm/processor.h>
37#include <asm/timer.h>
38#include <asm/vmi_time.h>
39#include <asm/kmap_types.h>
40
41/* Convenient for calling VMI functions indirectly in the ROM */
42typedef u32 __attribute__((regparm(1))) (VROMFUNC)(void);
43typedef u64 __attribute__((regparm(2))) (VROMLONGFUNC)(int);
44
45#define call_vrom_func(rom,func) \
46 (((VROMFUNC *)(rom->func))())
47
48#define call_vrom_long_func(rom,func,arg) \
49 (((VROMLONGFUNC *)(rom->func)) (arg))
50
51static struct vrom_header *vmi_rom;
52static int disable_pge;
53static int disable_pse;
54static int disable_sep;
55static int disable_tsc;
56static int disable_mtrr;
57static int disable_noidle;
58static int disable_vmi_timer;
59
60/* Cached VMI operations */
61static struct {
62 void (*cpuid)(void /* non-c */);
63 void (*_set_ldt)(u32 selector);
64 void (*set_tr)(u32 selector);
65 void (*set_kernel_stack)(u32 selector, u32 esp0);
66 void (*allocate_page)(u32, u32, u32, u32, u32);
67 void (*release_page)(u32, u32);
68 void (*set_pte)(pte_t, pte_t *, unsigned);
69 void (*update_pte)(pte_t *, unsigned);
70 void (*set_linear_mapping)(int, void *, u32, u32);
71 void (*_flush_tlb)(int);
72 void (*set_initial_ap_state)(int, int);
73 void (*halt)(void);
74 void (*set_lazy_mode)(int mode);
75} vmi_ops;
76
77/* Cached VMI operations */
78struct vmi_timer_ops vmi_timer_ops;
79
80/*
81 * VMI patching routines.
82 */
83#define MNEM_CALL 0xe8
84#define MNEM_JMP 0xe9
85#define MNEM_RET 0xc3
86
87#define IRQ_PATCH_INT_MASK 0
88#define IRQ_PATCH_DISABLE 5
89
90static inline void patch_offset(void *insnbuf,
91 unsigned long eip, unsigned long dest)
92{
93 *(unsigned long *)(insnbuf+1) = dest-eip-5;
94}
95
96static unsigned patch_internal(int call, unsigned len, void *insnbuf,
97 unsigned long eip)
98{
99 u64 reloc;
100 struct vmi_relocation_info *const rel = (struct vmi_relocation_info *)&reloc;
101 reloc = call_vrom_long_func(vmi_rom, get_reloc, call);
102 switch(rel->type) {
103 case VMI_RELOCATION_CALL_REL:
104 BUG_ON(len < 5);
105 *(char *)insnbuf = MNEM_CALL;
106 patch_offset(insnbuf, eip, (unsigned long)rel->eip);
107 return 5;
108
109 case VMI_RELOCATION_JUMP_REL:
110 BUG_ON(len < 5);
111 *(char *)insnbuf = MNEM_JMP;
112 patch_offset(insnbuf, eip, (unsigned long)rel->eip);
113 return 5;
114
115 case VMI_RELOCATION_NOP:
116 /* obliterate the whole thing */
117 return 0;
118
119 case VMI_RELOCATION_NONE:
120 /* leave native code in place */
121 break;
122
123 default:
124 BUG();
125 }
126 return len;
127}
128
129/*
130 * Apply patch if appropriate, return length of new instruction
131 * sequence. The callee does nop padding for us.
132 */
133static unsigned vmi_patch(u8 type, u16 clobbers, void *insns,
134 unsigned long eip, unsigned len)
135{
136 switch (type) {
137 case PARAVIRT_PATCH(irq_disable):
138 return patch_internal(VMI_CALL_DisableInterrupts, len,
139 insns, eip);
140 case PARAVIRT_PATCH(irq_enable):
141 return patch_internal(VMI_CALL_EnableInterrupts, len,
142 insns, eip);
143 case PARAVIRT_PATCH(restore_fl):
144 return patch_internal(VMI_CALL_SetInterruptMask, len,
145 insns, eip);
146 case PARAVIRT_PATCH(save_fl):
147 return patch_internal(VMI_CALL_GetInterruptMask, len,
148 insns, eip);
149 case PARAVIRT_PATCH(iret):
150 return patch_internal(VMI_CALL_IRET, len, insns, eip);
151 case PARAVIRT_PATCH(irq_enable_sysexit):
152 return patch_internal(VMI_CALL_SYSEXIT, len, insns, eip);
153 default:
154 break;
155 }
156 return len;
157}
158
159/* CPUID has non-C semantics, and paravirt-ops API doesn't match hardware ISA */
160static void vmi_cpuid(unsigned int *eax, unsigned int *ebx,
161 unsigned int *ecx, unsigned int *edx)
162{
163 int override = 0;
164 if (*eax == 1)
165 override = 1;
166 asm volatile ("call *%6"
167 : "=a" (*eax),
168 "=b" (*ebx),
169 "=c" (*ecx),
170 "=d" (*edx)
171 : "0" (*eax), "2" (*ecx), "r" (vmi_ops.cpuid));
172 if (override) {
173 if (disable_pse)
174 *edx &= ~X86_FEATURE_PSE;
175 if (disable_pge)
176 *edx &= ~X86_FEATURE_PGE;
177 if (disable_sep)
178 *edx &= ~X86_FEATURE_SEP;
179 if (disable_tsc)
180 *edx &= ~X86_FEATURE_TSC;
181 if (disable_mtrr)
182 *edx &= ~X86_FEATURE_MTRR;
183 }
184}
185
186static inline void vmi_maybe_load_tls(struct desc_struct *gdt, int nr, struct desc_struct *new)
187{
188 if (gdt[nr].a != new->a || gdt[nr].b != new->b)
189 write_gdt_entry(gdt, nr, new->a, new->b);
190}
191
192static void vmi_load_tls(struct thread_struct *t, unsigned int cpu)
193{
194 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
195 vmi_maybe_load_tls(gdt, GDT_ENTRY_TLS_MIN + 0, &t->tls_array[0]);
196 vmi_maybe_load_tls(gdt, GDT_ENTRY_TLS_MIN + 1, &t->tls_array[1]);
197 vmi_maybe_load_tls(gdt, GDT_ENTRY_TLS_MIN + 2, &t->tls_array[2]);
198}
199
200static void vmi_set_ldt(const void *addr, unsigned entries)
201{
202 unsigned cpu = smp_processor_id();
203 u32 low, high;
204
205 pack_descriptor(&low, &high, (unsigned long)addr,
206 entries * sizeof(struct desc_struct) - 1,
207 DESCTYPE_LDT, 0);
208 write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_LDT, low, high);
209 vmi_ops._set_ldt(entries ? GDT_ENTRY_LDT*sizeof(struct desc_struct) : 0);
210}
211
212static void vmi_set_tr(void)
213{
214 vmi_ops.set_tr(GDT_ENTRY_TSS*sizeof(struct desc_struct));
215}
216
217static void vmi_load_esp0(struct tss_struct *tss,
218 struct thread_struct *thread)
219{
220 tss->x86_tss.esp0 = thread->esp0;
221
222 /* This can only happen when SEP is enabled, no need to test "SEP"arately */
223 if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
224 tss->x86_tss.ss1 = thread->sysenter_cs;
225 wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
226 }
227 vmi_ops.set_kernel_stack(__KERNEL_DS, tss->x86_tss.esp0);
228}
229
230static void vmi_flush_tlb_user(void)
231{
232 vmi_ops._flush_tlb(VMI_FLUSH_TLB);
233}
234
235static void vmi_flush_tlb_kernel(void)
236{
237 vmi_ops._flush_tlb(VMI_FLUSH_TLB | VMI_FLUSH_GLOBAL);
238}
239
240/* Stub to do nothing at all; used for delays and unimplemented calls */
241static void vmi_nop(void)
242{
243}
244
245#ifdef CONFIG_DEBUG_PAGE_TYPE
246
247#ifdef CONFIG_X86_PAE
248#define MAX_BOOT_PTS (2048+4+1)
249#else
250#define MAX_BOOT_PTS (1024+1)
251#endif
252
253/*
254 * During boot, mem_map is not yet available in paging_init, so stash
255 * all the boot page allocations here.
256 */
257static struct {
258 u32 pfn;
259 int type;
260} boot_page_allocations[MAX_BOOT_PTS];
261static int num_boot_page_allocations;
262static int boot_allocations_applied;
263
264void vmi_apply_boot_page_allocations(void)
265{
266 int i;
267 BUG_ON(!mem_map);
268 for (i = 0; i < num_boot_page_allocations; i++) {
269 struct page *page = pfn_to_page(boot_page_allocations[i].pfn);
270 page->type = boot_page_allocations[i].type;
271 page->type = boot_page_allocations[i].type &
272 ~(VMI_PAGE_ZEROED | VMI_PAGE_CLONE);
273 }
274 boot_allocations_applied = 1;
275}
276
277static void record_page_type(u32 pfn, int type)
278{
279 BUG_ON(num_boot_page_allocations >= MAX_BOOT_PTS);
280 boot_page_allocations[num_boot_page_allocations].pfn = pfn;
281 boot_page_allocations[num_boot_page_allocations].type = type;
282 num_boot_page_allocations++;
283}
284
285static void check_zeroed_page(u32 pfn, int type, struct page *page)
286{
287 u32 *ptr;
288 int i;
289 int limit = PAGE_SIZE / sizeof(int);
290
291 if (page_address(page))
292 ptr = (u32 *)page_address(page);
293 else
294 ptr = (u32 *)__va(pfn << PAGE_SHIFT);
295 /*
296 * When cloning the root in non-PAE mode, only the userspace
297 * pdes need to be zeroed.
298 */
299 if (type & VMI_PAGE_CLONE)
300 limit = USER_PTRS_PER_PGD;
301 for (i = 0; i < limit; i++)
302 BUG_ON(ptr[i]);
303}
304
305/*
306 * We stash the page type into struct page so we can verify the page
307 * types are used properly.
308 */
309static void vmi_set_page_type(u32 pfn, int type)
310{
311 /* PAE can have multiple roots per page - don't track */
312 if (PTRS_PER_PMD > 1 && (type & VMI_PAGE_PDP))
313 return;
314
315 if (boot_allocations_applied) {
316 struct page *page = pfn_to_page(pfn);
317 if (type != VMI_PAGE_NORMAL)
318 BUG_ON(page->type);
319 else
320 BUG_ON(page->type == VMI_PAGE_NORMAL);
321 page->type = type & ~(VMI_PAGE_ZEROED | VMI_PAGE_CLONE);
322 if (type & VMI_PAGE_ZEROED)
323 check_zeroed_page(pfn, type, page);
324 } else {
325 record_page_type(pfn, type);
326 }
327}
328
329static void vmi_check_page_type(u32 pfn, int type)
330{
331 /* PAE can have multiple roots per page - skip checks */
332 if (PTRS_PER_PMD > 1 && (type & VMI_PAGE_PDP))
333 return;
334
335 type &= ~(VMI_PAGE_ZEROED | VMI_PAGE_CLONE);
336 if (boot_allocations_applied) {
337 struct page *page = pfn_to_page(pfn);
338 BUG_ON((page->type ^ type) & VMI_PAGE_PAE);
339 BUG_ON(type == VMI_PAGE_NORMAL && page->type);
340 BUG_ON((type & page->type) == 0);
341 }
342}
343#else
344#define vmi_set_page_type(p,t) do { } while (0)
345#define vmi_check_page_type(p,t) do { } while (0)
346#endif
347
348#ifdef CONFIG_HIGHPTE
349static void *vmi_kmap_atomic_pte(struct page *page, enum km_type type)
350{
351 void *va = kmap_atomic(page, type);
352
353 /*
354 * Internally, the VMI ROM must map virtual addresses to physical
355 * addresses for processing MMU updates. By the time MMU updates
356 * are issued, this information is typically already lost.
357 * Fortunately, the VMI provides a cache of mapping slots for active
358 * page tables.
359 *
360 * We use slot zero for the linear mapping of physical memory, and
361 * in HIGHPTE kernels, slot 1 and 2 for KM_PTE0 and KM_PTE1.
362 *
363 * args: SLOT VA COUNT PFN
364 */
365 BUG_ON(type != KM_PTE0 && type != KM_PTE1);
366 vmi_ops.set_linear_mapping((type - KM_PTE0)+1, va, 1, page_to_pfn(page));
367
368 return va;
369}
370#endif
371
372static void vmi_allocate_pt(struct mm_struct *mm, u32 pfn)
373{
374 vmi_set_page_type(pfn, VMI_PAGE_L1);
375 vmi_ops.allocate_page(pfn, VMI_PAGE_L1, 0, 0, 0);
376}
377
378static void vmi_allocate_pd(u32 pfn)
379{
380 /*
381 * This call comes in very early, before mem_map is setup.
382 * It is called only for swapper_pg_dir, which already has
383 * data on it.
384 */
385 vmi_set_page_type(pfn, VMI_PAGE_L2);
386 vmi_ops.allocate_page(pfn, VMI_PAGE_L2, 0, 0, 0);
387}
388
389static void vmi_allocate_pd_clone(u32 pfn, u32 clonepfn, u32 start, u32 count)
390{
391 vmi_set_page_type(pfn, VMI_PAGE_L2 | VMI_PAGE_CLONE);
392 vmi_check_page_type(clonepfn, VMI_PAGE_L2);
393 vmi_ops.allocate_page(pfn, VMI_PAGE_L2 | VMI_PAGE_CLONE, clonepfn, start, count);
394}
395
396static void vmi_release_pt(u32 pfn)
397{
398 vmi_ops.release_page(pfn, VMI_PAGE_L1);
399 vmi_set_page_type(pfn, VMI_PAGE_NORMAL);
400}
401
402static void vmi_release_pd(u32 pfn)
403{
404 vmi_ops.release_page(pfn, VMI_PAGE_L2);
405 vmi_set_page_type(pfn, VMI_PAGE_NORMAL);
406}
407
408/*
409 * Helper macros for MMU update flags. We can defer updates until a flush
410 * or page invalidation only if the update is to the current address space
411 * (otherwise, there is no flush). We must check against init_mm, since
412 * this could be a kernel update, which usually passes init_mm, although
413 * sometimes this check can be skipped if we know the particular function
414 * is only called on user mode PTEs. We could change the kernel to pass
415 * current->active_mm here, but in particular, I was unsure if changing
416 * mm/highmem.c to do this would still be correct on other architectures.
417 */
418#define is_current_as(mm, mustbeuser) ((mm) == current->active_mm || \
419 (!mustbeuser && (mm) == &init_mm))
420#define vmi_flags_addr(mm, addr, level, user) \
421 ((level) | (is_current_as(mm, user) ? \
422 (VMI_PAGE_CURRENT_AS | ((addr) & VMI_PAGE_VA_MASK)) : 0))
423#define vmi_flags_addr_defer(mm, addr, level, user) \
424 ((level) | (is_current_as(mm, user) ? \
425 (VMI_PAGE_DEFER | VMI_PAGE_CURRENT_AS | ((addr) & VMI_PAGE_VA_MASK)) : 0))
426
427static void vmi_update_pte(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
428{
429 vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE);
430 vmi_ops.update_pte(ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0));
431}
432
433static void vmi_update_pte_defer(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
434{
435 vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE);
436 vmi_ops.update_pte(ptep, vmi_flags_addr_defer(mm, addr, VMI_PAGE_PT, 0));
437}
438
439static void vmi_set_pte(pte_t *ptep, pte_t pte)
440{
441 /* XXX because of set_pmd_pte, this can be called on PT or PD layers */
442 vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE | VMI_PAGE_PD);
443 vmi_ops.set_pte(pte, ptep, VMI_PAGE_PT);
444}
445
446static void vmi_set_pte_at(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pte)
447{
448 vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE);
449 vmi_ops.set_pte(pte, ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0));
450}
451
452static void vmi_set_pmd(pmd_t *pmdp, pmd_t pmdval)
453{
454#ifdef CONFIG_X86_PAE
455 const pte_t pte = { pmdval.pmd, pmdval.pmd >> 32 };
456 vmi_check_page_type(__pa(pmdp) >> PAGE_SHIFT, VMI_PAGE_PMD);
457#else
458 const pte_t pte = { pmdval.pud.pgd.pgd };
459 vmi_check_page_type(__pa(pmdp) >> PAGE_SHIFT, VMI_PAGE_PGD);
460#endif
461 vmi_ops.set_pte(pte, (pte_t *)pmdp, VMI_PAGE_PD);
462}
463
464#ifdef CONFIG_X86_PAE
465
466static void vmi_set_pte_atomic(pte_t *ptep, pte_t pteval)
467{
468 /*
469 * XXX This is called from set_pmd_pte, but at both PT
470 * and PD layers so the VMI_PAGE_PT flag is wrong. But
471 * it is only called for large page mapping changes,
472 * the Xen backend, doesn't support large pages, and the
473 * ESX backend doesn't depend on the flag.
474 */
475 set_64bit((unsigned long long *)ptep,pte_val(pteval));
476 vmi_ops.update_pte(ptep, VMI_PAGE_PT);
477}
478
479static void vmi_set_pte_present(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pte)
480{
481 vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE);
482 vmi_ops.set_pte(pte, ptep, vmi_flags_addr_defer(mm, addr, VMI_PAGE_PT, 1));
483}
484
485static void vmi_set_pud(pud_t *pudp, pud_t pudval)
486{
487 /* Um, eww */
488 const pte_t pte = { pudval.pgd.pgd, pudval.pgd.pgd >> 32 };
489 vmi_check_page_type(__pa(pudp) >> PAGE_SHIFT, VMI_PAGE_PGD);
490 vmi_ops.set_pte(pte, (pte_t *)pudp, VMI_PAGE_PDP);
491}
492
493static void vmi_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
494{
495 const pte_t pte = { 0 };
496 vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE);
497 vmi_ops.set_pte(pte, ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0));
498}
499
500static void vmi_pmd_clear(pmd_t *pmd)
501{
502 const pte_t pte = { 0 };
503 vmi_check_page_type(__pa(pmd) >> PAGE_SHIFT, VMI_PAGE_PMD);
504 vmi_ops.set_pte(pte, (pte_t *)pmd, VMI_PAGE_PD);
505}
506#endif
507
508#ifdef CONFIG_SMP
509static void __devinit
510vmi_startup_ipi_hook(int phys_apicid, unsigned long start_eip,
511 unsigned long start_esp)
512{
513 struct vmi_ap_state ap;
514
515 /* Default everything to zero. This is fine for most GPRs. */
516 memset(&ap, 0, sizeof(struct vmi_ap_state));
517
518 ap.gdtr_limit = GDT_SIZE - 1;
519 ap.gdtr_base = (unsigned long) get_cpu_gdt_table(phys_apicid);
520
521 ap.idtr_limit = IDT_ENTRIES * 8 - 1;
522 ap.idtr_base = (unsigned long) idt_table;
523
524 ap.ldtr = 0;
525
526 ap.cs = __KERNEL_CS;
527 ap.eip = (unsigned long) start_eip;
528 ap.ss = __KERNEL_DS;
529 ap.esp = (unsigned long) start_esp;
530
531 ap.ds = __USER_DS;
532 ap.es = __USER_DS;
533 ap.fs = __KERNEL_PERCPU;
534 ap.gs = 0;
535
536 ap.eflags = 0;
537
538#ifdef CONFIG_X86_PAE
539 /* efer should match BSP efer. */
540 if (cpu_has_nx) {
541 unsigned l, h;
542 rdmsr(MSR_EFER, l, h);
543 ap.efer = (unsigned long long) h << 32 | l;
544 }
545#endif
546
547 ap.cr3 = __pa(swapper_pg_dir);
548 /* Protected mode, paging, AM, WP, NE, MP. */
549 ap.cr0 = 0x80050023;
550 ap.cr4 = mmu_cr4_features;
551 vmi_ops.set_initial_ap_state((u32)&ap, phys_apicid);
552}
553#endif
554
555static void vmi_set_lazy_mode(enum paravirt_lazy_mode mode)
556{
557 static DEFINE_PER_CPU(enum paravirt_lazy_mode, lazy_mode);
558
559 if (!vmi_ops.set_lazy_mode)
560 return;
561
562 /* Modes should never nest or overlap */
563 BUG_ON(__get_cpu_var(lazy_mode) && !(mode == PARAVIRT_LAZY_NONE ||
564 mode == PARAVIRT_LAZY_FLUSH));
565
566 if (mode == PARAVIRT_LAZY_FLUSH) {
567 vmi_ops.set_lazy_mode(0);
568 vmi_ops.set_lazy_mode(__get_cpu_var(lazy_mode));
569 } else {
570 vmi_ops.set_lazy_mode(mode);
571 __get_cpu_var(lazy_mode) = mode;
572 }
573}
574
575static inline int __init check_vmi_rom(struct vrom_header *rom)
576{
577 struct pci_header *pci;
578 struct pnp_header *pnp;
579 const char *manufacturer = "UNKNOWN";
580 const char *product = "UNKNOWN";
581 const char *license = "unspecified";
582
583 if (rom->rom_signature != 0xaa55)
584 return 0;
585 if (rom->vrom_signature != VMI_SIGNATURE)
586 return 0;
587 if (rom->api_version_maj != VMI_API_REV_MAJOR ||
588 rom->api_version_min+1 < VMI_API_REV_MINOR+1) {
589 printk(KERN_WARNING "VMI: Found mismatched rom version %d.%d\n",
590 rom->api_version_maj,
591 rom->api_version_min);
592 return 0;
593 }
594
595 /*
596 * Relying on the VMI_SIGNATURE field is not 100% safe, so check
597 * the PCI header and device type to make sure this is really a
598 * VMI device.
599 */
600 if (!rom->pci_header_offs) {
601 printk(KERN_WARNING "VMI: ROM does not contain PCI header.\n");
602 return 0;
603 }
604
605 pci = (struct pci_header *)((char *)rom+rom->pci_header_offs);
606 if (pci->vendorID != PCI_VENDOR_ID_VMWARE ||
607 pci->deviceID != PCI_DEVICE_ID_VMWARE_VMI) {
608 /* Allow it to run... anyways, but warn */
609 printk(KERN_WARNING "VMI: ROM from unknown manufacturer\n");
610 }
611
612 if (rom->pnp_header_offs) {
613 pnp = (struct pnp_header *)((char *)rom+rom->pnp_header_offs);
614 if (pnp->manufacturer_offset)
615 manufacturer = (const char *)rom+pnp->manufacturer_offset;
616 if (pnp->product_offset)
617 product = (const char *)rom+pnp->product_offset;
618 }
619
620 if (rom->license_offs)
621 license = (char *)rom+rom->license_offs;
622
623 printk(KERN_INFO "VMI: Found %s %s, API version %d.%d, ROM version %d.%d\n",
624 manufacturer, product,
625 rom->api_version_maj, rom->api_version_min,
626 pci->rom_version_maj, pci->rom_version_min);
627
628 /* Don't allow BSD/MIT here for now because we don't want to end up
629 with any binary only shim layers */
630 if (strcmp(license, "GPL") && strcmp(license, "GPL v2")) {
631 printk(KERN_WARNING "VMI: Non GPL license `%s' found for ROM. Not used.\n",
632 license);
633 return 0;
634 }
635
636 return 1;
637}
638
639/*
640 * Probe for the VMI option ROM
641 */
642static 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 */
661void vmi_bringup(void)
662{
663 /* We must establish the lowmem mapping for MMU ops to work */
664 if (vmi_ops.set_linear_mapping)
665 vmi_ops.set_linear_mapping(0, (void *)__PAGE_OFFSET, max_low_pfn, 0);
666}
667
668/*
669 * Return a pointer to a VMI function or NULL if unimplemented
670 */
671static 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 NULL;
681}
682
683/*
684 * Helper macro for making the VMI paravirt-ops fill code readable.
685 * For unimplemented operations, fall back to default, unless nop
686 * is returned by the ROM.
687 */
688#define para_fill(opname, vmicall) \
689do { \
690 reloc = call_vrom_long_func(vmi_rom, get_reloc, \
691 VMI_CALL_##vmicall); \
692 if (rel->type == VMI_RELOCATION_CALL_REL) \
693 paravirt_ops.opname = (void *)rel->eip; \
694 else if (rel->type == VMI_RELOCATION_NOP) \
695 paravirt_ops.opname = (void *)vmi_nop; \
696 else if (rel->type != VMI_RELOCATION_NONE) \
697 printk(KERN_WARNING "VMI: Unknown relocation " \
698 "type %d for " #vmicall"\n",\
699 rel->type); \
700} while (0)
701
702/*
703 * Helper macro for making the VMI paravirt-ops fill code readable.
704 * For cached operations which do not match the VMI ROM ABI and must
705 * go through a tranlation stub. Ignore NOPs, since it is not clear
706 * a NOP * VMI function corresponds to a NOP paravirt-op when the
707 * functions are not in 1-1 correspondence.
708 */
709#define para_wrap(opname, wrapper, cache, vmicall) \
710do { \
711 reloc = call_vrom_long_func(vmi_rom, get_reloc, \
712 VMI_CALL_##vmicall); \
713 BUG_ON(rel->type == VMI_RELOCATION_JUMP_REL); \
714 if (rel->type == VMI_RELOCATION_CALL_REL) { \
715 paravirt_ops.opname = wrapper; \
716 vmi_ops.cache = (void *)rel->eip; \
717 } \
718} while (0)
719
720/*
721 * Activate the VMI interface and switch into paravirtualized mode
722 */
723static inline int __init activate_vmi(void)
724{
725 short kernel_cs;
726 u64 reloc;
727 const struct vmi_relocation_info *rel = (struct vmi_relocation_info *)&reloc;
728
729 if (call_vrom_func(vmi_rom, vmi_init) != 0) {
730 printk(KERN_ERR "VMI ROM failed to initialize!");
731 return 0;
732 }
733 savesegment(cs, kernel_cs);
734
735 paravirt_ops.paravirt_enabled = 1;
736 paravirt_ops.kernel_rpl = kernel_cs & SEGMENT_RPL_MASK;
737
738 paravirt_ops.patch = vmi_patch;
739 paravirt_ops.name = "vmi";
740
741 /*
742 * Many of these operations are ABI compatible with VMI.
743 * This means we can fill in the paravirt-ops with direct
744 * pointers into the VMI ROM. If the calling convention for
745 * these operations changes, this code needs to be updated.
746 *
747 * Exceptions
748 * CPUID paravirt-op uses pointers, not the native ISA
749 * halt has no VMI equivalent; all VMI halts are "safe"
750 * no MSR support yet - just trap and emulate. VMI uses the
751 * same ABI as the native ISA, but Linux wants exceptions
752 * from bogus MSR read / write handled
753 * rdpmc is not yet used in Linux
754 */
755
756 /* CPUID is special, so very special it gets wrapped like a present */
757 para_wrap(cpuid, vmi_cpuid, cpuid, CPUID);
758
759 para_fill(clts, CLTS);
760 para_fill(get_debugreg, GetDR);
761 para_fill(set_debugreg, SetDR);
762 para_fill(read_cr0, GetCR0);
763 para_fill(read_cr2, GetCR2);
764 para_fill(read_cr3, GetCR3);
765 para_fill(read_cr4, GetCR4);
766 para_fill(write_cr0, SetCR0);
767 para_fill(write_cr2, SetCR2);
768 para_fill(write_cr3, SetCR3);
769 para_fill(write_cr4, SetCR4);
770 para_fill(save_fl, GetInterruptMask);
771 para_fill(restore_fl, SetInterruptMask);
772 para_fill(irq_disable, DisableInterrupts);
773 para_fill(irq_enable, EnableInterrupts);
774
775 para_fill(wbinvd, WBINVD);
776 para_fill(read_tsc, RDTSC);
777
778 /* The following we emulate with trap and emulate for now */
779 /* paravirt_ops.read_msr = vmi_rdmsr */
780 /* paravirt_ops.write_msr = vmi_wrmsr */
781 /* paravirt_ops.rdpmc = vmi_rdpmc */
782
783 /* TR interface doesn't pass TR value, wrap */
784 para_wrap(load_tr_desc, vmi_set_tr, set_tr, SetTR);
785
786 /* LDT is special, too */
787 para_wrap(set_ldt, vmi_set_ldt, _set_ldt, SetLDT);
788
789 para_fill(load_gdt, SetGDT);
790 para_fill(load_idt, SetIDT);
791 para_fill(store_gdt, GetGDT);
792 para_fill(store_idt, GetIDT);
793 para_fill(store_tr, GetTR);
794 paravirt_ops.load_tls = vmi_load_tls;
795 para_fill(write_ldt_entry, WriteLDTEntry);
796 para_fill(write_gdt_entry, WriteGDTEntry);
797 para_fill(write_idt_entry, WriteIDTEntry);
798 para_wrap(load_esp0, vmi_load_esp0, set_kernel_stack, UpdateKernelStack);
799 para_fill(set_iopl_mask, SetIOPLMask);
800 para_fill(io_delay, IODelay);
801 para_wrap(set_lazy_mode, vmi_set_lazy_mode, set_lazy_mode, SetLazyMode);
802
803 /* user and kernel flush are just handled with different flags to FlushTLB */
804 para_wrap(flush_tlb_user, vmi_flush_tlb_user, _flush_tlb, FlushTLB);
805 para_wrap(flush_tlb_kernel, vmi_flush_tlb_kernel, _flush_tlb, FlushTLB);
806 para_fill(flush_tlb_single, InvalPage);
807
808 /*
809 * Until a standard flag format can be agreed on, we need to
810 * implement these as wrappers in Linux. Get the VMI ROM
811 * function pointers for the two backend calls.
812 */
813#ifdef CONFIG_X86_PAE
814 vmi_ops.set_pte = vmi_get_function(VMI_CALL_SetPxELong);
815 vmi_ops.update_pte = vmi_get_function(VMI_CALL_UpdatePxELong);
816#else
817 vmi_ops.set_pte = vmi_get_function(VMI_CALL_SetPxE);
818 vmi_ops.update_pte = vmi_get_function(VMI_CALL_UpdatePxE);
819#endif
820
821 if (vmi_ops.set_pte) {
822 paravirt_ops.set_pte = vmi_set_pte;
823 paravirt_ops.set_pte_at = vmi_set_pte_at;
824 paravirt_ops.set_pmd = vmi_set_pmd;
825#ifdef CONFIG_X86_PAE
826 paravirt_ops.set_pte_atomic = vmi_set_pte_atomic;
827 paravirt_ops.set_pte_present = vmi_set_pte_present;
828 paravirt_ops.set_pud = vmi_set_pud;
829 paravirt_ops.pte_clear = vmi_pte_clear;
830 paravirt_ops.pmd_clear = vmi_pmd_clear;
831#endif
832 }
833
834 if (vmi_ops.update_pte) {
835 paravirt_ops.pte_update = vmi_update_pte;
836 paravirt_ops.pte_update_defer = vmi_update_pte_defer;
837 }
838
839 vmi_ops.allocate_page = vmi_get_function(VMI_CALL_AllocatePage);
840 if (vmi_ops.allocate_page) {
841 paravirt_ops.alloc_pt = vmi_allocate_pt;
842 paravirt_ops.alloc_pd = vmi_allocate_pd;
843 paravirt_ops.alloc_pd_clone = vmi_allocate_pd_clone;
844 }
845
846 vmi_ops.release_page = vmi_get_function(VMI_CALL_ReleasePage);
847 if (vmi_ops.release_page) {
848 paravirt_ops.release_pt = vmi_release_pt;
849 paravirt_ops.release_pd = vmi_release_pd;
850 }
851
852 /* Set linear is needed in all cases */
853 vmi_ops.set_linear_mapping = vmi_get_function(VMI_CALL_SetLinearMapping);
854#ifdef CONFIG_HIGHPTE
855 if (vmi_ops.set_linear_mapping)
856 paravirt_ops.kmap_atomic_pte = vmi_kmap_atomic_pte;
857#endif
858
859 /*
860 * These MUST always be patched. Don't support indirect jumps
861 * through these operations, as the VMI interface may use either
862 * a jump or a call to get to these operations, depending on
863 * the backend. They are performance critical anyway, so requiring
864 * a patch is not a big problem.
865 */
866 paravirt_ops.irq_enable_sysexit = (void *)0xfeedbab0;
867 paravirt_ops.iret = (void *)0xbadbab0;
868
869#ifdef CONFIG_SMP
870 para_wrap(startup_ipi_hook, vmi_startup_ipi_hook, set_initial_ap_state, SetInitialAPState);
871#endif
872
873#ifdef CONFIG_X86_LOCAL_APIC
874 para_fill(apic_read, APICRead);
875 para_fill(apic_write, APICWrite);
876 para_fill(apic_write_atomic, APICWrite);
877#endif
878
879 /*
880 * Check for VMI timer functionality by probing for a cycle frequency method
881 */
882 reloc = call_vrom_long_func(vmi_rom, get_reloc, VMI_CALL_GetCycleFrequency);
883 if (!disable_vmi_timer && rel->type != VMI_RELOCATION_NONE) {
884 vmi_timer_ops.get_cycle_frequency = (void *)rel->eip;
885 vmi_timer_ops.get_cycle_counter =
886 vmi_get_function(VMI_CALL_GetCycleCounter);
887 vmi_timer_ops.get_wallclock =
888 vmi_get_function(VMI_CALL_GetWallclockTime);
889 vmi_timer_ops.wallclock_updated =
890 vmi_get_function(VMI_CALL_WallclockUpdated);
891 vmi_timer_ops.set_alarm = vmi_get_function(VMI_CALL_SetAlarm);
892 vmi_timer_ops.cancel_alarm =
893 vmi_get_function(VMI_CALL_CancelAlarm);
894 paravirt_ops.time_init = vmi_time_init;
895 paravirt_ops.get_wallclock = vmi_get_wallclock;
896 paravirt_ops.set_wallclock = vmi_set_wallclock;
897#ifdef CONFIG_X86_LOCAL_APIC
898 paravirt_ops.setup_boot_clock = vmi_time_bsp_init;
899 paravirt_ops.setup_secondary_clock = vmi_time_ap_init;
900#endif
901 paravirt_ops.sched_clock = vmi_sched_clock;
902 paravirt_ops.get_cpu_khz = vmi_cpu_khz;
903
904 /* We have true wallclock functions; disable CMOS clock sync */
905 no_sync_cmos_clock = 1;
906 } else {
907 disable_noidle = 1;
908 disable_vmi_timer = 1;
909 }
910
911 para_fill(safe_halt, Halt);
912
913 /*
914 * Alternative instruction rewriting doesn't happen soon enough
915 * to convert VMI_IRET to a call instead of a jump; so we have
916 * to do this before IRQs get reenabled. Fortunately, it is
917 * idempotent.
918 */
919 apply_paravirt(__parainstructions, __parainstructions_end);
920
921 vmi_bringup();
922
923 return 1;
924}
925
926#undef para_fill
927
928void __init vmi_init(void)
929{
930 unsigned long flags;
931
932 if (!vmi_rom)
933 probe_vmi_rom();
934 else
935 check_vmi_rom(vmi_rom);
936
937 /* In case probing for or validating the ROM failed, basil */
938 if (!vmi_rom)
939 return;
940
941 reserve_top_address(-vmi_rom->virtual_top);
942
943 local_irq_save(flags);
944 activate_vmi();
945
946#ifdef CONFIG_X86_IO_APIC
947 /* This is virtual hardware; timer routing is wired correctly */
948 no_timer_check = 1;
949#endif
950 local_irq_restore(flags & X86_EFLAGS_IF);
951}
952
953static int __init parse_vmi(char *arg)
954{
955 if (!arg)
956 return -EINVAL;
957
958 if (!strcmp(arg, "disable_pge")) {
959 clear_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability);
960 disable_pge = 1;
961 } else if (!strcmp(arg, "disable_pse")) {
962 clear_bit(X86_FEATURE_PSE, boot_cpu_data.x86_capability);
963 disable_pse = 1;
964 } else if (!strcmp(arg, "disable_sep")) {
965 clear_bit(X86_FEATURE_SEP, boot_cpu_data.x86_capability);
966 disable_sep = 1;
967 } else if (!strcmp(arg, "disable_tsc")) {
968 clear_bit(X86_FEATURE_TSC, boot_cpu_data.x86_capability);
969 disable_tsc = 1;
970 } else if (!strcmp(arg, "disable_mtrr")) {
971 clear_bit(X86_FEATURE_MTRR, boot_cpu_data.x86_capability);
972 disable_mtrr = 1;
973 } else if (!strcmp(arg, "disable_timer")) {
974 disable_vmi_timer = 1;
975 disable_noidle = 1;
976 } else if (!strcmp(arg, "disable_noidle"))
977 disable_noidle = 1;
978 return 0;
979}
980
981early_param("vmi", parse_vmi);