blob: d430e4c3019300d02cdd0854d74e444b74e2a7b2 [file] [log] [blame]
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +01001/*
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>
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +010026#include <linux/cpu.h>
27#include <linux/bootmem.h>
28#include <linux/mm.h>
Zachary Amsdeneeef9c62007-05-02 19:27:16 +020029#include <linux/highmem.h>
Alexey Dobriyanfa0aa862007-06-01 00:46:27 -070030#include <linux/sched.h>
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +010031#include <asm/vmi.h>
32#include <asm/io.h>
33#include <asm/fixmap.h>
34#include <asm/apicdef.h>
Ingo Molnar7b6aa332009-02-17 13:58:15 +010035#include <asm/apic.h>
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +010036#include <asm/processor.h>
37#include <asm/timer.h>
Zachary Amsdenbbab4f32007-02-13 13:26:21 +010038#include <asm/vmi_time.h>
Adrian Bunk8f485612007-03-05 00:30:56 -080039#include <asm/kmap_types.h>
Alok Kataria31343d82008-08-08 12:15:57 -070040#include <asm/setup.h>
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +010041
42/* Convenient for calling VMI functions indirectly in the ROM */
43typedef u32 __attribute__((regparm(1))) (VROMFUNC)(void);
44typedef u64 __attribute__((regparm(2))) (VROMLONGFUNC)(int);
45
46#define call_vrom_func(rom,func) \
47 (((VROMFUNC *)(rom->func))())
48
49#define call_vrom_long_func(rom,func,arg) \
50 (((VROMLONGFUNC *)(rom->func)) (arg))
51
52static struct vrom_header *vmi_rom;
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +010053static int disable_pge;
54static int disable_pse;
55static int disable_sep;
56static int disable_tsc;
57static int disable_mtrr;
Zachary Amsden7507ba32007-03-05 00:30:34 -080058static int disable_noidle;
Zachary Amsden772205f2007-03-05 00:30:41 -080059static int disable_vmi_timer;
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +010060
61/* Cached VMI operations */
Adrian Bunk30a15282007-05-02 19:27:08 +020062static struct {
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +010063 void (*cpuid)(void /* non-c */);
64 void (*_set_ldt)(u32 selector);
65 void (*set_tr)(u32 selector);
Glauber de Oliveira Costa8d947342008-01-30 13:31:12 +010066 void (*write_idt_entry)(struct desc_struct *, int, u32, u32);
Glauber de Oliveira Costa014b15b2008-01-30 13:31:13 +010067 void (*write_gdt_entry)(struct desc_struct *, int, u32, u32);
Glauber de Oliveira Costa75b8bb32008-01-30 13:31:13 +010068 void (*write_ldt_entry)(struct desc_struct *, int, u32, u32);
H. Peter Anvinfaca6222008-01-30 13:31:02 +010069 void (*set_kernel_stack)(u32 selector, u32 sp0);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +010070 void (*allocate_page)(u32, u32, u32, u32, u32);
71 void (*release_page)(u32, u32);
72 void (*set_pte)(pte_t, pte_t *, unsigned);
73 void (*update_pte)(pte_t *, unsigned);
Zachary Amsdeneeef9c62007-05-02 19:27:16 +020074 void (*set_linear_mapping)(int, void *, u32, u32);
75 void (*_flush_tlb)(int);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +010076 void (*set_initial_ap_state)(int, int);
Zachary Amsdenbbab4f32007-02-13 13:26:21 +010077 void (*halt)(void);
Zachary Amsden49f19712007-04-08 16:04:01 -070078 void (*set_lazy_mode)(int mode);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +010079} vmi_ops;
80
Zachary Amsdene0bb8642007-05-02 19:27:16 +020081/* Cached VMI operations */
82struct vmi_timer_ops vmi_timer_ops;
83
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +010084/*
85 * VMI patching routines.
86 */
87#define MNEM_CALL 0xe8
88#define MNEM_JMP 0xe9
89#define MNEM_RET 0xc3
90
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +010091#define IRQ_PATCH_INT_MASK 0
92#define IRQ_PATCH_DISABLE 5
93
Andi Kleenab144f52007-08-10 22:31:03 +020094static inline void patch_offset(void *insnbuf,
H. Peter Anvin65ea5b02008-01-30 13:30:56 +010095 unsigned long ip, unsigned long dest)
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +010096{
H. Peter Anvin65ea5b02008-01-30 13:30:56 +010097 *(unsigned long *)(insnbuf+1) = dest-ip-5;
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +010098}
99
Andi Kleenab144f52007-08-10 22:31:03 +0200100static unsigned patch_internal(int call, unsigned len, void *insnbuf,
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100101 unsigned long ip)
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100102{
103 u64 reloc;
104 struct vmi_relocation_info *const rel = (struct vmi_relocation_info *)&reloc;
105 reloc = call_vrom_long_func(vmi_rom, get_reloc, call);
106 switch(rel->type) {
107 case VMI_RELOCATION_CALL_REL:
108 BUG_ON(len < 5);
Andi Kleenab144f52007-08-10 22:31:03 +0200109 *(char *)insnbuf = MNEM_CALL;
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100110 patch_offset(insnbuf, ip, (unsigned long)rel->eip);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100111 return 5;
112
113 case VMI_RELOCATION_JUMP_REL:
114 BUG_ON(len < 5);
Andi Kleenab144f52007-08-10 22:31:03 +0200115 *(char *)insnbuf = MNEM_JMP;
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100116 patch_offset(insnbuf, ip, (unsigned long)rel->eip);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100117 return 5;
118
119 case VMI_RELOCATION_NOP:
120 /* obliterate the whole thing */
121 return 0;
122
123 case VMI_RELOCATION_NONE:
124 /* leave native code in place */
125 break;
126
127 default:
128 BUG();
129 }
130 return len;
131}
132
133/*
134 * Apply patch if appropriate, return length of new instruction
135 * sequence. The callee does nop padding for us.
136 */
Andi Kleenab144f52007-08-10 22:31:03 +0200137static unsigned vmi_patch(u8 type, u16 clobbers, void *insns,
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100138 unsigned long ip, unsigned len)
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100139{
140 switch (type) {
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700141 case PARAVIRT_PATCH(pv_irq_ops.irq_disable):
Andi Kleenab144f52007-08-10 22:31:03 +0200142 return patch_internal(VMI_CALL_DisableInterrupts, len,
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100143 insns, ip);
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700144 case PARAVIRT_PATCH(pv_irq_ops.irq_enable):
Andi Kleenab144f52007-08-10 22:31:03 +0200145 return patch_internal(VMI_CALL_EnableInterrupts, len,
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100146 insns, ip);
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700147 case PARAVIRT_PATCH(pv_irq_ops.restore_fl):
Andi Kleenab144f52007-08-10 22:31:03 +0200148 return patch_internal(VMI_CALL_SetInterruptMask, len,
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100149 insns, ip);
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700150 case PARAVIRT_PATCH(pv_irq_ops.save_fl):
Andi Kleenab144f52007-08-10 22:31:03 +0200151 return patch_internal(VMI_CALL_GetInterruptMask, len,
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100152 insns, ip);
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700153 case PARAVIRT_PATCH(pv_cpu_ops.iret):
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100154 return patch_internal(VMI_CALL_IRET, len, insns, ip);
Jeremy Fitzhardinged75cd222008-06-25 00:19:26 -0400155 case PARAVIRT_PATCH(pv_cpu_ops.irq_enable_sysexit):
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100156 return patch_internal(VMI_CALL_SYSEXIT, len, insns, ip);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100157 default:
158 break;
159 }
160 return len;
161}
162
163/* CPUID has non-C semantics, and paravirt-ops API doesn't match hardware ISA */
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100164static void vmi_cpuid(unsigned int *ax, unsigned int *bx,
165 unsigned int *cx, unsigned int *dx)
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100166{
167 int override = 0;
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100168 if (*ax == 1)
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100169 override = 1;
170 asm volatile ("call *%6"
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100171 : "=a" (*ax),
172 "=b" (*bx),
173 "=c" (*cx),
174 "=d" (*dx)
175 : "0" (*ax), "2" (*cx), "r" (vmi_ops.cpuid));
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100176 if (override) {
177 if (disable_pse)
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100178 *dx &= ~X86_FEATURE_PSE;
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100179 if (disable_pge)
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100180 *dx &= ~X86_FEATURE_PGE;
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100181 if (disable_sep)
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100182 *dx &= ~X86_FEATURE_SEP;
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100183 if (disable_tsc)
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100184 *dx &= ~X86_FEATURE_TSC;
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100185 if (disable_mtrr)
H. Peter Anvin65ea5b02008-01-30 13:30:56 +0100186 *dx &= ~X86_FEATURE_MTRR;
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100187 }
188}
189
190static inline void vmi_maybe_load_tls(struct desc_struct *gdt, int nr, struct desc_struct *new)
191{
192 if (gdt[nr].a != new->a || gdt[nr].b != new->b)
Glauber de Oliveira Costa014b15b2008-01-30 13:31:13 +0100193 write_gdt_entry(gdt, nr, new, 0);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100194}
195
196static void vmi_load_tls(struct thread_struct *t, unsigned int cpu)
197{
198 struct desc_struct *gdt = get_cpu_gdt_table(cpu);
199 vmi_maybe_load_tls(gdt, GDT_ENTRY_TLS_MIN + 0, &t->tls_array[0]);
200 vmi_maybe_load_tls(gdt, GDT_ENTRY_TLS_MIN + 1, &t->tls_array[1]);
201 vmi_maybe_load_tls(gdt, GDT_ENTRY_TLS_MIN + 2, &t->tls_array[2]);
202}
203
204static void vmi_set_ldt(const void *addr, unsigned entries)
205{
206 unsigned cpu = smp_processor_id();
Glauber de Oliveira Costa014b15b2008-01-30 13:31:13 +0100207 struct desc_struct desc;
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100208
Glauber de Oliveira Costa014b15b2008-01-30 13:31:13 +0100209 pack_descriptor(&desc, (unsigned long)addr,
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100210 entries * sizeof(struct desc_struct) - 1,
Glauber de Oliveira Costa014b15b2008-01-30 13:31:13 +0100211 DESC_LDT, 0);
212 write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_LDT, &desc, DESC_LDT);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100213 vmi_ops._set_ldt(entries ? GDT_ENTRY_LDT*sizeof(struct desc_struct) : 0);
214}
215
216static void vmi_set_tr(void)
217{
218 vmi_ops.set_tr(GDT_ENTRY_TSS*sizeof(struct desc_struct));
219}
220
Glauber de Oliveira Costa8d947342008-01-30 13:31:12 +0100221static void vmi_write_idt_entry(gate_desc *dt, int entry, const gate_desc *g)
222{
223 u32 *idt_entry = (u32 *)g;
Ingo Molnar262d5ee2008-02-04 16:47:54 +0100224 vmi_ops.write_idt_entry(dt, entry, idt_entry[0], idt_entry[1]);
Glauber de Oliveira Costa8d947342008-01-30 13:31:12 +0100225}
226
Glauber de Oliveira Costa014b15b2008-01-30 13:31:13 +0100227static void vmi_write_gdt_entry(struct desc_struct *dt, int entry,
228 const void *desc, int type)
229{
230 u32 *gdt_entry = (u32 *)desc;
Ingo Molnar262d5ee2008-02-04 16:47:54 +0100231 vmi_ops.write_gdt_entry(dt, entry, gdt_entry[0], gdt_entry[1]);
Glauber de Oliveira Costa014b15b2008-01-30 13:31:13 +0100232}
233
Glauber de Oliveira Costa75b8bb32008-01-30 13:31:13 +0100234static void vmi_write_ldt_entry(struct desc_struct *dt, int entry,
235 const void *desc)
236{
237 u32 *ldt_entry = (u32 *)desc;
Zachary Amsdende599852008-09-30 11:02:12 -0700238 vmi_ops.write_ldt_entry(dt, entry, ldt_entry[0], ldt_entry[1]);
Glauber de Oliveira Costa75b8bb32008-01-30 13:31:13 +0100239}
240
H. Peter Anvinfaca6222008-01-30 13:31:02 +0100241static void vmi_load_sp0(struct tss_struct *tss,
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100242 struct thread_struct *thread)
243{
H. Peter Anvinfaca6222008-01-30 13:31:02 +0100244 tss->x86_tss.sp0 = thread->sp0;
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100245
246 /* This can only happen when SEP is enabled, no need to test "SEP"arately */
Rusty Russella75c54f2007-05-02 19:27:13 +0200247 if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
248 tss->x86_tss.ss1 = thread->sysenter_cs;
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100249 wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
250 }
H. Peter Anvinfaca6222008-01-30 13:31:02 +0100251 vmi_ops.set_kernel_stack(__KERNEL_DS, tss->x86_tss.sp0);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100252}
253
254static void vmi_flush_tlb_user(void)
255{
Zachary Amsdeneeef9c62007-05-02 19:27:16 +0200256 vmi_ops._flush_tlb(VMI_FLUSH_TLB);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100257}
258
259static void vmi_flush_tlb_kernel(void)
260{
Zachary Amsdeneeef9c62007-05-02 19:27:16 +0200261 vmi_ops._flush_tlb(VMI_FLUSH_TLB | VMI_FLUSH_GLOBAL);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100262}
263
264/* Stub to do nothing at all; used for delays and unimplemented calls */
265static void vmi_nop(void)
266{
267}
268
Zachary Amsdeneeef9c62007-05-02 19:27:16 +0200269#ifdef CONFIG_HIGHPTE
270static void *vmi_kmap_atomic_pte(struct page *page, enum km_type type)
Zachary Amsden9a1c13e2007-03-05 00:30:37 -0800271{
Zachary Amsdeneeef9c62007-05-02 19:27:16 +0200272 void *va = kmap_atomic(page, type);
273
Zachary Amsden9a1c13e2007-03-05 00:30:37 -0800274 /*
275 * Internally, the VMI ROM must map virtual addresses to physical
276 * addresses for processing MMU updates. By the time MMU updates
277 * are issued, this information is typically already lost.
278 * Fortunately, the VMI provides a cache of mapping slots for active
279 * page tables.
280 *
281 * We use slot zero for the linear mapping of physical memory, and
282 * in HIGHPTE kernels, slot 1 and 2 for KM_PTE0 and KM_PTE1.
283 *
284 * args: SLOT VA COUNT PFN
285 */
286 BUG_ON(type != KM_PTE0 && type != KM_PTE1);
Zachary Amsdeneeef9c62007-05-02 19:27:16 +0200287 vmi_ops.set_linear_mapping((type - KM_PTE0)+1, va, 1, page_to_pfn(page));
288
289 return va;
Zachary Amsden9a1c13e2007-03-05 00:30:37 -0800290}
Zachary Amsdeneeef9c62007-05-02 19:27:16 +0200291#endif
Zachary Amsden9a1c13e2007-03-05 00:30:37 -0800292
Eduardo Habkostf8639932008-07-30 18:32:27 -0300293static void vmi_allocate_pte(struct mm_struct *mm, unsigned long pfn)
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100294{
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100295 vmi_ops.allocate_page(pfn, VMI_PAGE_L1, 0, 0, 0);
296}
297
Eduardo Habkostf8639932008-07-30 18:32:27 -0300298static void vmi_allocate_pmd(struct mm_struct *mm, unsigned long pfn)
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100299{
300 /*
301 * This call comes in very early, before mem_map is setup.
302 * It is called only for swapper_pg_dir, which already has
303 * data on it.
304 */
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100305 vmi_ops.allocate_page(pfn, VMI_PAGE_L2, 0, 0, 0);
306}
307
Eduardo Habkostf8639932008-07-30 18:32:27 -0300308static void vmi_allocate_pmd_clone(unsigned long pfn, unsigned long clonepfn, unsigned long start, unsigned long count)
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100309{
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100310 vmi_ops.allocate_page(pfn, VMI_PAGE_L2 | VMI_PAGE_CLONE, clonepfn, start, count);
311}
312
Eduardo Habkostf8639932008-07-30 18:32:27 -0300313static void vmi_release_pte(unsigned long pfn)
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100314{
315 vmi_ops.release_page(pfn, VMI_PAGE_L1);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100316}
317
Eduardo Habkostf8639932008-07-30 18:32:27 -0300318static void vmi_release_pmd(unsigned long pfn)
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100319{
320 vmi_ops.release_page(pfn, VMI_PAGE_L2);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100321}
322
323/*
Alok Kataria55a8ba42009-02-06 10:29:35 -0800324 * We use the pgd_free hook for releasing the pgd page:
325 */
326static void vmi_pgd_free(struct mm_struct *mm, pgd_t *pgd)
327{
328 unsigned long pfn = __pa(pgd) >> PAGE_SHIFT;
329
330 vmi_ops.release_page(pfn, VMI_PAGE_L2);
331}
332
333/*
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100334 * Helper macros for MMU update flags. We can defer updates until a flush
335 * or page invalidation only if the update is to the current address space
336 * (otherwise, there is no flush). We must check against init_mm, since
337 * this could be a kernel update, which usually passes init_mm, although
338 * sometimes this check can be skipped if we know the particular function
339 * is only called on user mode PTEs. We could change the kernel to pass
340 * current->active_mm here, but in particular, I was unsure if changing
341 * mm/highmem.c to do this would still be correct on other architectures.
342 */
343#define is_current_as(mm, mustbeuser) ((mm) == current->active_mm || \
344 (!mustbeuser && (mm) == &init_mm))
345#define vmi_flags_addr(mm, addr, level, user) \
346 ((level) | (is_current_as(mm, user) ? \
347 (VMI_PAGE_CURRENT_AS | ((addr) & VMI_PAGE_VA_MASK)) : 0))
348#define vmi_flags_addr_defer(mm, addr, level, user) \
349 ((level) | (is_current_as(mm, user) ? \
350 (VMI_PAGE_DEFER | VMI_PAGE_CURRENT_AS | ((addr) & VMI_PAGE_VA_MASK)) : 0))
351
Jeremy Fitzhardinge3dc494e2007-05-02 19:27:13 +0200352static void vmi_update_pte(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100353{
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100354 vmi_ops.update_pte(ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0));
355}
356
Jeremy Fitzhardinge3dc494e2007-05-02 19:27:13 +0200357static void vmi_update_pte_defer(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100358{
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100359 vmi_ops.update_pte(ptep, vmi_flags_addr_defer(mm, addr, VMI_PAGE_PT, 0));
360}
361
362static void vmi_set_pte(pte_t *ptep, pte_t pte)
363{
364 /* XXX because of set_pmd_pte, this can be called on PT or PD layers */
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100365 vmi_ops.set_pte(pte, ptep, VMI_PAGE_PT);
366}
367
Jeremy Fitzhardinge3dc494e2007-05-02 19:27:13 +0200368static void vmi_set_pte_at(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pte)
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100369{
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100370 vmi_ops.set_pte(pte, ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0));
371}
372
373static void vmi_set_pmd(pmd_t *pmdp, pmd_t pmdval)
374{
375#ifdef CONFIG_X86_PAE
Jeremy Fitzhardingee3328702008-01-30 13:32:58 +0100376 const pte_t pte = { .pte = pmdval.pmd };
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100377#else
378 const pte_t pte = { pmdval.pud.pgd.pgd };
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100379#endif
380 vmi_ops.set_pte(pte, (pte_t *)pmdp, VMI_PAGE_PD);
381}
382
383#ifdef CONFIG_X86_PAE
384
385static void vmi_set_pte_atomic(pte_t *ptep, pte_t pteval)
386{
387 /*
388 * XXX This is called from set_pmd_pte, but at both PT
389 * and PD layers so the VMI_PAGE_PT flag is wrong. But
390 * it is only called for large page mapping changes,
391 * the Xen backend, doesn't support large pages, and the
392 * ESX backend doesn't depend on the flag.
393 */
394 set_64bit((unsigned long long *)ptep,pte_val(pteval));
395 vmi_ops.update_pte(ptep, VMI_PAGE_PT);
396}
397
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100398static void vmi_set_pud(pud_t *pudp, pud_t pudval)
399{
400 /* Um, eww */
Jeremy Fitzhardingee3328702008-01-30 13:32:58 +0100401 const pte_t pte = { .pte = pudval.pgd.pgd };
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100402 vmi_ops.set_pte(pte, (pte_t *)pudp, VMI_PAGE_PDP);
403}
404
405static void vmi_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
406{
Jeremy Fitzhardingee3328702008-01-30 13:32:58 +0100407 const pte_t pte = { .pte = 0 };
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100408 vmi_ops.set_pte(pte, ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0));
409}
410
Adrian Bunk8eb68fa2007-05-02 19:27:09 +0200411static void vmi_pmd_clear(pmd_t *pmd)
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100412{
Jeremy Fitzhardingee3328702008-01-30 13:32:58 +0100413 const pte_t pte = { .pte = 0 };
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100414 vmi_ops.set_pte(pte, (pte_t *)pmd, VMI_PAGE_PD);
415}
416#endif
417
418#ifdef CONFIG_SMP
Zachary Amsdenc6b36e92007-03-05 00:30:43 -0800419static void __devinit
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100420vmi_startup_ipi_hook(int phys_apicid, unsigned long start_eip,
421 unsigned long start_esp)
422{
Zachary Amsdenc6b36e92007-03-05 00:30:43 -0800423 struct vmi_ap_state ap;
424
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100425 /* Default everything to zero. This is fine for most GPRs. */
426 memset(&ap, 0, sizeof(struct vmi_ap_state));
427
428 ap.gdtr_limit = GDT_SIZE - 1;
429 ap.gdtr_base = (unsigned long) get_cpu_gdt_table(phys_apicid);
430
431 ap.idtr_limit = IDT_ENTRIES * 8 - 1;
432 ap.idtr_base = (unsigned long) idt_table;
433
434 ap.ldtr = 0;
435
436 ap.cs = __KERNEL_CS;
437 ap.eip = (unsigned long) start_eip;
438 ap.ss = __KERNEL_DS;
439 ap.esp = (unsigned long) start_esp;
440
441 ap.ds = __USER_DS;
442 ap.es = __USER_DS;
Jeremy Fitzhardinge7c3576d2007-05-02 19:27:16 +0200443 ap.fs = __KERNEL_PERCPU;
Alok Kataria7d5b0052009-08-04 15:34:22 -0700444 ap.gs = __KERNEL_STACK_CANARY;
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100445
446 ap.eflags = 0;
447
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100448#ifdef CONFIG_X86_PAE
449 /* efer should match BSP efer. */
450 if (cpu_has_nx) {
451 unsigned l, h;
452 rdmsr(MSR_EFER, l, h);
453 ap.efer = (unsigned long long) h << 32 | l;
454 }
455#endif
456
457 ap.cr3 = __pa(swapper_pg_dir);
458 /* Protected mode, paging, AM, WP, NE, MP. */
459 ap.cr0 = 0x80050023;
460 ap.cr4 = mmu_cr4_features;
Zachary Amsdenc6b36e92007-03-05 00:30:43 -0800461 vmi_ops.set_initial_ap_state((u32)&ap, phys_apicid);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100462}
463#endif
464
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -0800465static void vmi_start_context_switch(struct task_struct *prev)
Zachary Amsden49f19712007-04-08 16:04:01 -0700466{
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -0800467 paravirt_start_context_switch(prev);
Jeremy Fitzhardinge8965c1c2007-10-16 11:51:29 -0700468 vmi_ops.set_lazy_mode(2);
469}
Zachary Amsden49f19712007-04-08 16:04:01 -0700470
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -0800471static void vmi_end_context_switch(struct task_struct *next)
Jeremy Fitzhardingeb407fc52009-02-17 23:46:21 -0800472{
473 vmi_ops.set_lazy_mode(0);
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -0800474 paravirt_end_context_switch(next);
Jeremy Fitzhardingeb407fc52009-02-17 23:46:21 -0800475}
476
Jeremy Fitzhardinge8965c1c2007-10-16 11:51:29 -0700477static void vmi_enter_lazy_mmu(void)
478{
479 paravirt_enter_lazy_mmu();
480 vmi_ops.set_lazy_mode(1);
481}
Zachary Amsden49f19712007-04-08 16:04:01 -0700482
Jeremy Fitzhardingeb407fc52009-02-17 23:46:21 -0800483static void vmi_leave_lazy_mmu(void)
Jeremy Fitzhardinge8965c1c2007-10-16 11:51:29 -0700484{
Jeremy Fitzhardinge8965c1c2007-10-16 11:51:29 -0700485 vmi_ops.set_lazy_mode(0);
Jeremy Fitzhardingeb407fc52009-02-17 23:46:21 -0800486 paravirt_leave_lazy_mmu();
Zachary Amsden49f19712007-04-08 16:04:01 -0700487}
488
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100489static inline int __init check_vmi_rom(struct vrom_header *rom)
490{
491 struct pci_header *pci;
492 struct pnp_header *pnp;
493 const char *manufacturer = "UNKNOWN";
494 const char *product = "UNKNOWN";
495 const char *license = "unspecified";
496
497 if (rom->rom_signature != 0xaa55)
498 return 0;
499 if (rom->vrom_signature != VMI_SIGNATURE)
500 return 0;
501 if (rom->api_version_maj != VMI_API_REV_MAJOR ||
502 rom->api_version_min+1 < VMI_API_REV_MINOR+1) {
503 printk(KERN_WARNING "VMI: Found mismatched rom version %d.%d\n",
504 rom->api_version_maj,
505 rom->api_version_min);
506 return 0;
507 }
508
509 /*
510 * Relying on the VMI_SIGNATURE field is not 100% safe, so check
511 * the PCI header and device type to make sure this is really a
512 * VMI device.
513 */
514 if (!rom->pci_header_offs) {
515 printk(KERN_WARNING "VMI: ROM does not contain PCI header.\n");
516 return 0;
517 }
518
519 pci = (struct pci_header *)((char *)rom+rom->pci_header_offs);
520 if (pci->vendorID != PCI_VENDOR_ID_VMWARE ||
521 pci->deviceID != PCI_DEVICE_ID_VMWARE_VMI) {
522 /* Allow it to run... anyways, but warn */
523 printk(KERN_WARNING "VMI: ROM from unknown manufacturer\n");
524 }
525
526 if (rom->pnp_header_offs) {
527 pnp = (struct pnp_header *)((char *)rom+rom->pnp_header_offs);
528 if (pnp->manufacturer_offset)
529 manufacturer = (const char *)rom+pnp->manufacturer_offset;
530 if (pnp->product_offset)
531 product = (const char *)rom+pnp->product_offset;
532 }
533
534 if (rom->license_offs)
535 license = (char *)rom+rom->license_offs;
536
537 printk(KERN_INFO "VMI: Found %s %s, API version %d.%d, ROM version %d.%d\n",
538 manufacturer, product,
539 rom->api_version_maj, rom->api_version_min,
540 pci->rom_version_maj, pci->rom_version_min);
541
Andi Kleen302cf932007-03-16 21:07:36 +0100542 /* Don't allow BSD/MIT here for now because we don't want to end up
543 with any binary only shim layers */
544 if (strcmp(license, "GPL") && strcmp(license, "GPL v2")) {
545 printk(KERN_WARNING "VMI: Non GPL license `%s' found for ROM. Not used.\n",
546 license);
547 return 0;
548 }
549
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100550 return 1;
551}
552
553/*
554 * Probe for the VMI option ROM
555 */
556static inline int __init probe_vmi_rom(void)
557{
558 unsigned long base;
559
560 /* VMI ROM is in option ROM area, check signature */
561 for (base = 0xC0000; base < 0xE0000; base += 2048) {
562 struct vrom_header *romstart;
563 romstart = (struct vrom_header *)isa_bus_to_virt(base);
564 if (check_vmi_rom(romstart)) {
565 vmi_rom = romstart;
566 return 1;
567 }
568 }
569 return 0;
570}
571
572/*
573 * VMI setup common to all processors
574 */
575void vmi_bringup(void)
576{
577 /* We must establish the lowmem mapping for MMU ops to work */
Zachary Amsden772205f2007-03-05 00:30:41 -0800578 if (vmi_ops.set_linear_mapping)
Alok Kataria31343d82008-08-08 12:15:57 -0700579 vmi_ops.set_linear_mapping(0, (void *)__PAGE_OFFSET, MAXMEM_PFN, 0);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100580}
581
582/*
Zachary Amsden772205f2007-03-05 00:30:41 -0800583 * Return a pointer to a VMI function or NULL if unimplemented
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100584 */
585static void *vmi_get_function(int vmicall)
586{
587 u64 reloc;
588 const struct vmi_relocation_info *rel = (struct vmi_relocation_info *)&reloc;
589 reloc = call_vrom_long_func(vmi_rom, get_reloc, vmicall);
590 BUG_ON(rel->type == VMI_RELOCATION_JUMP_REL);
591 if (rel->type == VMI_RELOCATION_CALL_REL)
592 return (void *)rel->eip;
593 else
Zachary Amsden772205f2007-03-05 00:30:41 -0800594 return NULL;
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100595}
596
597/*
598 * Helper macro for making the VMI paravirt-ops fill code readable.
Zachary Amsden772205f2007-03-05 00:30:41 -0800599 * For unimplemented operations, fall back to default, unless nop
600 * is returned by the ROM.
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100601 */
602#define para_fill(opname, vmicall) \
603do { \
604 reloc = call_vrom_long_func(vmi_rom, get_reloc, \
605 VMI_CALL_##vmicall); \
Zachary Amsden0492c372007-04-12 19:28:46 -0700606 if (rel->type == VMI_RELOCATION_CALL_REL) \
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700607 opname = (void *)rel->eip; \
Zachary Amsden0492c372007-04-12 19:28:46 -0700608 else if (rel->type == VMI_RELOCATION_NOP) \
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700609 opname = (void *)vmi_nop; \
Zachary Amsden0492c372007-04-12 19:28:46 -0700610 else if (rel->type != VMI_RELOCATION_NONE) \
611 printk(KERN_WARNING "VMI: Unknown relocation " \
612 "type %d for " #vmicall"\n",\
613 rel->type); \
Zachary Amsden772205f2007-03-05 00:30:41 -0800614} while (0)
615
616/*
617 * Helper macro for making the VMI paravirt-ops fill code readable.
618 * For cached operations which do not match the VMI ROM ABI and must
619 * go through a tranlation stub. Ignore NOPs, since it is not clear
620 * a NOP * VMI function corresponds to a NOP paravirt-op when the
621 * functions are not in 1-1 correspondence.
622 */
623#define para_wrap(opname, wrapper, cache, vmicall) \
624do { \
625 reloc = call_vrom_long_func(vmi_rom, get_reloc, \
626 VMI_CALL_##vmicall); \
627 BUG_ON(rel->type == VMI_RELOCATION_JUMP_REL); \
628 if (rel->type == VMI_RELOCATION_CALL_REL) { \
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700629 opname = wrapper; \
Zachary Amsden772205f2007-03-05 00:30:41 -0800630 vmi_ops.cache = (void *)rel->eip; \
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100631 } \
632} while (0)
633
634/*
635 * Activate the VMI interface and switch into paravirtualized mode
636 */
637static inline int __init activate_vmi(void)
638{
639 short kernel_cs;
640 u64 reloc;
641 const struct vmi_relocation_info *rel = (struct vmi_relocation_info *)&reloc;
642
643 if (call_vrom_func(vmi_rom, vmi_init) != 0) {
644 printk(KERN_ERR "VMI ROM failed to initialize!");
645 return 0;
646 }
647 savesegment(cs, kernel_cs);
648
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700649 pv_info.paravirt_enabled = 1;
650 pv_info.kernel_rpl = kernel_cs & SEGMENT_RPL_MASK;
Alok Katariad0153ca2009-09-29 10:25:24 -0700651 pv_info.name = "vmi [deprecated]";
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100652
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700653 pv_init_ops.patch = vmi_patch;
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100654
655 /*
656 * Many of these operations are ABI compatible with VMI.
657 * This means we can fill in the paravirt-ops with direct
658 * pointers into the VMI ROM. If the calling convention for
659 * these operations changes, this code needs to be updated.
660 *
661 * Exceptions
662 * CPUID paravirt-op uses pointers, not the native ISA
663 * halt has no VMI equivalent; all VMI halts are "safe"
664 * no MSR support yet - just trap and emulate. VMI uses the
665 * same ABI as the native ISA, but Linux wants exceptions
666 * from bogus MSR read / write handled
667 * rdpmc is not yet used in Linux
668 */
669
Zachary Amsden772205f2007-03-05 00:30:41 -0800670 /* CPUID is special, so very special it gets wrapped like a present */
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700671 para_wrap(pv_cpu_ops.cpuid, vmi_cpuid, cpuid, CPUID);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100672
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700673 para_fill(pv_cpu_ops.clts, CLTS);
674 para_fill(pv_cpu_ops.get_debugreg, GetDR);
675 para_fill(pv_cpu_ops.set_debugreg, SetDR);
676 para_fill(pv_cpu_ops.read_cr0, GetCR0);
677 para_fill(pv_mmu_ops.read_cr2, GetCR2);
678 para_fill(pv_mmu_ops.read_cr3, GetCR3);
679 para_fill(pv_cpu_ops.read_cr4, GetCR4);
680 para_fill(pv_cpu_ops.write_cr0, SetCR0);
681 para_fill(pv_mmu_ops.write_cr2, SetCR2);
682 para_fill(pv_mmu_ops.write_cr3, SetCR3);
683 para_fill(pv_cpu_ops.write_cr4, SetCR4);
Jeremy Fitzhardinge664c7952009-01-30 23:18:41 -0800684
685 para_fill(pv_irq_ops.save_fl.func, GetInterruptMask);
686 para_fill(pv_irq_ops.restore_fl.func, SetInterruptMask);
687 para_fill(pv_irq_ops.irq_disable.func, DisableInterrupts);
688 para_fill(pv_irq_ops.irq_enable.func, EnableInterrupts);
Zachary Amsden772205f2007-03-05 00:30:41 -0800689
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700690 para_fill(pv_cpu_ops.wbinvd, WBINVD);
691 para_fill(pv_cpu_ops.read_tsc, RDTSC);
Zachary Amsden772205f2007-03-05 00:30:41 -0800692
693 /* The following we emulate with trap and emulate for now */
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100694 /* paravirt_ops.read_msr = vmi_rdmsr */
695 /* paravirt_ops.write_msr = vmi_wrmsr */
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100696 /* paravirt_ops.rdpmc = vmi_rdpmc */
697
Zachary Amsden772205f2007-03-05 00:30:41 -0800698 /* TR interface doesn't pass TR value, wrap */
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700699 para_wrap(pv_cpu_ops.load_tr_desc, vmi_set_tr, set_tr, SetTR);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100700
701 /* LDT is special, too */
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700702 para_wrap(pv_cpu_ops.set_ldt, vmi_set_ldt, _set_ldt, SetLDT);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100703
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700704 para_fill(pv_cpu_ops.load_gdt, SetGDT);
705 para_fill(pv_cpu_ops.load_idt, SetIDT);
706 para_fill(pv_cpu_ops.store_gdt, GetGDT);
707 para_fill(pv_cpu_ops.store_idt, GetIDT);
708 para_fill(pv_cpu_ops.store_tr, GetTR);
709 pv_cpu_ops.load_tls = vmi_load_tls;
Glauber de Oliveira Costa75b8bb32008-01-30 13:31:13 +0100710 para_wrap(pv_cpu_ops.write_ldt_entry, vmi_write_ldt_entry,
711 write_ldt_entry, WriteLDTEntry);
Glauber de Oliveira Costa014b15b2008-01-30 13:31:13 +0100712 para_wrap(pv_cpu_ops.write_gdt_entry, vmi_write_gdt_entry,
713 write_gdt_entry, WriteGDTEntry);
Glauber de Oliveira Costa8d947342008-01-30 13:31:12 +0100714 para_wrap(pv_cpu_ops.write_idt_entry, vmi_write_idt_entry,
715 write_idt_entry, WriteIDTEntry);
H. Peter Anvinfaca6222008-01-30 13:31:02 +0100716 para_wrap(pv_cpu_ops.load_sp0, vmi_load_sp0, set_kernel_stack, UpdateKernelStack);
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700717 para_fill(pv_cpu_ops.set_iopl_mask, SetIOPLMask);
718 para_fill(pv_cpu_ops.io_delay, IODelay);
Jeremy Fitzhardinge8965c1c2007-10-16 11:51:29 -0700719
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -0800720 para_wrap(pv_cpu_ops.start_context_switch, vmi_start_context_switch,
Jeremy Fitzhardinge8965c1c2007-10-16 11:51:29 -0700721 set_lazy_mode, SetLazyMode);
Jeremy Fitzhardinge224101e2009-02-18 11:18:57 -0800722 para_wrap(pv_cpu_ops.end_context_switch, vmi_end_context_switch,
Jeremy Fitzhardinge8965c1c2007-10-16 11:51:29 -0700723 set_lazy_mode, SetLazyMode);
724
725 para_wrap(pv_mmu_ops.lazy_mode.enter, vmi_enter_lazy_mmu,
726 set_lazy_mode, SetLazyMode);
Jeremy Fitzhardingeb407fc52009-02-17 23:46:21 -0800727 para_wrap(pv_mmu_ops.lazy_mode.leave, vmi_leave_lazy_mmu,
Jeremy Fitzhardinge8965c1c2007-10-16 11:51:29 -0700728 set_lazy_mode, SetLazyMode);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100729
Zachary Amsden772205f2007-03-05 00:30:41 -0800730 /* user and kernel flush are just handled with different flags to FlushTLB */
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700731 para_wrap(pv_mmu_ops.flush_tlb_user, vmi_flush_tlb_user, _flush_tlb, FlushTLB);
732 para_wrap(pv_mmu_ops.flush_tlb_kernel, vmi_flush_tlb_kernel, _flush_tlb, FlushTLB);
733 para_fill(pv_mmu_ops.flush_tlb_single, InvalPage);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100734
735 /*
736 * Until a standard flag format can be agreed on, we need to
737 * implement these as wrappers in Linux. Get the VMI ROM
738 * function pointers for the two backend calls.
739 */
740#ifdef CONFIG_X86_PAE
741 vmi_ops.set_pte = vmi_get_function(VMI_CALL_SetPxELong);
742 vmi_ops.update_pte = vmi_get_function(VMI_CALL_UpdatePxELong);
743#else
744 vmi_ops.set_pte = vmi_get_function(VMI_CALL_SetPxE);
745 vmi_ops.update_pte = vmi_get_function(VMI_CALL_UpdatePxE);
746#endif
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100747
Zachary Amsden772205f2007-03-05 00:30:41 -0800748 if (vmi_ops.set_pte) {
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700749 pv_mmu_ops.set_pte = vmi_set_pte;
750 pv_mmu_ops.set_pte_at = vmi_set_pte_at;
751 pv_mmu_ops.set_pmd = vmi_set_pmd;
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100752#ifdef CONFIG_X86_PAE
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700753 pv_mmu_ops.set_pte_atomic = vmi_set_pte_atomic;
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700754 pv_mmu_ops.set_pud = vmi_set_pud;
755 pv_mmu_ops.pte_clear = vmi_pte_clear;
756 pv_mmu_ops.pmd_clear = vmi_pmd_clear;
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100757#endif
Zachary Amsden772205f2007-03-05 00:30:41 -0800758 }
759
760 if (vmi_ops.update_pte) {
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700761 pv_mmu_ops.pte_update = vmi_update_pte;
762 pv_mmu_ops.pte_update_defer = vmi_update_pte_defer;
Zachary Amsden772205f2007-03-05 00:30:41 -0800763 }
764
765 vmi_ops.allocate_page = vmi_get_function(VMI_CALL_AllocatePage);
766 if (vmi_ops.allocate_page) {
Jeremy Fitzhardinge6944a9c2008-03-17 16:37:01 -0700767 pv_mmu_ops.alloc_pte = vmi_allocate_pte;
768 pv_mmu_ops.alloc_pmd = vmi_allocate_pmd;
769 pv_mmu_ops.alloc_pmd_clone = vmi_allocate_pmd_clone;
Zachary Amsden772205f2007-03-05 00:30:41 -0800770 }
771
772 vmi_ops.release_page = vmi_get_function(VMI_CALL_ReleasePage);
773 if (vmi_ops.release_page) {
Jeremy Fitzhardinge6944a9c2008-03-17 16:37:01 -0700774 pv_mmu_ops.release_pte = vmi_release_pte;
775 pv_mmu_ops.release_pmd = vmi_release_pmd;
Alok Kataria55a8ba42009-02-06 10:29:35 -0800776 pv_mmu_ops.pgd_free = vmi_pgd_free;
Zachary Amsden772205f2007-03-05 00:30:41 -0800777 }
Zachary Amsdeneeef9c62007-05-02 19:27:16 +0200778
779 /* Set linear is needed in all cases */
780 vmi_ops.set_linear_mapping = vmi_get_function(VMI_CALL_SetLinearMapping);
781#ifdef CONFIG_HIGHPTE
782 if (vmi_ops.set_linear_mapping)
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700783 pv_mmu_ops.kmap_atomic_pte = vmi_kmap_atomic_pte;
Jeremy Fitzhardingea27fe802007-05-02 19:27:15 +0200784#endif
Zachary Amsden772205f2007-03-05 00:30:41 -0800785
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100786 /*
787 * These MUST always be patched. Don't support indirect jumps
788 * through these operations, as the VMI interface may use either
789 * a jump or a call to get to these operations, depending on
790 * the backend. They are performance critical anyway, so requiring
791 * a patch is not a big problem.
792 */
Jeremy Fitzhardinged75cd222008-06-25 00:19:26 -0400793 pv_cpu_ops.irq_enable_sysexit = (void *)0xfeedbab0;
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700794 pv_cpu_ops.iret = (void *)0xbadbab0;
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100795
796#ifdef CONFIG_SMP
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700797 para_wrap(pv_apic_ops.startup_ipi_hook, vmi_startup_ipi_hook, set_initial_ap_state, SetInitialAPState);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100798#endif
799
800#ifdef CONFIG_X86_LOCAL_APIC
Yinghai Luc1eeb2d2009-02-16 23:02:14 -0800801 para_fill(apic->read, APICRead);
802 para_fill(apic->write, APICWrite);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100803#endif
804
805 /*
Zachary Amsdenbbab4f32007-02-13 13:26:21 +0100806 * Check for VMI timer functionality by probing for a cycle frequency method
807 */
808 reloc = call_vrom_long_func(vmi_rom, get_reloc, VMI_CALL_GetCycleFrequency);
Zachary Amsden772205f2007-03-05 00:30:41 -0800809 if (!disable_vmi_timer && rel->type != VMI_RELOCATION_NONE) {
Zachary Amsdenbbab4f32007-02-13 13:26:21 +0100810 vmi_timer_ops.get_cycle_frequency = (void *)rel->eip;
811 vmi_timer_ops.get_cycle_counter =
812 vmi_get_function(VMI_CALL_GetCycleCounter);
813 vmi_timer_ops.get_wallclock =
814 vmi_get_function(VMI_CALL_GetWallclockTime);
815 vmi_timer_ops.wallclock_updated =
816 vmi_get_function(VMI_CALL_WallclockUpdated);
817 vmi_timer_ops.set_alarm = vmi_get_function(VMI_CALL_SetAlarm);
818 vmi_timer_ops.cancel_alarm =
819 vmi_get_function(VMI_CALL_CancelAlarm);
Thomas Gleixner845b3942009-08-19 15:37:03 +0200820 x86_init.timers.timer_init = vmi_time_init;
Zachary Amsdenbbab4f32007-02-13 13:26:21 +0100821#ifdef CONFIG_X86_LOCAL_APIC
Thomas Gleixner736deca2009-08-19 12:35:53 +0200822 x86_init.timers.setup_percpu_clockev = vmi_time_bsp_init;
823 x86_cpuinit.setup_percpu_clockev = vmi_time_ap_init;
Zachary Amsdenbbab4f32007-02-13 13:26:21 +0100824#endif
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700825 pv_time_ops.sched_clock = vmi_sched_clock;
Thomas Gleixner2d826402009-08-20 17:06:25 +0200826 x86_platform.calibrate_tsc = vmi_tsc_khz;
Feng Tang7bd867d2009-09-10 10:48:56 +0800827 x86_platform.get_wallclock = vmi_get_wallclock;
828 x86_platform.set_wallclock = vmi_set_wallclock;
Zachary Amsden772205f2007-03-05 00:30:41 -0800829
830 /* We have true wallclock functions; disable CMOS clock sync */
831 no_sync_cmos_clock = 1;
832 } else {
833 disable_noidle = 1;
834 disable_vmi_timer = 1;
Zachary Amsdenbbab4f32007-02-13 13:26:21 +0100835 }
Zachary Amsden772205f2007-03-05 00:30:41 -0800836
Jeremy Fitzhardinge93b1eab2007-10-16 11:51:29 -0700837 para_fill(pv_irq_ops.safe_halt, Halt);
Zachary Amsdenbbab4f32007-02-13 13:26:21 +0100838
839 /*
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100840 * Alternative instruction rewriting doesn't happen soon enough
841 * to convert VMI_IRET to a call instead of a jump; so we have
842 * to do this before IRQs get reenabled. Fortunately, it is
843 * idempotent.
844 */
Jeremy Fitzhardinge441d40d2007-05-02 19:27:16 +0200845 apply_paravirt(__parainstructions, __parainstructions_end);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100846
847 vmi_bringup();
848
849 return 1;
850}
851
852#undef para_fill
853
854void __init vmi_init(void)
855{
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100856 if (!vmi_rom)
857 probe_vmi_rom();
858 else
859 check_vmi_rom(vmi_rom);
860
861 /* In case probing for or validating the ROM failed, basil */
862 if (!vmi_rom)
863 return;
864
865 reserve_top_address(-vmi_rom->virtual_top);
866
Zachary Amsden7507ba32007-03-05 00:30:34 -0800867#ifdef CONFIG_X86_IO_APIC
Zachary Amsden772205f2007-03-05 00:30:41 -0800868 /* This is virtual hardware; timer routing is wired correctly */
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100869 no_timer_check = 1;
870#endif
Zachary Amsdenae8d04e2008-12-13 12:36:58 -0800871}
872
Rakib Mullick659d2612009-01-24 01:46:03 +0600873void __init vmi_activate(void)
Zachary Amsdenae8d04e2008-12-13 12:36:58 -0800874{
875 unsigned long flags;
876
877 if (!vmi_rom)
878 return;
879
880 local_irq_save(flags);
881 activate_vmi();
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100882 local_irq_restore(flags & X86_EFLAGS_IF);
883}
884
885static int __init parse_vmi(char *arg)
886{
887 if (!arg)
888 return -EINVAL;
889
Zachary Amsdeneda08b12007-03-05 00:30:38 -0800890 if (!strcmp(arg, "disable_pge")) {
Jeremy Fitzhardinge53756d32008-01-30 13:30:55 +0100891 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_PGE);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100892 disable_pge = 1;
893 } else if (!strcmp(arg, "disable_pse")) {
Jeremy Fitzhardinge53756d32008-01-30 13:30:55 +0100894 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_PSE);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100895 disable_pse = 1;
896 } else if (!strcmp(arg, "disable_sep")) {
Jeremy Fitzhardinge53756d32008-01-30 13:30:55 +0100897 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_SEP);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100898 disable_sep = 1;
899 } else if (!strcmp(arg, "disable_tsc")) {
Jeremy Fitzhardinge53756d32008-01-30 13:30:55 +0100900 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_TSC);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100901 disable_tsc = 1;
902 } else if (!strcmp(arg, "disable_mtrr")) {
Jeremy Fitzhardinge53756d32008-01-30 13:30:55 +0100903 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_MTRR);
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100904 disable_mtrr = 1;
Zachary Amsden772205f2007-03-05 00:30:41 -0800905 } else if (!strcmp(arg, "disable_timer")) {
906 disable_vmi_timer = 1;
907 disable_noidle = 1;
Zachary Amsden7507ba32007-03-05 00:30:34 -0800908 } else if (!strcmp(arg, "disable_noidle"))
909 disable_noidle = 1;
Zachary Amsden7ce0bcf2007-02-13 13:26:21 +0100910 return 0;
911}
912
913early_param("vmi", parse_vmi);