/arch/x86/kernel/vmi_32.c
C | 893 lines | 620 code | 120 blank | 153 comment | 82 complexity | 5bafb4b4fcc51975a8784d177dfb69e1 MD5 | raw file
Possible License(s): LGPL-2.0, AGPL-1.0, GPL-2.0
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 <linux/gfp.h>
32#include <asm/vmi.h>
33#include <asm/io.h>
34#include <asm/fixmap.h>
35#include <asm/apicdef.h>
36#include <asm/apic.h>
37#include <asm/pgalloc.h>
38#include <asm/processor.h>
39#include <asm/timer.h>
40#include <asm/vmi_time.h>
41#include <asm/kmap_types.h>
42#include <asm/setup.h>
43
44/* Convenient for calling VMI functions indirectly in the ROM */
45typedef u32 __attribute__((regparm(1))) (VROMFUNC)(void);
46typedef u64 __attribute__((regparm(2))) (VROMLONGFUNC)(int);
47
48#define call_vrom_func(rom,func) \
49 (((VROMFUNC *)(rom->func))())
50
51#define call_vrom_long_func(rom,func,arg) \
52 (((VROMLONGFUNC *)(rom->func)) (arg))
53
54static struct vrom_header *vmi_rom;
55static int disable_pge;
56static int disable_pse;
57static int disable_sep;
58static int disable_tsc;
59static int disable_mtrr;
60static int disable_noidle;
61static int disable_vmi_timer;
62
63/* Cached VMI operations */
64static struct {
65 void (*cpuid)(void /* non-c */);
66 void (*_set_ldt)(u32 selector);
67 void (*set_tr)(u32 selector);
68 void (*write_idt_entry)(struct desc_struct *, int, u32, u32);
69 void (*write_gdt_entry)(struct desc_struct *, int, u32, u32);
70 void (*write_ldt_entry)(struct desc_struct *, int, u32, u32);
71 void (*set_kernel_stack)(u32 selector, u32 sp0);
72 void (*allocate_page)(u32, u32, u32, u32, u32);
73 void (*release_page)(u32, u32);
74 void (*set_pte)(pte_t, pte_t *, unsigned);
75 void (*update_pte)(pte_t *, unsigned);
76 void (*set_linear_mapping)(int, void *, u32, u32);
77 void (*_flush_tlb)(int);
78 void (*set_initial_ap_state)(int, int);
79 void (*halt)(void);
80 void (*set_lazy_mode)(int mode);
81} vmi_ops;
82
83/* Cached VMI operations */
84struct vmi_timer_ops vmi_timer_ops;
85
86/*
87 * VMI patching routines.
88 */
89#define MNEM_CALL 0xe8
90#define MNEM_JMP 0xe9
91#define MNEM_RET 0xc3
92
93#define IRQ_PATCH_INT_MASK 0
94#define IRQ_PATCH_DISABLE 5
95
96static inline void patch_offset(void *insnbuf,
97 unsigned long ip, unsigned long dest)
98{
99 *(unsigned long *)(insnbuf+1) = dest-ip-5;
100}
101
102static unsigned patch_internal(int call, unsigned len, void *insnbuf,
103 unsigned long ip)
104{
105 u64 reloc;
106 struct vmi_relocation_info *const rel = (struct vmi_relocation_info *)&reloc;
107 reloc = call_vrom_long_func(vmi_rom, get_reloc, call);
108 switch(rel->type) {
109 case VMI_RELOCATION_CALL_REL:
110 BUG_ON(len < 5);
111 *(char *)insnbuf = MNEM_CALL;
112 patch_offset(insnbuf, ip, (unsigned long)rel->eip);
113 return 5;
114
115 case VMI_RELOCATION_JUMP_REL:
116 BUG_ON(len < 5);
117 *(char *)insnbuf = MNEM_JMP;
118 patch_offset(insnbuf, ip, (unsigned long)rel->eip);
119 return 5;
120
121 case VMI_RELOCATION_NOP:
122 /* obliterate the whole thing */
123 return 0;
124
125 case VMI_RELOCATION_NONE:
126 /* leave native code in place */
127 break;
128
129 default:
130 BUG();
131 }
132 return len;
133}
134
135/*
136 * Apply patch if appropriate, return length of new instruction
137 * sequence. The callee does nop padding for us.
138 */
139static unsigned vmi_patch(u8 type, u16 clobbers, void *insns,
140 unsigned long ip, unsigned len)
141{
142 switch (type) {
143 case PARAVIRT_PATCH(pv_irq_ops.irq_disable):
144 return patch_internal(VMI_CALL_DisableInterrupts, len,
145 insns, ip);
146 case PARAVIRT_PATCH(pv_irq_ops.irq_enable):
147 return patch_internal(VMI_CALL_EnableInterrupts, len,
148 insns, ip);
149 case PARAVIRT_PATCH(pv_irq_ops.restore_fl):
150 return patch_internal(VMI_CALL_SetInterruptMask, len,
151 insns, ip);
152 case PARAVIRT_PATCH(pv_irq_ops.save_fl):
153 return patch_internal(VMI_CALL_GetInterruptMask, len,
154 insns, ip);
155 case PARAVIRT_PATCH(pv_cpu_ops.iret):
156 return patch_internal(VMI_CALL_IRET, len, insns, ip);
157 case PARAVIRT_PATCH(pv_cpu_ops.irq_enable_sysexit):
158 return patch_internal(VMI_CALL_SYSEXIT, len, insns, ip);
159 default:
160 break;
161 }
162 return len;
163}
164
165/* CPUID has non-C semantics, and paravirt-ops API doesn't match hardware ISA */
166static void vmi_cpuid(unsigned int *ax, unsigned int *bx,
167 unsigned int *cx, unsigned int *dx)
168{
169 int override = 0;
170 if (*ax == 1)
171 override = 1;
172 asm volatile ("call *%6"
173 : "=a" (*ax),
174 "=b" (*bx),
175 "=c" (*cx),
176 "=d" (*dx)
177 : "0" (*ax), "2" (*cx), "r" (vmi_ops.cpuid));
178 if (override) {
179 if (disable_pse)
180 *dx &= ~X86_FEATURE_PSE;
181 if (disable_pge)
182 *dx &= ~X86_FEATURE_PGE;
183 if (disable_sep)
184 *dx &= ~X86_FEATURE_SEP;
185 if (disable_tsc)
186 *dx &= ~X86_FEATURE_TSC;
187 if (disable_mtrr)
188 *dx &= ~X86_FEATURE_MTRR;
189 }
190}
191
192static inline void vmi_maybe_load_tls(struct desc_struct *gdt, int nr, struct desc_struct *new)
193{
194 if (gdt[nr].a != new->a || gdt[nr].b != new->b)
195 write_gdt_entry(gdt, nr, new, 0);
196}
197
198static void vmi_load_tls(struct thread_struct *t, unsigned int cpu)
199{
200 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
201 vmi_maybe_load_tls(gdt, GDT_ENTRY_TLS_MIN + 0, &t->tls_array[0]);
202 vmi_maybe_load_tls(gdt, GDT_ENTRY_TLS_MIN + 1, &t->tls_array[1]);
203 vmi_maybe_load_tls(gdt, GDT_ENTRY_TLS_MIN + 2, &t->tls_array[2]);
204}
205
206static void vmi_set_ldt(const void *addr, unsigned entries)
207{
208 unsigned cpu = smp_processor_id();
209 struct desc_struct desc;
210
211 pack_descriptor(&desc, (unsigned long)addr,
212 entries * sizeof(struct desc_struct) - 1,
213 DESC_LDT, 0);
214 write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_LDT, &desc, DESC_LDT);
215 vmi_ops._set_ldt(entries ? GDT_ENTRY_LDT*sizeof(struct desc_struct) : 0);
216}
217
218static void vmi_set_tr(void)
219{
220 vmi_ops.set_tr(GDT_ENTRY_TSS*sizeof(struct desc_struct));
221}
222
223static void vmi_write_idt_entry(gate_desc *dt, int entry, const gate_desc *g)
224{
225 u32 *idt_entry = (u32 *)g;
226 vmi_ops.write_idt_entry(dt, entry, idt_entry[0], idt_entry[1]);
227}
228
229static void vmi_write_gdt_entry(struct desc_struct *dt, int entry,
230 const void *desc, int type)
231{
232 u32 *gdt_entry = (u32 *)desc;
233 vmi_ops.write_gdt_entry(dt, entry, gdt_entry[0], gdt_entry[1]);
234}
235
236static void vmi_write_ldt_entry(struct desc_struct *dt, int entry,
237 const void *desc)
238{
239 u32 *ldt_entry = (u32 *)desc;
240 vmi_ops.write_ldt_entry(dt, entry, ldt_entry[0], ldt_entry[1]);
241}
242
243static void vmi_load_sp0(struct tss_struct *tss,
244 struct thread_struct *thread)
245{
246 tss->x86_tss.sp0 = thread->sp0;
247
248 /* This can only happen when SEP is enabled, no need to test "SEP"arately */
249 if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
250 tss->x86_tss.ss1 = thread->sysenter_cs;
251 wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
252 }
253 vmi_ops.set_kernel_stack(__KERNEL_DS, tss->x86_tss.sp0);
254}
255
256static void vmi_flush_tlb_user(void)
257{
258 vmi_ops._flush_tlb(VMI_FLUSH_TLB);
259}
260
261static void vmi_flush_tlb_kernel(void)
262{
263 vmi_ops._flush_tlb(VMI_FLUSH_TLB | VMI_FLUSH_GLOBAL);
264}
265
266/* Stub to do nothing at all; used for delays and unimplemented calls */
267static void vmi_nop(void)
268{
269}
270
271static void vmi_allocate_pte(struct mm_struct *mm, unsigned long pfn)
272{
273 vmi_ops.allocate_page(pfn, VMI_PAGE_L1, 0, 0, 0);
274}
275
276static void vmi_allocate_pmd(struct mm_struct *mm, unsigned long pfn)
277{
278 /*
279 * This call comes in very early, before mem_map is setup.
280 * It is called only for swapper_pg_dir, which already has
281 * data on it.
282 */
283 vmi_ops.allocate_page(pfn, VMI_PAGE_L2, 0, 0, 0);
284}
285
286static void vmi_allocate_pmd_clone(unsigned long pfn, unsigned long clonepfn, unsigned long start, unsigned long count)
287{
288 vmi_ops.allocate_page(pfn, VMI_PAGE_L2 | VMI_PAGE_CLONE, clonepfn, start, count);
289}
290
291static void vmi_release_pte(unsigned long pfn)
292{
293 vmi_ops.release_page(pfn, VMI_PAGE_L1);
294}
295
296static void vmi_release_pmd(unsigned long pfn)
297{
298 vmi_ops.release_page(pfn, VMI_PAGE_L2);
299}
300
301/*
302 * We use the pgd_free hook for releasing the pgd page:
303 */
304static void vmi_pgd_free(struct mm_struct *mm, pgd_t *pgd)
305{
306 unsigned long pfn = __pa(pgd) >> PAGE_SHIFT;
307
308 vmi_ops.release_page(pfn, VMI_PAGE_L2);
309}
310
311/*
312 * Helper macros for MMU update flags. We can defer updates until a flush
313 * or page invalidation only if the update is to the current address space
314 * (otherwise, there is no flush). We must check against init_mm, since
315 * this could be a kernel update, which usually passes init_mm, although
316 * sometimes this check can be skipped if we know the particular function
317 * is only called on user mode PTEs. We could change the kernel to pass
318 * current->active_mm here, but in particular, I was unsure if changing
319 * mm/highmem.c to do this would still be correct on other architectures.
320 */
321#define is_current_as(mm, mustbeuser) ((mm) == current->active_mm || \
322 (!mustbeuser && (mm) == &init_mm))
323#define vmi_flags_addr(mm, addr, level, user) \
324 ((level) | (is_current_as(mm, user) ? \
325 (VMI_PAGE_CURRENT_AS | ((addr) & VMI_PAGE_VA_MASK)) : 0))
326#define vmi_flags_addr_defer(mm, addr, level, user) \
327 ((level) | (is_current_as(mm, user) ? \
328 (VMI_PAGE_DEFER | VMI_PAGE_CURRENT_AS | ((addr) & VMI_PAGE_VA_MASK)) : 0))
329
330static void vmi_update_pte(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
331{
332 vmi_ops.update_pte(ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0));
333}
334
335static void vmi_update_pte_defer(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
336{
337 vmi_ops.update_pte(ptep, vmi_flags_addr_defer(mm, addr, VMI_PAGE_PT, 0));
338}
339
340static void vmi_set_pte(pte_t *ptep, pte_t pte)
341{
342 /* XXX because of set_pmd_pte, this can be called on PT or PD layers */
343 vmi_ops.set_pte(pte, ptep, VMI_PAGE_PT);
344}
345
346static void vmi_set_pte_at(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pte)
347{
348 vmi_ops.set_pte(pte, ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0));
349}
350
351static void vmi_set_pmd(pmd_t *pmdp, pmd_t pmdval)
352{
353#ifdef CONFIG_X86_PAE
354 const pte_t pte = { .pte = pmdval.pmd };
355#else
356 const pte_t pte = { pmdval.pud.pgd.pgd };
357#endif
358 vmi_ops.set_pte(pte, (pte_t *)pmdp, VMI_PAGE_PD);
359}
360
361#ifdef CONFIG_X86_PAE
362
363static void vmi_set_pte_atomic(pte_t *ptep, pte_t pteval)
364{
365 /*
366 * XXX This is called from set_pmd_pte, but at both PT
367 * and PD layers so the VMI_PAGE_PT flag is wrong. But
368 * it is only called for large page mapping changes,
369 * the Xen backend, doesn't support large pages, and the
370 * ESX backend doesn't depend on the flag.
371 */
372 set_64bit((unsigned long long *)ptep,pte_val(pteval));
373 vmi_ops.update_pte(ptep, VMI_PAGE_PT);
374}
375
376static void vmi_set_pud(pud_t *pudp, pud_t pudval)
377{
378 /* Um, eww */
379 const pte_t pte = { .pte = pudval.pgd.pgd };
380 vmi_ops.set_pte(pte, (pte_t *)pudp, VMI_PAGE_PDP);
381}
382
383static void vmi_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
384{
385 const pte_t pte = { .pte = 0 };
386 vmi_ops.set_pte(pte, ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0));
387}
388
389static void vmi_pmd_clear(pmd_t *pmd)
390{
391 const pte_t pte = { .pte = 0 };
392 vmi_ops.set_pte(pte, (pte_t *)pmd, VMI_PAGE_PD);
393}
394#endif
395
396#ifdef CONFIG_SMP
397static void __devinit
398vmi_startup_ipi_hook(int phys_apicid, unsigned long start_eip,
399 unsigned long start_esp)
400{
401 struct vmi_ap_state ap;
402
403 /* Default everything to zero. This is fine for most GPRs. */
404 memset(&ap, 0, sizeof(struct vmi_ap_state));
405
406 ap.gdtr_limit = GDT_SIZE - 1;
407 ap.gdtr_base = (unsigned long) get_cpu_gdt_table(phys_apicid);
408
409 ap.idtr_limit = IDT_ENTRIES * 8 - 1;
410 ap.idtr_base = (unsigned long) idt_table;
411
412 ap.ldtr = 0;
413
414 ap.cs = __KERNEL_CS;
415 ap.eip = (unsigned long) start_eip;
416 ap.ss = __KERNEL_DS;
417 ap.esp = (unsigned long) start_esp;
418
419 ap.ds = __USER_DS;
420 ap.es = __USER_DS;
421 ap.fs = __KERNEL_PERCPU;
422 ap.gs = __KERNEL_STACK_CANARY;
423
424 ap.eflags = 0;
425
426#ifdef CONFIG_X86_PAE
427 /* efer should match BSP efer. */
428 if (cpu_has_nx) {
429 unsigned l, h;
430 rdmsr(MSR_EFER, l, h);
431 ap.efer = (unsigned long long) h << 32 | l;
432 }
433#endif
434
435 ap.cr3 = __pa(swapper_pg_dir);
436 /* Protected mode, paging, AM, WP, NE, MP. */
437 ap.cr0 = 0x80050023;
438 ap.cr4 = mmu_cr4_features;
439 vmi_ops.set_initial_ap_state((u32)&ap, phys_apicid);
440}
441#endif
442
443static void vmi_start_context_switch(struct task_struct *prev)
444{
445 paravirt_start_context_switch(prev);
446 vmi_ops.set_lazy_mode(2);
447}
448
449static void vmi_end_context_switch(struct task_struct *next)
450{
451 vmi_ops.set_lazy_mode(0);
452 paravirt_end_context_switch(next);
453}
454
455static void vmi_enter_lazy_mmu(void)
456{
457 paravirt_enter_lazy_mmu();
458 vmi_ops.set_lazy_mode(1);
459}
460
461static void vmi_leave_lazy_mmu(void)
462{
463 vmi_ops.set_lazy_mode(0);
464 paravirt_leave_lazy_mmu();
465}
466
467static inline int __init check_vmi_rom(struct vrom_header *rom)
468{
469 struct pci_header *pci;
470 struct pnp_header *pnp;
471 const char *manufacturer = "UNKNOWN";
472 const char *product = "UNKNOWN";
473 const char *license = "unspecified";
474
475 if (rom->rom_signature != 0xaa55)
476 return 0;
477 if (rom->vrom_signature != VMI_SIGNATURE)
478 return 0;
479 if (rom->api_version_maj != VMI_API_REV_MAJOR ||
480 rom->api_version_min+1 < VMI_API_REV_MINOR+1) {
481 printk(KERN_WARNING "VMI: Found mismatched rom version %d.%d\n",
482 rom->api_version_maj,
483 rom->api_version_min);
484 return 0;
485 }
486
487 /*
488 * Relying on the VMI_SIGNATURE field is not 100% safe, so check
489 * the PCI header and device type to make sure this is really a
490 * VMI device.
491 */
492 if (!rom->pci_header_offs) {
493 printk(KERN_WARNING "VMI: ROM does not contain PCI header.\n");
494 return 0;
495 }
496
497 pci = (struct pci_header *)((char *)rom+rom->pci_header_offs);
498 if (pci->vendorID != PCI_VENDOR_ID_VMWARE ||
499 pci->deviceID != PCI_DEVICE_ID_VMWARE_VMI) {
500 /* Allow it to run... anyways, but warn */
501 printk(KERN_WARNING "VMI: ROM from unknown manufacturer\n");
502 }
503
504 if (rom->pnp_header_offs) {
505 pnp = (struct pnp_header *)((char *)rom+rom->pnp_header_offs);
506 if (pnp->manufacturer_offset)
507 manufacturer = (const char *)rom+pnp->manufacturer_offset;
508 if (pnp->product_offset)
509 product = (const char *)rom+pnp->product_offset;
510 }
511
512 if (rom->license_offs)
513 license = (char *)rom+rom->license_offs;
514
515 printk(KERN_INFO "VMI: Found %s %s, API version %d.%d, ROM version %d.%d\n",
516 manufacturer, product,
517 rom->api_version_maj, rom->api_version_min,
518 pci->rom_version_maj, pci->rom_version_min);
519
520 /* Don't allow BSD/MIT here for now because we don't want to end up
521 with any binary only shim layers */
522 if (strcmp(license, "GPL") && strcmp(license, "GPL v2")) {
523 printk(KERN_WARNING "VMI: Non GPL license `%s' found for ROM. Not used.\n",
524 license);
525 return 0;
526 }
527
528 return 1;
529}
530
531/*
532 * Probe for the VMI option ROM
533 */
534static inline int __init probe_vmi_rom(void)
535{
536 unsigned long base;
537
538 /* VMI ROM is in option ROM area, check signature */
539 for (base = 0xC0000; base < 0xE0000; base += 2048) {
540 struct vrom_header *romstart;
541 romstart = (struct vrom_header *)isa_bus_to_virt(base);
542 if (check_vmi_rom(romstart)) {
543 vmi_rom = romstart;
544 return 1;
545 }
546 }
547 return 0;
548}
549
550/*
551 * VMI setup common to all processors
552 */
553void vmi_bringup(void)
554{
555 /* We must establish the lowmem mapping for MMU ops to work */
556 if (vmi_ops.set_linear_mapping)
557 vmi_ops.set_linear_mapping(0, (void *)__PAGE_OFFSET, MAXMEM_PFN, 0);
558}
559
560/*
561 * Return a pointer to a VMI function or NULL if unimplemented
562 */
563static void *vmi_get_function(int vmicall)
564{
565 u64 reloc;
566 const struct vmi_relocation_info *rel = (struct vmi_relocation_info *)&reloc;
567 reloc = call_vrom_long_func(vmi_rom, get_reloc, vmicall);
568 BUG_ON(rel->type == VMI_RELOCATION_JUMP_REL);
569 if (rel->type == VMI_RELOCATION_CALL_REL)
570 return (void *)rel->eip;
571 else
572 return NULL;
573}
574
575/*
576 * Helper macro for making the VMI paravirt-ops fill code readable.
577 * For unimplemented operations, fall back to default, unless nop
578 * is returned by the ROM.
579 */
580#define para_fill(opname, vmicall) \
581do { \
582 reloc = call_vrom_long_func(vmi_rom, get_reloc, \
583 VMI_CALL_##vmicall); \
584 if (rel->type == VMI_RELOCATION_CALL_REL) \
585 opname = (void *)rel->eip; \
586 else if (rel->type == VMI_RELOCATION_NOP) \
587 opname = (void *)vmi_nop; \
588 else if (rel->type != VMI_RELOCATION_NONE) \
589 printk(KERN_WARNING "VMI: Unknown relocation " \
590 "type %d for " #vmicall"\n",\
591 rel->type); \
592} while (0)
593
594/*
595 * Helper macro for making the VMI paravirt-ops fill code readable.
596 * For cached operations which do not match the VMI ROM ABI and must
597 * go through a tranlation stub. Ignore NOPs, since it is not clear
598 * a NOP * VMI function corresponds to a NOP paravirt-op when the
599 * functions are not in 1-1 correspondence.
600 */
601#define para_wrap(opname, wrapper, cache, vmicall) \
602do { \
603 reloc = call_vrom_long_func(vmi_rom, get_reloc, \
604 VMI_CALL_##vmicall); \
605 BUG_ON(rel->type == VMI_RELOCATION_JUMP_REL); \
606 if (rel->type == VMI_RELOCATION_CALL_REL) { \
607 opname = wrapper; \
608 vmi_ops.cache = (void *)rel->eip; \
609 } \
610} while (0)
611
612/*
613 * Activate the VMI interface and switch into paravirtualized mode
614 */
615static inline int __init activate_vmi(void)
616{
617 short kernel_cs;
618 u64 reloc;
619 const struct vmi_relocation_info *rel = (struct vmi_relocation_info *)&reloc;
620
621 /*
622 * Prevent page tables from being allocated in highmem, even if
623 * CONFIG_HIGHPTE is enabled.
624 */
625 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
626
627 if (call_vrom_func(vmi_rom, vmi_init) != 0) {
628 printk(KERN_ERR "VMI ROM failed to initialize!");
629 return 0;
630 }
631 savesegment(cs, kernel_cs);
632
633 pv_info.paravirt_enabled = 1;
634 pv_info.kernel_rpl = kernel_cs & SEGMENT_RPL_MASK;
635 pv_info.name = "vmi [deprecated]";
636
637 pv_init_ops.patch = vmi_patch;
638
639 /*
640 * Many of these operations are ABI compatible with VMI.
641 * This means we can fill in the paravirt-ops with direct
642 * pointers into the VMI ROM. If the calling convention for
643 * these operations changes, this code needs to be updated.
644 *
645 * Exceptions
646 * CPUID paravirt-op uses pointers, not the native ISA
647 * halt has no VMI equivalent; all VMI halts are "safe"
648 * no MSR support yet - just trap and emulate. VMI uses the
649 * same ABI as the native ISA, but Linux wants exceptions
650 * from bogus MSR read / write handled
651 * rdpmc is not yet used in Linux
652 */
653
654 /* CPUID is special, so very special it gets wrapped like a present */
655 para_wrap(pv_cpu_ops.cpuid, vmi_cpuid, cpuid, CPUID);
656
657 para_fill(pv_cpu_ops.clts, CLTS);
658 para_fill(pv_cpu_ops.get_debugreg, GetDR);
659 para_fill(pv_cpu_ops.set_debugreg, SetDR);
660 para_fill(pv_cpu_ops.read_cr0, GetCR0);
661 para_fill(pv_mmu_ops.read_cr2, GetCR2);
662 para_fill(pv_mmu_ops.read_cr3, GetCR3);
663 para_fill(pv_cpu_ops.read_cr4, GetCR4);
664 para_fill(pv_cpu_ops.write_cr0, SetCR0);
665 para_fill(pv_mmu_ops.write_cr2, SetCR2);
666 para_fill(pv_mmu_ops.write_cr3, SetCR3);
667 para_fill(pv_cpu_ops.write_cr4, SetCR4);
668
669 para_fill(pv_irq_ops.save_fl.func, GetInterruptMask);
670 para_fill(pv_irq_ops.restore_fl.func, SetInterruptMask);
671 para_fill(pv_irq_ops.irq_disable.func, DisableInterrupts);
672 para_fill(pv_irq_ops.irq_enable.func, EnableInterrupts);
673
674 para_fill(pv_cpu_ops.wbinvd, WBINVD);
675 para_fill(pv_cpu_ops.read_tsc, RDTSC);
676
677 /* The following we emulate with trap and emulate for now */
678 /* paravirt_ops.read_msr = vmi_rdmsr */
679 /* paravirt_ops.write_msr = vmi_wrmsr */
680 /* paravirt_ops.rdpmc = vmi_rdpmc */
681
682 /* TR interface doesn't pass TR value, wrap */
683 para_wrap(pv_cpu_ops.load_tr_desc, vmi_set_tr, set_tr, SetTR);
684
685 /* LDT is special, too */
686 para_wrap(pv_cpu_ops.set_ldt, vmi_set_ldt, _set_ldt, SetLDT);
687
688 para_fill(pv_cpu_ops.load_gdt, SetGDT);
689 para_fill(pv_cpu_ops.load_idt, SetIDT);
690 para_fill(pv_cpu_ops.store_gdt, GetGDT);
691 para_fill(pv_cpu_ops.store_idt, GetIDT);
692 para_fill(pv_cpu_ops.store_tr, GetTR);
693 pv_cpu_ops.load_tls = vmi_load_tls;
694 para_wrap(pv_cpu_ops.write_ldt_entry, vmi_write_ldt_entry,
695 write_ldt_entry, WriteLDTEntry);
696 para_wrap(pv_cpu_ops.write_gdt_entry, vmi_write_gdt_entry,
697 write_gdt_entry, WriteGDTEntry);
698 para_wrap(pv_cpu_ops.write_idt_entry, vmi_write_idt_entry,
699 write_idt_entry, WriteIDTEntry);
700 para_wrap(pv_cpu_ops.load_sp0, vmi_load_sp0, set_kernel_stack, UpdateKernelStack);
701 para_fill(pv_cpu_ops.set_iopl_mask, SetIOPLMask);
702 para_fill(pv_cpu_ops.io_delay, IODelay);
703
704 para_wrap(pv_cpu_ops.start_context_switch, vmi_start_context_switch,
705 set_lazy_mode, SetLazyMode);
706 para_wrap(pv_cpu_ops.end_context_switch, vmi_end_context_switch,
707 set_lazy_mode, SetLazyMode);
708
709 para_wrap(pv_mmu_ops.lazy_mode.enter, vmi_enter_lazy_mmu,
710 set_lazy_mode, SetLazyMode);
711 para_wrap(pv_mmu_ops.lazy_mode.leave, vmi_leave_lazy_mmu,
712 set_lazy_mode, SetLazyMode);
713
714 /* user and kernel flush are just handled with different flags to FlushTLB */
715 para_wrap(pv_mmu_ops.flush_tlb_user, vmi_flush_tlb_user, _flush_tlb, FlushTLB);
716 para_wrap(pv_mmu_ops.flush_tlb_kernel, vmi_flush_tlb_kernel, _flush_tlb, FlushTLB);
717 para_fill(pv_mmu_ops.flush_tlb_single, InvalPage);
718
719 /*
720 * Until a standard flag format can be agreed on, we need to
721 * implement these as wrappers in Linux. Get the VMI ROM
722 * function pointers for the two backend calls.
723 */
724#ifdef CONFIG_X86_PAE
725 vmi_ops.set_pte = vmi_get_function(VMI_CALL_SetPxELong);
726 vmi_ops.update_pte = vmi_get_function(VMI_CALL_UpdatePxELong);
727#else
728 vmi_ops.set_pte = vmi_get_function(VMI_CALL_SetPxE);
729 vmi_ops.update_pte = vmi_get_function(VMI_CALL_UpdatePxE);
730#endif
731
732 if (vmi_ops.set_pte) {
733 pv_mmu_ops.set_pte = vmi_set_pte;
734 pv_mmu_ops.set_pte_at = vmi_set_pte_at;
735 pv_mmu_ops.set_pmd = vmi_set_pmd;
736#ifdef CONFIG_X86_PAE
737 pv_mmu_ops.set_pte_atomic = vmi_set_pte_atomic;
738 pv_mmu_ops.set_pud = vmi_set_pud;
739 pv_mmu_ops.pte_clear = vmi_pte_clear;
740 pv_mmu_ops.pmd_clear = vmi_pmd_clear;
741#endif
742 }
743
744 if (vmi_ops.update_pte) {
745 pv_mmu_ops.pte_update = vmi_update_pte;
746 pv_mmu_ops.pte_update_defer = vmi_update_pte_defer;
747 }
748
749 vmi_ops.allocate_page = vmi_get_function(VMI_CALL_AllocatePage);
750 if (vmi_ops.allocate_page) {
751 pv_mmu_ops.alloc_pte = vmi_allocate_pte;
752 pv_mmu_ops.alloc_pmd = vmi_allocate_pmd;
753 pv_mmu_ops.alloc_pmd_clone = vmi_allocate_pmd_clone;
754 }
755
756 vmi_ops.release_page = vmi_get_function(VMI_CALL_ReleasePage);
757 if (vmi_ops.release_page) {
758 pv_mmu_ops.release_pte = vmi_release_pte;
759 pv_mmu_ops.release_pmd = vmi_release_pmd;
760 pv_mmu_ops.pgd_free = vmi_pgd_free;
761 }
762
763 /* Set linear is needed in all cases */
764 vmi_ops.set_linear_mapping = vmi_get_function(VMI_CALL_SetLinearMapping);
765
766 /*
767 * These MUST always be patched. Don't support indirect jumps
768 * through these operations, as the VMI interface may use either
769 * a jump or a call to get to these operations, depending on
770 * the backend. They are performance critical anyway, so requiring
771 * a patch is not a big problem.
772 */
773 pv_cpu_ops.irq_enable_sysexit = (void *)0xfeedbab0;
774 pv_cpu_ops.iret = (void *)0xbadbab0;
775
776#ifdef CONFIG_SMP
777 para_wrap(pv_apic_ops.startup_ipi_hook, vmi_startup_ipi_hook, set_initial_ap_state, SetInitialAPState);
778#endif
779
780#ifdef CONFIG_X86_LOCAL_APIC
781 para_fill(apic->read, APICRead);
782 para_fill(apic->write, APICWrite);
783#endif
784
785 /*
786 * Check for VMI timer functionality by probing for a cycle frequency method
787 */
788 reloc = call_vrom_long_func(vmi_rom, get_reloc, VMI_CALL_GetCycleFrequency);
789 if (!disable_vmi_timer && rel->type != VMI_RELOCATION_NONE) {
790 vmi_timer_ops.get_cycle_frequency = (void *)rel->eip;
791 vmi_timer_ops.get_cycle_counter =
792 vmi_get_function(VMI_CALL_GetCycleCounter);
793 vmi_timer_ops.get_wallclock =
794 vmi_get_function(VMI_CALL_GetWallclockTime);
795 vmi_timer_ops.wallclock_updated =
796 vmi_get_function(VMI_CALL_WallclockUpdated);
797 vmi_timer_ops.set_alarm = vmi_get_function(VMI_CALL_SetAlarm);
798 vmi_timer_ops.cancel_alarm =
799 vmi_get_function(VMI_CALL_CancelAlarm);
800 x86_init.timers.timer_init = vmi_time_init;
801#ifdef CONFIG_X86_LOCAL_APIC
802 x86_init.timers.setup_percpu_clockev = vmi_time_bsp_init;
803 x86_cpuinit.setup_percpu_clockev = vmi_time_ap_init;
804#endif
805 pv_time_ops.sched_clock = vmi_sched_clock;
806 x86_platform.calibrate_tsc = vmi_tsc_khz;
807 x86_platform.get_wallclock = vmi_get_wallclock;
808 x86_platform.set_wallclock = vmi_set_wallclock;
809
810 /* We have true wallclock functions; disable CMOS clock sync */
811 no_sync_cmos_clock = 1;
812 } else {
813 disable_noidle = 1;
814 disable_vmi_timer = 1;
815 }
816
817 para_fill(pv_irq_ops.safe_halt, Halt);
818
819 /*
820 * Alternative instruction rewriting doesn't happen soon enough
821 * to convert VMI_IRET to a call instead of a jump; so we have
822 * to do this before IRQs get reenabled. Fortunately, it is
823 * idempotent.
824 */
825 apply_paravirt(__parainstructions, __parainstructions_end);
826
827 vmi_bringup();
828
829 return 1;
830}
831
832#undef para_fill
833
834void __init vmi_init(void)
835{
836 if (!vmi_rom)
837 probe_vmi_rom();
838 else
839 check_vmi_rom(vmi_rom);
840
841 /* In case probing for or validating the ROM failed, basil */
842 if (!vmi_rom)
843 return;
844
845 reserve_top_address(-vmi_rom->virtual_top);
846
847#ifdef CONFIG_X86_IO_APIC
848 /* This is virtual hardware; timer routing is wired correctly */
849 no_timer_check = 1;
850#endif
851}
852
853void __init vmi_activate(void)
854{
855 unsigned long flags;
856
857 if (!vmi_rom)
858 return;
859
860 local_irq_save(flags);
861 activate_vmi();
862 local_irq_restore(flags & X86_EFLAGS_IF);
863}
864
865static int __init parse_vmi(char *arg)
866{
867 if (!arg)
868 return -EINVAL;
869
870 if (!strcmp(arg, "disable_pge")) {
871 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_PGE);
872 disable_pge = 1;
873 } else if (!strcmp(arg, "disable_pse")) {
874 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_PSE);
875 disable_pse = 1;
876 } else if (!strcmp(arg, "disable_sep")) {
877 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_SEP);
878 disable_sep = 1;
879 } else if (!strcmp(arg, "disable_tsc")) {
880 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_TSC);
881 disable_tsc = 1;
882 } else if (!strcmp(arg, "disable_mtrr")) {
883 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_MTRR);
884 disable_mtrr = 1;
885 } else if (!strcmp(arg, "disable_timer")) {
886 disable_vmi_timer = 1;
887 disable_noidle = 1;
888 } else if (!strcmp(arg, "disable_noidle"))
889 disable_noidle = 1;
890 return 0;
891}
892
893early_param("vmi", parse_vmi);