arch/x86/kernel/vmi_32.c

   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 #include <asm/setup.h>
  41
  42 /* Convenient for calling VMI functions indirectly in the ROM */
  43 typedef u32 __attribute__((regparm(1))) (VROMFUNC)(void);
  44 typedef 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
  52 static struct vrom_header *vmi_rom;
  53 static int disable_pge;
  54 static int disable_pse;
  55 static int disable_sep;
  56 static int disable_tsc;
  57 static int disable_mtrr;
  58 static int disable_noidle;
  59 static int disable_vmi_timer;
  60
  61 /* Cached VMI operations */
  62 static struct {
  63         void (*cpuid)(void /* non-c */);
  64         void (*_set_ldt)(u32 selector);
  65         void (*set_tr)(u32 selector);
  66         void (*write_idt_entry)(struct desc_struct *, int, u32, u32);
  67         void (*write_gdt_entry)(struct desc_struct *, int, u32, u32);
  68         void (*write_ldt_entry)(struct desc_struct *, int, u32, u32);
  69         void (*set_kernel_stack)(u32 selector, u32 sp0);
  70         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);
  74         void (*set_linear_mapping)(int, void *, u32, u32);
  75         void (*_flush_tlb)(int);
  76         void (*set_initial_ap_state)(int, int);
  77         void (*halt)(void);
  78         void (*set_lazy_mode)(int mode);
  79 } vmi_ops;
  80
  81 /* Cached VMI operations */
  82 struct vmi_timer_ops vmi_timer_ops;
  83
  84 /*
  85  * VMI patching routines.
  86  */
  87 #define MNEM_CALL 0xe8
  88 #define MNEM_JMP  0xe9
  89 #define MNEM_RET  0xc3
  90
  91 #define IRQ_PATCH_INT_MASK 0
  92 #define IRQ_PATCH_DISABLE  5
  93
  94 static inline void patch_offset(void *insnbuf,
  95                                 unsigned long ip, unsigned long dest)
  96 {
  97         *(unsigned long *)(insnbuf+1) = dest-ip-5;
  98 }
  99
 100 static unsigned patch_internal(int call, unsigned len, void *insnbuf,
 101                                unsigned long ip)
 102 {
 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);
 109                         *(char *)insnbuf = MNEM_CALL;
 110                         patch_offset(insnbuf, ip, (unsigned long)rel->eip);
 111                         return 5;
 112
 113                 case VMI_RELOCATION_JUMP_REL:
 114                         BUG_ON(len < 5);
 115                         *(char *)insnbuf = MNEM_JMP;
 116                         patch_offset(insnbuf, ip, (unsigned long)rel->eip);
 117                         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  */
 137 static unsigned vmi_patch(u8 type, u16 clobbers, void *insns,
 138                           unsigned long ip, unsigned len)
 139 {
 140         switch (type) {
 141                 case PARAVIRT_PATCH(pv_irq_ops.irq_disable):
 142                         return patch_internal(VMI_CALL_DisableInterrupts, len,
 143                                               insns, ip);
 144                 case PARAVIRT_PATCH(pv_irq_ops.irq_enable):
 145                         return patch_internal(VMI_CALL_EnableInterrupts, len,
 146                                               insns, ip);
 147                 case PARAVIRT_PATCH(pv_irq_ops.restore_fl):
 148                         return patch_internal(VMI_CALL_SetInterruptMask, len,
 149                                               insns, ip);
 150                 case PARAVIRT_PATCH(pv_irq_ops.save_fl):
 151                         return patch_internal(VMI_CALL_GetInterruptMask, len,
 152                                               insns, ip);
 153                 case PARAVIRT_PATCH(pv_cpu_ops.iret):
 154                         return patch_internal(VMI_CALL_IRET, len, insns, ip);
 155                 case PARAVIRT_PATCH(pv_cpu_ops.irq_enable_sysexit):
 156                         return patch_internal(VMI_CALL_SYSEXIT, len, insns, ip);
 157                 default:
 158                         break;
 159         }
 160         return len;
 161 }
 162
 163 /* CPUID has non-C semantics, and paravirt-ops API doesn't match hardware ISA */
 164 static void vmi_cpuid(unsigned int *ax, unsigned int *bx,
 165                                unsigned int *cx, unsigned int *dx)
 166 {
 167         int override = 0;
 168         if (*ax == 1)
 169                 override = 1;
 170         asm volatile ("call *%6"
 171                       : "=a" (*ax),
 172                         "=b" (*bx),
 173                         "=c" (*cx),
 174                         "=d" (*dx)
 175                       : "0" (*ax), "2" (*cx), "r" (vmi_ops.cpuid));
 176         if (override) {
 177                 if (disable_pse)
 178                         *dx &= ~X86_FEATURE_PSE;
 179                 if (disable_pge)
 180                         *dx &= ~X86_FEATURE_PGE;
 181                 if (disable_sep)
 182                         *dx &= ~X86_FEATURE_SEP;
 183                 if (disable_tsc)
 184                         *dx &= ~X86_FEATURE_TSC;
 185                 if (disable_mtrr)
 186                         *dx &= ~X86_FEATURE_MTRR;
 187         }
 188 }
 189
 190 static 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)
 193                 write_gdt_entry(gdt, nr, new, 0);
 194 }
 195
 196 static 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
 204 static void vmi_set_ldt(const void *addr, unsigned entries)
 205 {
 206         unsigned cpu = smp_processor_id();
 207         struct desc_struct desc;
 208
 209         pack_descriptor(&desc, (unsigned long)addr,
 210                         entries * sizeof(struct desc_struct) - 1,
 211                         DESC_LDT, 0);
 212         write_gdt_entry(get_cpu_gdt_table(cpu), GDT_ENTRY_LDT, &desc, DESC_LDT);
 213         vmi_ops._set_ldt(entries ? GDT_ENTRY_LDT*sizeof(struct desc_struct) : 0);
 214 }
 215
 216 static void vmi_set_tr(void)
 217 {
 218         vmi_ops.set_tr(GDT_ENTRY_TSS*sizeof(struct desc_struct));
 219 }
 220
 221 static void vmi_write_idt_entry(gate_desc *dt, int entry, const gate_desc *g)
 222 {
 223         u32 *idt_entry = (u32 *)g;
 224         vmi_ops.write_idt_entry(dt, entry, idt_entry[0], idt_entry[1]);
 225 }
 226
 227 static void vmi_write_gdt_entry(struct desc_struct *dt, int entry,
 228                                 const void *desc, int type)
 229 {
 230         u32 *gdt_entry = (u32 *)desc;
 231         vmi_ops.write_gdt_entry(dt, entry, gdt_entry[0], gdt_entry[1]);
 232 }
 233
 234 static void vmi_write_ldt_entry(struct desc_struct *dt, int entry,
 235                                 const void *desc)
 236 {
 237         u32 *ldt_entry = (u32 *)desc;
 238         vmi_ops.write_ldt_entry(dt, entry, ldt_entry[0], ldt_entry[1]);
 239 }
 240
 241 static void vmi_load_sp0(struct tss_struct *tss,
 242                                    struct thread_struct *thread)
 243 {
 244         tss->x86_tss.sp0 = thread->sp0;
 245
 246         /* This can only happen when SEP is enabled, no need to test "SEP"arately */
 247         if (unlikely(tss->x86_tss.ss1 != thread->sysenter_cs)) {
 248                 tss->x86_tss.ss1 = thread->sysenter_cs;
 249                 wrmsr(MSR_IA32_SYSENTER_CS, thread->sysenter_cs, 0);
 250         }
 251         vmi_ops.set_kernel_stack(__KERNEL_DS, tss->x86_tss.sp0);
 252 }
 253
 254 static void vmi_flush_tlb_user(void)
 255 {
 256         vmi_ops._flush_tlb(VMI_FLUSH_TLB);
 257 }
 258
 259 static void vmi_flush_tlb_kernel(void)
 260 {
 261         vmi_ops._flush_tlb(VMI_FLUSH_TLB | VMI_FLUSH_GLOBAL);
 262 }
 263
 264 /* Stub to do nothing at all; used for delays and unimplemented calls */
 265 static void vmi_nop(void)
 266 {
 267 }
 268
 269 #ifdef CONFIG_DEBUG_PAGE_TYPE
 270
 271 #ifdef CONFIG_X86_PAE
 272 #define MAX_BOOT_PTS (2048+4+1)
 273 #else
 274 #define MAX_BOOT_PTS (1024+1)
 275 #endif
 276
 277 /*
 278  * During boot, mem_map is not yet available in paging_init, so stash
 279  * all the boot page allocations here.
 280  */
 281 static struct {
 282         u32 pfn;
 283         int type;
 284 } boot_page_allocations[MAX_BOOT_PTS];
 285 static int num_boot_page_allocations;
 286 static int boot_allocations_applied;
 287
 288 void vmi_apply_boot_page_allocations(void)
 289 {
 290         int i;
 291         BUG_ON(!mem_map);
 292         for (i = 0; i < num_boot_page_allocations; i++) {
 293                 struct page *page = pfn_to_page(boot_page_allocations[i].pfn);
 294                 page->type = boot_page_allocations[i].type;
 295                 page->type = boot_page_allocations[i].type &
 296                                 ~(VMI_PAGE_ZEROED | VMI_PAGE_CLONE);
 297         }
 298         boot_allocations_applied = 1;
 299 }
 300
 301 static void record_page_type(u32 pfn, int type)
 302 {
 303         BUG_ON(num_boot_page_allocations >= MAX_BOOT_PTS);
 304         boot_page_allocations[num_boot_page_allocations].pfn = pfn;
 305         boot_page_allocations[num_boot_page_allocations].type = type;
 306         num_boot_page_allocations++;
 307 }
 308
 309 static void check_zeroed_page(u32 pfn, int type, struct page *page)
 310 {
 311         u32 *ptr;
 312         int i;
 313         int limit = PAGE_SIZE / sizeof(int);
 314
 315         if (page_address(page))
 316                 ptr = (u32 *)page_address(page);
 317         else
 318                 ptr = (u32 *)__va(pfn << PAGE_SHIFT);
 319         /*
 320          * When cloning the root in non-PAE mode, only the userspace
 321          * pdes need to be zeroed.
 322          */
 323         if (type & VMI_PAGE_CLONE)
 324                 limit = KERNEL_PGD_BOUNDARY;
 325         for (i = 0; i < limit; i++)
 326                 BUG_ON(ptr[i]);
 327 }
 328
 329 /*
 330  * We stash the page type into struct page so we can verify the page
 331  * types are used properly.
 332  */
 333 static void vmi_set_page_type(u32 pfn, int type)
 334 {
 335         /* PAE can have multiple roots per page - don't track */
 336         if (PTRS_PER_PMD > 1 && (type & VMI_PAGE_PDP))
 337                 return;
 338
 339         if (boot_allocations_applied) {
 340                 struct page *page = pfn_to_page(pfn);
 341                 if (type != VMI_PAGE_NORMAL)
 342                         BUG_ON(page->type);
 343                 else
 344                         BUG_ON(page->type == VMI_PAGE_NORMAL);
 345                 page->type = type & ~(VMI_PAGE_ZEROED | VMI_PAGE_CLONE);
 346                 if (type & VMI_PAGE_ZEROED)
 347                         check_zeroed_page(pfn, type, page);
 348         } else {
 349                 record_page_type(pfn, type);
 350         }
 351 }
 352
 353 static void vmi_check_page_type(u32 pfn, int type)
 354 {
 355         /* PAE can have multiple roots per page - skip checks */
 356         if (PTRS_PER_PMD > 1 && (type & VMI_PAGE_PDP))
 357                 return;
 358
 359         type &= ~(VMI_PAGE_ZEROED | VMI_PAGE_CLONE);
 360         if (boot_allocations_applied) {
 361                 struct page *page = pfn_to_page(pfn);
 362                 BUG_ON((page->type ^ type) & VMI_PAGE_PAE);
 363                 BUG_ON(type == VMI_PAGE_NORMAL && page->type);
 364                 BUG_ON((type & page->type) == 0);
 365         }
 366 }
 367 #else
 368 #define vmi_set_page_type(p,t) do { } while (0)
 369 #define vmi_check_page_type(p,t) do { } while (0)
 370 #endif
 371
 372 #ifdef CONFIG_HIGHPTE
 373 static void *vmi_kmap_atomic_pte(struct page *page, enum km_type type)
 374 {
 375         void *va = kmap_atomic(page, type);
 376
 377         /*
 378          * Internally, the VMI ROM must map virtual addresses to physical
 379          * addresses for processing MMU updates.  By the time MMU updates
 380          * are issued, this information is typically already lost.
 381          * Fortunately, the VMI provides a cache of mapping slots for active
 382          * page tables.
 383          *
 384          * We use slot zero for the linear mapping of physical memory, and
 385          * in HIGHPTE kernels, slot 1 and 2 for KM_PTE0 and KM_PTE1.
 386          *
 387          *  args:                 SLOT                 VA    COUNT PFN
 388          */
 389         BUG_ON(type != KM_PTE0 && type != KM_PTE1);
 390         vmi_ops.set_linear_mapping((type - KM_PTE0)+1, va, 1, page_to_pfn(page));
 391
 392         return va;
 393 }
 394 #endif
 395
 396 static void vmi_allocate_pte(struct mm_struct *mm, u32 pfn)
 397 {
 398         vmi_set_page_type(pfn, VMI_PAGE_L1);
 399         vmi_ops.allocate_page(pfn, VMI_PAGE_L1, 0, 0, 0);
 400 }
 401
 402 static void vmi_allocate_pmd(struct mm_struct *mm, u32 pfn)
 403 {
 404         /*
 405          * This call comes in very early, before mem_map is setup.
 406          * It is called only for swapper_pg_dir, which already has
 407          * data on it.
 408          */
 409         vmi_set_page_type(pfn, VMI_PAGE_L2);
 410         vmi_ops.allocate_page(pfn, VMI_PAGE_L2, 0, 0, 0);
 411 }
 412
 413 static void vmi_allocate_pmd_clone(u32 pfn, u32 clonepfn, u32 start, u32 count)
 414 {
 415         vmi_set_page_type(pfn, VMI_PAGE_L2 | VMI_PAGE_CLONE);
 416         vmi_check_page_type(clonepfn, VMI_PAGE_L2);
 417         vmi_ops.allocate_page(pfn, VMI_PAGE_L2 | VMI_PAGE_CLONE, clonepfn, start, count);
 418 }
 419
 420 static void vmi_release_pte(u32 pfn)
 421 {
 422         vmi_ops.release_page(pfn, VMI_PAGE_L1);
 423         vmi_set_page_type(pfn, VMI_PAGE_NORMAL);
 424 }
 425
 426 static void vmi_release_pmd(u32 pfn)
 427 {
 428         vmi_ops.release_page(pfn, VMI_PAGE_L2);
 429         vmi_set_page_type(pfn, VMI_PAGE_NORMAL);
 430 }
 431
 432 /*
 433  * Helper macros for MMU update flags.  We can defer updates until a flush
 434  * or page invalidation only if the update is to the current address space
 435  * (otherwise, there is no flush).  We must check against init_mm, since
 436  * this could be a kernel update, which usually passes init_mm, although
 437  * sometimes this check can be skipped if we know the particular function
 438  * is only called on user mode PTEs.  We could change the kernel to pass
 439  * current->active_mm here, but in particular, I was unsure if changing
 440  * mm/highmem.c to do this would still be correct on other architectures.
 441  */
 442 #define is_current_as(mm, mustbeuser) ((mm) == current->active_mm ||    \
 443                                        (!mustbeuser && (mm) == &init_mm))
 444 #define vmi_flags_addr(mm, addr, level, user)                           \
 445         ((level) | (is_current_as(mm, user) ?                           \
 446                 (VMI_PAGE_CURRENT_AS | ((addr) & VMI_PAGE_VA_MASK)) : 0))
 447 #define vmi_flags_addr_defer(mm, addr, level, user)                     \
 448         ((level) | (is_current_as(mm, user) ?                           \
 449                 (VMI_PAGE_DEFER | VMI_PAGE_CURRENT_AS | ((addr) & VMI_PAGE_VA_MASK)) : 0))
 450
 451 static void vmi_update_pte(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
 452 {
 453         vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE);
 454         vmi_ops.update_pte(ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0));
 455 }
 456
 457 static void vmi_update_pte_defer(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
 458 {
 459         vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE);
 460         vmi_ops.update_pte(ptep, vmi_flags_addr_defer(mm, addr, VMI_PAGE_PT, 0));
 461 }
 462
 463 static void vmi_set_pte(pte_t *ptep, pte_t pte)
 464 {
 465         /* XXX because of set_pmd_pte, this can be called on PT or PD layers */
 466         vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE | VMI_PAGE_PD);
 467         vmi_ops.set_pte(pte, ptep, VMI_PAGE_PT);
 468 }
 469
 470 static void vmi_set_pte_at(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pte)
 471 {
 472         vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE);
 473         vmi_ops.set_pte(pte, ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0));
 474 }
 475
 476 static void vmi_set_pmd(pmd_t *pmdp, pmd_t pmdval)
 477 {
 478 #ifdef CONFIG_X86_PAE
 479         const pte_t pte = { .pte = pmdval.pmd };
 480         vmi_check_page_type(__pa(pmdp) >> PAGE_SHIFT, VMI_PAGE_PMD);
 481 #else
 482         const pte_t pte = { pmdval.pud.pgd.pgd };
 483         vmi_check_page_type(__pa(pmdp) >> PAGE_SHIFT, VMI_PAGE_PGD);
 484 #endif
 485         vmi_ops.set_pte(pte, (pte_t *)pmdp, VMI_PAGE_PD);
 486 }
 487
 488 #ifdef CONFIG_X86_PAE
 489
 490 static void vmi_set_pte_atomic(pte_t *ptep, pte_t pteval)
 491 {
 492         /*
 493          * XXX This is called from set_pmd_pte, but at both PT
 494          * and PD layers so the VMI_PAGE_PT flag is wrong.  But
 495          * it is only called for large page mapping changes,
 496          * the Xen backend, doesn't support large pages, and the
 497          * ESX backend doesn't depend on the flag.
 498          */
 499         set_64bit((unsigned long long *)ptep,pte_val(pteval));
 500         vmi_ops.update_pte(ptep, VMI_PAGE_PT);
 501 }
 502
 503 static void vmi_set_pte_present(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pte)
 504 {
 505         vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE);
 506         vmi_ops.set_pte(pte, ptep, vmi_flags_addr_defer(mm, addr, VMI_PAGE_PT, 1));
 507 }
 508
 509 static void vmi_set_pud(pud_t *pudp, pud_t pudval)
 510 {
 511         /* Um, eww */
 512         const pte_t pte = { .pte = pudval.pgd.pgd };
 513         vmi_check_page_type(__pa(pudp) >> PAGE_SHIFT, VMI_PAGE_PGD);
 514         vmi_ops.set_pte(pte, (pte_t *)pudp, VMI_PAGE_PDP);
 515 }
 516
 517 static void vmi_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
 518 {
 519         const pte_t pte = { .pte = 0 };
 520         vmi_check_page_type(__pa(ptep) >> PAGE_SHIFT, VMI_PAGE_PTE);
 521         vmi_ops.set_pte(pte, ptep, vmi_flags_addr(mm, addr, VMI_PAGE_PT, 0));
 522 }
 523
 524 static void vmi_pmd_clear(pmd_t *pmd)
 525 {
 526         const pte_t pte = { .pte = 0 };
 527         vmi_check_page_type(__pa(pmd) >> PAGE_SHIFT, VMI_PAGE_PMD);
 528         vmi_ops.set_pte(pte, (pte_t *)pmd, VMI_PAGE_PD);
 529 }
 530 #endif
 531
 532 #ifdef CONFIG_SMP
 533 static void __devinit
 534 vmi_startup_ipi_hook(int phys_apicid, unsigned long start_eip,
 535                      unsigned long start_esp)
 536 {
 537         struct vmi_ap_state ap;
 538
 539         /* Default everything to zero.  This is fine for most GPRs. */
 540         memset(&ap, 0, sizeof(struct vmi_ap_state));
 541
 542         ap.gdtr_limit = GDT_SIZE - 1;
 543         ap.gdtr_base = (unsigned long) get_cpu_gdt_table(phys_apicid);
 544
 545         ap.idtr_limit = IDT_ENTRIES * 8 - 1;
 546         ap.idtr_base = (unsigned long) idt_table;
 547
 548         ap.ldtr = 0;
 549
 550         ap.cs = __KERNEL_CS;
 551         ap.eip = (unsigned long) start_eip;
 552         ap.ss = __KERNEL_DS;
 553         ap.esp = (unsigned long) start_esp;
 554
 555         ap.ds = __USER_DS;
 556         ap.es = __USER_DS;
 557         ap.fs = __KERNEL_PERCPU;
 558         ap.gs = 0;
 559
 560         ap.eflags = 0;
 561
 562 #ifdef CONFIG_X86_PAE
 563         /* efer should match BSP efer. */
 564         if (cpu_has_nx) {
 565                 unsigned l, h;
 566                 rdmsr(MSR_EFER, l, h);
 567                 ap.efer = (unsigned long long) h << 32 | l;
 568         }
 569 #endif
 570
 571         ap.cr3 = __pa(swapper_pg_dir);
 572         /* Protected mode, paging, AM, WP, NE, MP. */
 573         ap.cr0 = 0x80050023;
 574         ap.cr4 = mmu_cr4_features;
 575         vmi_ops.set_initial_ap_state((u32)&ap, phys_apicid);
 576 }
 577 #endif
 578
 579 static void vmi_enter_lazy_cpu(void)
 580 {
 581         paravirt_enter_lazy_cpu();
 582         vmi_ops.set_lazy_mode(2);
 583 }
 584
 585 static void vmi_enter_lazy_mmu(void)
 586 {
 587         paravirt_enter_lazy_mmu();
 588         vmi_ops.set_lazy_mode(1);
 589 }
 590
 591 static void vmi_leave_lazy(void)
 592 {
 593         paravirt_leave_lazy(paravirt_get_lazy_mode());
 594         vmi_ops.set_lazy_mode(0);
 595 }
 596
 597 static inline int __init check_vmi_rom(struct vrom_header *rom)
 598 {
 599         struct pci_header *pci;
 600         struct pnp_header *pnp;
 601         const char *manufacturer = "UNKNOWN";
 602         const char *product = "UNKNOWN";
 603         const char *license = "unspecified";
 604
 605         if (rom->rom_signature != 0xaa55)
 606                 return 0;
 607         if (rom->vrom_signature != VMI_SIGNATURE)
 608                 return 0;
 609         if (rom->api_version_maj != VMI_API_REV_MAJOR ||
 610             rom->api_version_min+1 < VMI_API_REV_MINOR+1) {
 611                 printk(KERN_WARNING "VMI: Found mismatched rom version %d.%d\n",
 612                                 rom->api_version_maj,
 613                                 rom->api_version_min);
 614                 return 0;
 615         }
 616
 617         /*
 618          * Relying on the VMI_SIGNATURE field is not 100% safe, so check
 619          * the PCI header and device type to make sure this is really a
 620          * VMI device.
 621          */
 622         if (!rom->pci_header_offs) {
 623                 printk(KERN_WARNING "VMI: ROM does not contain PCI header.\n");
 624                 return 0;
 625         }
 626
 627         pci = (struct pci_header *)((char *)rom+rom->pci_header_offs);
 628         if (pci->vendorID != PCI_VENDOR_ID_VMWARE ||
 629             pci->deviceID != PCI_DEVICE_ID_VMWARE_VMI) {
 630                 /* Allow it to run... anyways, but warn */
 631                 printk(KERN_WARNING "VMI: ROM from unknown manufacturer\n");
 632         }
 633
 634         if (rom->pnp_header_offs) {
 635                 pnp = (struct pnp_header *)((char *)rom+rom->pnp_header_offs);
 636                 if (pnp->manufacturer_offset)
 637                         manufacturer = (const char *)rom+pnp->manufacturer_offset;
 638                 if (pnp->product_offset)
 639                         product = (const char *)rom+pnp->product_offset;
 640         }
 641
 642         if (rom->license_offs)
 643                 license = (char *)rom+rom->license_offs;
 644
 645         printk(KERN_INFO "VMI: Found %s %s, API version %d.%d, ROM version %d.%d\n",
 646                 manufacturer, product,
 647                 rom->api_version_maj, rom->api_version_min,
 648                 pci->rom_version_maj, pci->rom_version_min);
 649
 650         /* Don't allow BSD/MIT here for now because we don't want to end up
 651            with any binary only shim layers */
 652         if (strcmp(license, "GPL") && strcmp(license, "GPL v2")) {
 653                 printk(KERN_WARNING "VMI: Non GPL license `%s' found for ROM. Not used.\n",
 654                         license);
 655                 return 0;
 656         }
 657
 658         return 1;
 659 }
 660
 661 /*
 662  * Probe for the VMI option ROM
 663  */
 664 static inline int __init probe_vmi_rom(void)
 665 {
 666         unsigned long base;
 667
 668         /* VMI ROM is in option ROM area, check signature */
 669         for (base = 0xC0000; base < 0xE0000; base += 2048) {
 670                 struct vrom_header *romstart;
 671                 romstart = (struct vrom_header *)isa_bus_to_virt(base);
 672                 if (check_vmi_rom(romstart)) {
 673                         vmi_rom = romstart;
 674                         return 1;
 675                 }
 676         }
 677         return 0;
 678 }
 679
 680 /*
 681  * VMI setup common to all processors
 682  */
 683 void vmi_bringup(void)
 684 {
 685         /* We must establish the lowmem mapping for MMU ops to work */
 686         if (vmi_ops.set_linear_mapping)
 687                 vmi_ops.set_linear_mapping(0, (void *)__PAGE_OFFSET, MAXMEM_PFN, 0);
 688 }
 689
 690 /*
 691  * Return a pointer to a VMI function or NULL if unimplemented
 692  */
 693 static void *vmi_get_function(int vmicall)
 694 {
 695         u64 reloc;
 696         const struct vmi_relocation_info *rel = (struct vmi_relocation_info *)&reloc;
 697         reloc = call_vrom_long_func(vmi_rom, get_reloc, vmicall);
 698         BUG_ON(rel->type == VMI_RELOCATION_JUMP_REL);
 699         if (rel->type == VMI_RELOCATION_CALL_REL)
 700                 return (void *)rel->eip;
 701         else
 702                 return NULL;
 703 }
 704
 705 /*
 706  * Helper macro for making the VMI paravirt-ops fill code readable.
 707  * For unimplemented operations, fall back to default, unless nop
 708  * is returned by the ROM.
 709  */
 710 #define para_fill(opname, vmicall)                              \
 711 do {                                                            \
 712         reloc = call_vrom_long_func(vmi_rom, get_reloc,         \
 713                                     VMI_CALL_##vmicall);        \
 714         if (rel->type == VMI_RELOCATION_CALL_REL)               \
 715                 opname = (void *)rel->eip;                      \
 716         else if (rel->type == VMI_RELOCATION_NOP)               \
 717                 opname = (void *)vmi_nop;                       \
 718         else if (rel->type != VMI_RELOCATION_NONE)              \
 719                 printk(KERN_WARNING "VMI: Unknown relocation "  \
 720                                     "type %d for " #vmicall"\n",\
 721                                         rel->type);             \
 722 } while (0)
 723
 724 /*
 725  * Helper macro for making the VMI paravirt-ops fill code readable.
 726  * For cached operations which do not match the VMI ROM ABI and must
 727  * go through a tranlation stub.  Ignore NOPs, since it is not clear
 728  * a NOP * VMI function corresponds to a NOP paravirt-op when the
 729  * functions are not in 1-1 correspondence.
 730  */
 731 #define para_wrap(opname, wrapper, cache, vmicall)              \
 732 do {                                                            \
 733         reloc = call_vrom_long_func(vmi_rom, get_reloc,         \
 734                                     VMI_CALL_##vmicall);        \
 735         BUG_ON(rel->type == VMI_RELOCATION_JUMP_REL);           \
 736         if (rel->type == VMI_RELOCATION_CALL_REL) {             \
 737                 opname = wrapper;                               \
 738                 vmi_ops.cache = (void *)rel->eip;               \
 739         }                                                       \
 740 } while (0)
 741
 742 /*
 743  * Activate the VMI interface and switch into paravirtualized mode
 744  */
 745 static inline int __init activate_vmi(void)
 746 {
 747         short kernel_cs;
 748         u64 reloc;
 749         const struct vmi_relocation_info *rel = (struct vmi_relocation_info *)&reloc;
 750
 751         if (call_vrom_func(vmi_rom, vmi_init) != 0) {
 752                 printk(KERN_ERR "VMI ROM failed to initialize!");
 753                 return 0;
 754         }
 755         savesegment(cs, kernel_cs);
 756
 757         pv_info.paravirt_enabled = 1;
 758         pv_info.kernel_rpl = kernel_cs & SEGMENT_RPL_MASK;
 759         pv_info.name = "vmi";
 760
 761         pv_init_ops.patch = vmi_patch;
 762
 763         /*
 764          * Many of these operations are ABI compatible with VMI.
 765          * This means we can fill in the paravirt-ops with direct
 766          * pointers into the VMI ROM.  If the calling convention for
 767          * these operations changes, this code needs to be updated.
 768          *
 769          * Exceptions
 770          *  CPUID paravirt-op uses pointers, not the native ISA
 771          *  halt has no VMI equivalent; all VMI halts are "safe"
 772          *  no MSR support yet - just trap and emulate.  VMI uses the
 773          *    same ABI as the native ISA, but Linux wants exceptions
 774          *    from bogus MSR read / write handled
 775          *  rdpmc is not yet used in Linux
 776          */
 777
 778         /* CPUID is special, so very special it gets wrapped like a present */
 779         para_wrap(pv_cpu_ops.cpuid, vmi_cpuid, cpuid, CPUID);
 780
 781         para_fill(pv_cpu_ops.clts, CLTS);
 782         para_fill(pv_cpu_ops.get_debugreg, GetDR);
 783         para_fill(pv_cpu_ops.set_debugreg, SetDR);
 784         para_fill(pv_cpu_ops.read_cr0, GetCR0);
 785         para_fill(pv_mmu_ops.read_cr2, GetCR2);
 786         para_fill(pv_mmu_ops.read_cr3, GetCR3);
 787         para_fill(pv_cpu_ops.read_cr4, GetCR4);
 788         para_fill(pv_cpu_ops.write_cr0, SetCR0);
 789         para_fill(pv_mmu_ops.write_cr2, SetCR2);
 790         para_fill(pv_mmu_ops.write_cr3, SetCR3);
 791         para_fill(pv_cpu_ops.write_cr4, SetCR4);
 792         para_fill(pv_irq_ops.save_fl, GetInterruptMask);
 793         para_fill(pv_irq_ops.restore_fl, SetInterruptMask);
 794         para_fill(pv_irq_ops.irq_disable, DisableInterrupts);
 795         para_fill(pv_irq_ops.irq_enable, EnableInterrupts);
 796
 797         para_fill(pv_cpu_ops.wbinvd, WBINVD);
 798         para_fill(pv_cpu_ops.read_tsc, RDTSC);
 799
 800         /* The following we emulate with trap and emulate for now */
 801         /* paravirt_ops.read_msr = vmi_rdmsr */
 802         /* paravirt_ops.write_msr = vmi_wrmsr */
 803         /* paravirt_ops.rdpmc = vmi_rdpmc */
 804
 805         /* TR interface doesn't pass TR value, wrap */
 806         para_wrap(pv_cpu_ops.load_tr_desc, vmi_set_tr, set_tr, SetTR);
 807
 808         /* LDT is special, too */
 809         para_wrap(pv_cpu_ops.set_ldt, vmi_set_ldt, _set_ldt, SetLDT);
 810
 811         para_fill(pv_cpu_ops.load_gdt, SetGDT);
 812         para_fill(pv_cpu_ops.load_idt, SetIDT);
 813         para_fill(pv_cpu_ops.store_gdt, GetGDT);
 814         para_fill(pv_cpu_ops.store_idt, GetIDT);
 815         para_fill(pv_cpu_ops.store_tr, GetTR);
 816         pv_cpu_ops.load_tls = vmi_load_tls;
 817         para_wrap(pv_cpu_ops.write_ldt_entry, vmi_write_ldt_entry,
 818                   write_ldt_entry, WriteLDTEntry);
 819         para_wrap(pv_cpu_ops.write_gdt_entry, vmi_write_gdt_entry,
 820                   write_gdt_entry, WriteGDTEntry);
 821         para_wrap(pv_cpu_ops.write_idt_entry, vmi_write_idt_entry,
 822                   write_idt_entry, WriteIDTEntry);
 823         para_wrap(pv_cpu_ops.load_sp0, vmi_load_sp0, set_kernel_stack, UpdateKernelStack);
 824         para_fill(pv_cpu_ops.set_iopl_mask, SetIOPLMask);
 825         para_fill(pv_cpu_ops.io_delay, IODelay);
 826
 827         para_wrap(pv_cpu_ops.lazy_mode.enter, vmi_enter_lazy_cpu,
 828                   set_lazy_mode, SetLazyMode);
 829         para_wrap(pv_cpu_ops.lazy_mode.leave, vmi_leave_lazy,
 830                   set_lazy_mode, SetLazyMode);
 831
 832         para_wrap(pv_mmu_ops.lazy_mode.enter, vmi_enter_lazy_mmu,
 833                   set_lazy_mode, SetLazyMode);
 834         para_wrap(pv_mmu_ops.lazy_mode.leave, vmi_leave_lazy,
 835                   set_lazy_mode, SetLazyMode);
 836
 837         /* user and kernel flush are just handled with different flags to FlushTLB */
 838         para_wrap(pv_mmu_ops.flush_tlb_user, vmi_flush_tlb_user, _flush_tlb, FlushTLB);
 839         para_wrap(pv_mmu_ops.flush_tlb_kernel, vmi_flush_tlb_kernel, _flush_tlb, FlushTLB);
 840         para_fill(pv_mmu_ops.flush_tlb_single, InvalPage);
 841
 842         /*
 843          * Until a standard flag format can be agreed on, we need to
 844          * implement these as wrappers in Linux.  Get the VMI ROM
 845          * function pointers for the two backend calls.
 846          */
 847 #ifdef CONFIG_X86_PAE
 848         vmi_ops.set_pte = vmi_get_function(VMI_CALL_SetPxELong);
 849         vmi_ops.update_pte = vmi_get_function(VMI_CALL_UpdatePxELong);
 850 #else
 851         vmi_ops.set_pte = vmi_get_function(VMI_CALL_SetPxE);
 852         vmi_ops.update_pte = vmi_get_function(VMI_CALL_UpdatePxE);
 853 #endif
 854
 855         if (vmi_ops.set_pte) {
 856                 pv_mmu_ops.set_pte = vmi_set_pte;
 857                 pv_mmu_ops.set_pte_at = vmi_set_pte_at;
 858                 pv_mmu_ops.set_pmd = vmi_set_pmd;
 859 #ifdef CONFIG_X86_PAE
 860                 pv_mmu_ops.set_pte_atomic = vmi_set_pte_atomic;
 861                 pv_mmu_ops.set_pte_present = vmi_set_pte_present;
 862                 pv_mmu_ops.set_pud = vmi_set_pud;
 863                 pv_mmu_ops.pte_clear = vmi_pte_clear;
 864                 pv_mmu_ops.pmd_clear = vmi_pmd_clear;
 865 #endif
 866         }
 867
 868         if (vmi_ops.update_pte) {
 869                 pv_mmu_ops.pte_update = vmi_update_pte;
 870                 pv_mmu_ops.pte_update_defer = vmi_update_pte_defer;
 871         }
 872
 873         vmi_ops.allocate_page = vmi_get_function(VMI_CALL_AllocatePage);
 874         if (vmi_ops.allocate_page) {
 875                 pv_mmu_ops.alloc_pte = vmi_allocate_pte;
 876                 pv_mmu_ops.alloc_pmd = vmi_allocate_pmd;
 877                 pv_mmu_ops.alloc_pmd_clone = vmi_allocate_pmd_clone;
 878         }
 879
 880         vmi_ops.release_page = vmi_get_function(VMI_CALL_ReleasePage);
 881         if (vmi_ops.release_page) {
 882                 pv_mmu_ops.release_pte = vmi_release_pte;
 883                 pv_mmu_ops.release_pmd = vmi_release_pmd;
 884         }
 885
 886         /* Set linear is needed in all cases */
 887         vmi_ops.set_linear_mapping = vmi_get_function(VMI_CALL_SetLinearMapping);
 888 #ifdef CONFIG_HIGHPTE
 889         if (vmi_ops.set_linear_mapping)
 890                 pv_mmu_ops.kmap_atomic_pte = vmi_kmap_atomic_pte;
 891 #endif
 892
 893         /*
 894          * These MUST always be patched.  Don't support indirect jumps
 895          * through these operations, as the VMI interface may use either
 896          * a jump or a call to get to these operations, depending on
 897          * the backend.  They are performance critical anyway, so requiring
 898          * a patch is not a big problem.
 899          */
 900         pv_cpu_ops.irq_enable_sysexit = (void *)0xfeedbab0;
 901         pv_cpu_ops.iret = (void *)0xbadbab0;
 902
 903 #ifdef CONFIG_SMP
 904         para_wrap(pv_apic_ops.startup_ipi_hook, vmi_startup_ipi_hook, set_initial_ap_state, SetInitialAPState);
 905 #endif
 906
 907 #ifdef CONFIG_X86_LOCAL_APIC
 908         para_fill(pv_apic_ops.apic_read, APICRead);
 909         para_fill(pv_apic_ops.apic_write, APICWrite);
 910 #endif
 911
 912         /*
 913          * Check for VMI timer functionality by probing for a cycle frequency method
 914          */
 915         reloc = call_vrom_long_func(vmi_rom, get_reloc, VMI_CALL_GetCycleFrequency);
 916         if (!disable_vmi_timer && rel->type != VMI_RELOCATION_NONE) {
 917                 vmi_timer_ops.get_cycle_frequency = (void *)rel->eip;
 918                 vmi_timer_ops.get_cycle_counter =
 919                         vmi_get_function(VMI_CALL_GetCycleCounter);
 920                 vmi_timer_ops.get_wallclock =
 921                         vmi_get_function(VMI_CALL_GetWallclockTime);
 922                 vmi_timer_ops.wallclock_updated =
 923                         vmi_get_function(VMI_CALL_WallclockUpdated);
 924                 vmi_timer_ops.set_alarm = vmi_get_function(VMI_CALL_SetAlarm);
 925                 vmi_timer_ops.cancel_alarm =
 926                          vmi_get_function(VMI_CALL_CancelAlarm);
 927                 pv_time_ops.time_init = vmi_time_init;
 928                 pv_time_ops.get_wallclock = vmi_get_wallclock;
 929                 pv_time_ops.set_wallclock = vmi_set_wallclock;
 930 #ifdef CONFIG_X86_LOCAL_APIC
 931                 pv_apic_ops.setup_boot_clock = vmi_time_bsp_init;
 932                 pv_apic_ops.setup_secondary_clock = vmi_time_ap_init;
 933 #endif
 934                 pv_time_ops.sched_clock = vmi_sched_clock;
 935                 pv_time_ops.get_tsc_khz = vmi_tsc_khz;
 936
 937                 /* We have true wallclock functions; disable CMOS clock sync */
 938                 no_sync_cmos_clock = 1;
 939         } else {
 940                 disable_noidle = 1;
 941                 disable_vmi_timer = 1;
 942         }
 943
 944         para_fill(pv_irq_ops.safe_halt, Halt);
 945
 946         /*
 947          * Alternative instruction rewriting doesn't happen soon enough
 948          * to convert VMI_IRET to a call instead of a jump; so we have
 949          * to do this before IRQs get reenabled.  Fortunately, it is
 950          * idempotent.
 951          */
 952         apply_paravirt(__parainstructions, __parainstructions_end);
 953
 954         vmi_bringup();
 955
 956         return 1;
 957 }
 958
 959 #undef para_fill
 960
 961 void __init vmi_init(void)
 962 {
 963         unsigned long flags;
 964
 965         if (!vmi_rom)
 966                 probe_vmi_rom();
 967         else
 968                 check_vmi_rom(vmi_rom);
 969
 970         /* In case probing for or validating the ROM failed, basil */
 971         if (!vmi_rom)
 972                 return;
 973
 974         reserve_top_address(-vmi_rom->virtual_top);
 975
 976         local_irq_save(flags);
 977         activate_vmi();
 978
 979 #ifdef CONFIG_X86_IO_APIC
 980         /* This is virtual hardware; timer routing is wired correctly */
 981         no_timer_check = 1;
 982 #endif
 983         local_irq_restore(flags & X86_EFLAGS_IF);
 984 }
 985
 986 static int __init parse_vmi(char *arg)
 987 {
 988         if (!arg)
 989                 return -EINVAL;
 990
 991         if (!strcmp(arg, "disable_pge")) {
 992                 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_PGE);
 993                 disable_pge = 1;
 994         } else if (!strcmp(arg, "disable_pse")) {
 995                 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_PSE);
 996                 disable_pse = 1;
 997         } else if (!strcmp(arg, "disable_sep")) {
 998                 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_SEP);
 999                 disable_sep = 1;
1000         } else if (!strcmp(arg, "disable_tsc")) {
1001                 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_TSC);
1002                 disable_tsc = 1;
1003         } else if (!strcmp(arg, "disable_mtrr")) {
1004                 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_MTRR);
1005                 disable_mtrr = 1;
1006         } else if (!strcmp(arg, "disable_timer")) {
1007                 disable_vmi_timer = 1;
1008                 disable_noidle = 1;
1009         } else if (!strcmp(arg, "disable_noidle"))
1010                 disable_noidle = 1;
1011         return 0;
1012 }
1013
1014 early_param("vmi", parse_vmi);