2 * VMI specific paravirt-ops implementation
4 * Copyright (C) 2005, VMware, Inc.
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.
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
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.
21 * Send feedback to zach@vmware.com
25 #include <linux/module.h>
26 #include <linux/cpu.h>
27 #include <linux/bootmem.h>
29 #include <linux/highmem.h>
30 #include <linux/sched.h>
33 #include <asm/fixmap.h>
34 #include <asm/apicdef.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>
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);
46 #define call_vrom_func(rom,func) \
47 (((VROMFUNC *)(rom->func))())
49 #define call_vrom_long_func(rom,func,arg) \
50 (((VROMLONGFUNC *)(rom->func)) (arg))
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;
61 /* Cached VMI operations */
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);
78 void (*set_lazy_mode)(int mode);
81 /* Cached VMI operations */
82 struct vmi_timer_ops vmi_timer_ops;
85 * VMI patching routines.
87 #define MNEM_CALL 0xe8
91 #define IRQ_PATCH_INT_MASK 0
92 #define IRQ_PATCH_DISABLE 5
94 static inline void patch_offset(void *insnbuf,
95 unsigned long ip, unsigned long dest)
97 *(unsigned long *)(insnbuf+1) = dest-ip-5;
100 static unsigned patch_internal(int call, unsigned len, void *insnbuf,
104 struct vmi_relocation_info *const rel = (struct vmi_relocation_info *)&reloc;
105 reloc = call_vrom_long_func(vmi_rom, get_reloc, call);
107 case VMI_RELOCATION_CALL_REL:
109 *(char *)insnbuf = MNEM_CALL;
110 patch_offset(insnbuf, ip, (unsigned long)rel->eip);
113 case VMI_RELOCATION_JUMP_REL:
115 *(char *)insnbuf = MNEM_JMP;
116 patch_offset(insnbuf, ip, (unsigned long)rel->eip);
119 case VMI_RELOCATION_NOP:
120 /* obliterate the whole thing */
123 case VMI_RELOCATION_NONE:
124 /* leave native code in place */
134 * Apply patch if appropriate, return length of new instruction
135 * sequence. The callee does nop padding for us.
137 static unsigned vmi_patch(u8 type, u16 clobbers, void *insns,
138 unsigned long ip, unsigned len)
141 case PARAVIRT_PATCH(pv_irq_ops.irq_disable):
142 return patch_internal(VMI_CALL_DisableInterrupts, len,
144 case PARAVIRT_PATCH(pv_irq_ops.irq_enable):
145 return patch_internal(VMI_CALL_EnableInterrupts, len,
147 case PARAVIRT_PATCH(pv_irq_ops.restore_fl):
148 return patch_internal(VMI_CALL_SetInterruptMask, len,
150 case PARAVIRT_PATCH(pv_irq_ops.save_fl):
151 return patch_internal(VMI_CALL_GetInterruptMask, len,
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);
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)
170 asm volatile ("call *%6"
175 : "0" (*ax), "2" (*cx), "r" (vmi_ops.cpuid));
178 *dx &= ~X86_FEATURE_PSE;
180 *dx &= ~X86_FEATURE_PGE;
182 *dx &= ~X86_FEATURE_SEP;
184 *dx &= ~X86_FEATURE_TSC;
186 *dx &= ~X86_FEATURE_MTRR;
190 static inline void vmi_maybe_load_tls(struct desc_struct *gdt, int nr, struct desc_struct *new)
192 if (gdt[nr].a != new->a || gdt[nr].b != new->b)
193 write_gdt_entry(gdt, nr, new, 0);
196 static void vmi_load_tls(struct thread_struct *t, unsigned int cpu)
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]);
204 static void vmi_set_ldt(const void *addr, unsigned entries)
206 unsigned cpu = smp_processor_id();
207 struct desc_struct desc;
209 pack_descriptor(&desc, (unsigned long)addr,
210 entries * sizeof(struct desc_struct) - 1,
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);
216 static void vmi_set_tr(void)
218 vmi_ops.set_tr(GDT_ENTRY_TSS*sizeof(struct desc_struct));
221 static void vmi_write_idt_entry(gate_desc *dt, int entry, const gate_desc *g)
223 u32 *idt_entry = (u32 *)g;
224 vmi_ops.write_idt_entry(dt, entry, idt_entry[0], idt_entry[1]);
227 static void vmi_write_gdt_entry(struct desc_struct *dt, int entry,
228 const void *desc, int type)
230 u32 *gdt_entry = (u32 *)desc;
231 vmi_ops.write_gdt_entry(dt, entry, gdt_entry[0], gdt_entry[1]);
234 static void vmi_write_ldt_entry(struct desc_struct *dt, int entry,
237 u32 *ldt_entry = (u32 *)desc;
238 vmi_ops.write_idt_entry(dt, entry, ldt_entry[0], ldt_entry[1]);
241 static void vmi_load_sp0(struct tss_struct *tss,
242 struct thread_struct *thread)
244 tss->x86_tss.sp0 = thread->sp0;
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);
251 vmi_ops.set_kernel_stack(__KERNEL_DS, tss->x86_tss.sp0);
254 static void vmi_flush_tlb_user(void)
256 vmi_ops._flush_tlb(VMI_FLUSH_TLB);
259 static void vmi_flush_tlb_kernel(void)
261 vmi_ops._flush_tlb(VMI_FLUSH_TLB | VMI_FLUSH_GLOBAL);
264 /* Stub to do nothing at all; used for delays and unimplemented calls */
265 static void vmi_nop(void)
269 #ifdef CONFIG_DEBUG_PAGE_TYPE
271 #ifdef CONFIG_X86_PAE
272 #define MAX_BOOT_PTS (2048+4+1)
274 #define MAX_BOOT_PTS (1024+1)
278 * During boot, mem_map is not yet available in paging_init, so stash
279 * all the boot page allocations here.
284 } boot_page_allocations[MAX_BOOT_PTS];
285 static int num_boot_page_allocations;
286 static int boot_allocations_applied;
288 void vmi_apply_boot_page_allocations(void)
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);
298 boot_allocations_applied = 1;
301 static void record_page_type(u32 pfn, int type)
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++;
309 static void check_zeroed_page(u32 pfn, int type, struct page *page)
313 int limit = PAGE_SIZE / sizeof(int);
315 if (page_address(page))
316 ptr = (u32 *)page_address(page);
318 ptr = (u32 *)__va(pfn << PAGE_SHIFT);
320 * When cloning the root in non-PAE mode, only the userspace
321 * pdes need to be zeroed.
323 if (type & VMI_PAGE_CLONE)
324 limit = KERNEL_PGD_BOUNDARY;
325 for (i = 0; i < limit; i++)
330 * We stash the page type into struct page so we can verify the page
331 * types are used properly.
333 static void vmi_set_page_type(u32 pfn, int type)
335 /* PAE can have multiple roots per page - don't track */
336 if (PTRS_PER_PMD > 1 && (type & VMI_PAGE_PDP))
339 if (boot_allocations_applied) {
340 struct page *page = pfn_to_page(pfn);
341 if (type != VMI_PAGE_NORMAL)
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);
349 record_page_type(pfn, type);
353 static void vmi_check_page_type(u32 pfn, int type)
355 /* PAE can have multiple roots per page - skip checks */
356 if (PTRS_PER_PMD > 1 && (type & VMI_PAGE_PDP))
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);
368 #define vmi_set_page_type(p,t) do { } while (0)
369 #define vmi_check_page_type(p,t) do { } while (0)
372 #ifdef CONFIG_HIGHPTE
373 static void *vmi_kmap_atomic_pte(struct page *page, enum km_type type)
375 void *va = kmap_atomic(page, type);
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
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.
387 * args: SLOT VA COUNT PFN
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));
396 static void vmi_allocate_pte(struct mm_struct *mm, u32 pfn)
398 vmi_set_page_type(pfn, VMI_PAGE_L1);
399 vmi_ops.allocate_page(pfn, VMI_PAGE_L1, 0, 0, 0);
402 static void vmi_allocate_pmd(struct mm_struct *mm, u32 pfn)
405 * This call comes in very early, before mem_map is setup.
406 * It is called only for swapper_pg_dir, which already has
409 vmi_set_page_type(pfn, VMI_PAGE_L2);
410 vmi_ops.allocate_page(pfn, VMI_PAGE_L2, 0, 0, 0);
413 static void vmi_allocate_pmd_clone(u32 pfn, u32 clonepfn, u32 start, u32 count)
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);
420 static void vmi_release_pte(u32 pfn)
422 vmi_ops.release_page(pfn, VMI_PAGE_L1);
423 vmi_set_page_type(pfn, VMI_PAGE_NORMAL);
426 static void vmi_release_pmd(u32 pfn)
428 vmi_ops.release_page(pfn, VMI_PAGE_L2);
429 vmi_set_page_type(pfn, VMI_PAGE_NORMAL);
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.
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))
451 static void vmi_update_pte(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
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));
457 static void vmi_update_pte_defer(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
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));
463 static void vmi_set_pte(pte_t *ptep, pte_t pte)
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);
470 static void vmi_set_pte_at(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pte)
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));
476 static void vmi_set_pmd(pmd_t *pmdp, pmd_t pmdval)
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);
482 const pte_t pte = { pmdval.pud.pgd.pgd };
483 vmi_check_page_type(__pa(pmdp) >> PAGE_SHIFT, VMI_PAGE_PGD);
485 vmi_ops.set_pte(pte, (pte_t *)pmdp, VMI_PAGE_PD);
488 #ifdef CONFIG_X86_PAE
490 static void vmi_set_pte_atomic(pte_t *ptep, pte_t pteval)
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.
499 set_64bit((unsigned long long *)ptep,pte_val(pteval));
500 vmi_ops.update_pte(ptep, VMI_PAGE_PT);
503 static void vmi_set_pte_present(struct mm_struct *mm, unsigned long addr, pte_t *ptep, pte_t pte)
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));
509 static void vmi_set_pud(pud_t *pudp, pud_t pudval)
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);
517 static void vmi_pte_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
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));
524 static void vmi_pmd_clear(pmd_t *pmd)
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);
533 static void __devinit
534 vmi_startup_ipi_hook(int phys_apicid, unsigned long start_eip,
535 unsigned long start_esp)
537 struct vmi_ap_state ap;
539 /* Default everything to zero. This is fine for most GPRs. */
540 memset(&ap, 0, sizeof(struct vmi_ap_state));
542 ap.gdtr_limit = GDT_SIZE - 1;
543 ap.gdtr_base = (unsigned long) get_cpu_gdt_table(phys_apicid);
545 ap.idtr_limit = IDT_ENTRIES * 8 - 1;
546 ap.idtr_base = (unsigned long) idt_table;
551 ap.eip = (unsigned long) start_eip;
553 ap.esp = (unsigned long) start_esp;
557 ap.fs = __KERNEL_PERCPU;
562 #ifdef CONFIG_X86_PAE
563 /* efer should match BSP efer. */
566 rdmsr(MSR_EFER, l, h);
567 ap.efer = (unsigned long long) h << 32 | l;
571 ap.cr3 = __pa(swapper_pg_dir);
572 /* Protected mode, paging, AM, WP, NE, MP. */
574 ap.cr4 = mmu_cr4_features;
575 vmi_ops.set_initial_ap_state((u32)&ap, phys_apicid);
579 static void vmi_enter_lazy_cpu(void)
581 paravirt_enter_lazy_cpu();
582 vmi_ops.set_lazy_mode(2);
585 static void vmi_enter_lazy_mmu(void)
587 paravirt_enter_lazy_mmu();
588 vmi_ops.set_lazy_mode(1);
591 static void vmi_leave_lazy(void)
593 paravirt_leave_lazy(paravirt_get_lazy_mode());
594 vmi_ops.set_lazy_mode(0);
597 static inline int __init check_vmi_rom(struct vrom_header *rom)
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";
605 if (rom->rom_signature != 0xaa55)
607 if (rom->vrom_signature != VMI_SIGNATURE)
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);
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
622 if (!rom->pci_header_offs) {
623 printk(KERN_WARNING "VMI: ROM does not contain PCI header.\n");
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");
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;
642 if (rom->license_offs)
643 license = (char *)rom+rom->license_offs;
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);
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",
662 * Probe for the VMI option ROM
664 static inline int __init probe_vmi_rom(void)
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)) {
681 * VMI setup common to all processors
683 void vmi_bringup(void)
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);
691 * Return a pointer to a VMI function or NULL if unimplemented
693 static void *vmi_get_function(int vmicall)
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;
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.
710 #define para_fill(opname, vmicall) \
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",\
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.
731 #define para_wrap(opname, wrapper, cache, vmicall) \
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) { \
738 vmi_ops.cache = (void *)rel->eip; \
743 * Activate the VMI interface and switch into paravirtualized mode
745 static inline int __init activate_vmi(void)
749 const struct vmi_relocation_info *rel = (struct vmi_relocation_info *)&reloc;
751 if (call_vrom_func(vmi_rom, vmi_init) != 0) {
752 printk(KERN_ERR "VMI ROM failed to initialize!");
755 savesegment(cs, kernel_cs);
757 pv_info.paravirt_enabled = 1;
758 pv_info.kernel_rpl = kernel_cs & SEGMENT_RPL_MASK;
759 pv_info.name = "vmi";
761 pv_init_ops.patch = vmi_patch;
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.
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
778 /* CPUID is special, so very special it gets wrapped like a present */
779 para_wrap(pv_cpu_ops.cpuid, vmi_cpuid, cpuid, CPUID);
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);
797 para_fill(pv_cpu_ops.wbinvd, WBINVD);
798 para_fill(pv_cpu_ops.read_tsc, RDTSC);
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 */
805 /* TR interface doesn't pass TR value, wrap */
806 para_wrap(pv_cpu_ops.load_tr_desc, vmi_set_tr, set_tr, SetTR);
808 /* LDT is special, too */
809 para_wrap(pv_cpu_ops.set_ldt, vmi_set_ldt, _set_ldt, SetLDT);
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);
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);
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);
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);
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.
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);
851 vmi_ops.set_pte = vmi_get_function(VMI_CALL_SetPxE);
852 vmi_ops.update_pte = vmi_get_function(VMI_CALL_UpdatePxE);
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;
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;
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;
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;
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;
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.
900 pv_cpu_ops.irq_enable_sysexit = (void *)0xfeedbab0;
901 pv_cpu_ops.iret = (void *)0xbadbab0;
904 para_wrap(pv_apic_ops.startup_ipi_hook, vmi_startup_ipi_hook, set_initial_ap_state, SetInitialAPState);
907 #ifdef CONFIG_X86_LOCAL_APIC
908 para_fill(pv_apic_ops.apic_read, APICRead);
909 para_fill(pv_apic_ops.apic_write, APICWrite);
913 * Check for VMI timer functionality by probing for a cycle frequency method
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;
934 pv_time_ops.sched_clock = vmi_sched_clock;
935 pv_time_ops.get_tsc_khz = vmi_tsc_khz;
937 /* We have true wallclock functions; disable CMOS clock sync */
938 no_sync_cmos_clock = 1;
941 disable_vmi_timer = 1;
944 para_fill(pv_irq_ops.safe_halt, Halt);
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
952 apply_paravirt(__parainstructions, __parainstructions_end);
961 void __init vmi_init(void)
968 check_vmi_rom(vmi_rom);
970 /* In case probing for or validating the ROM failed, basil */
974 reserve_top_address(-vmi_rom->virtual_top);
976 local_irq_save(flags);
979 #ifdef CONFIG_X86_IO_APIC
980 /* This is virtual hardware; timer routing is wired correctly */
983 local_irq_restore(flags & X86_EFLAGS_IF);
986 static int __init parse_vmi(char *arg)
991 if (!strcmp(arg, "disable_pge")) {
992 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_PGE);
994 } else if (!strcmp(arg, "disable_pse")) {
995 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_PSE);
997 } else if (!strcmp(arg, "disable_sep")) {
998 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_SEP);
1000 } else if (!strcmp(arg, "disable_tsc")) {
1001 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_TSC);
1003 } else if (!strcmp(arg, "disable_mtrr")) {
1004 clear_cpu_cap(&boot_cpu_data, X86_FEATURE_MTRR);
1006 } else if (!strcmp(arg, "disable_timer")) {
1007 disable_vmi_timer = 1;
1009 } else if (!strcmp(arg, "disable_noidle"))
1014 early_param("vmi", parse_vmi);