Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2006, Rusty Russell <rusty@rustcorp.com.au> IBM Corporation. |
| 3 | * Copyright (C) 2007, Jes Sorensen <jes@sgi.com> SGI. |
| 4 | * |
| 5 | * This program is free software; you can redistribute it and/or modify |
| 6 | * it under the terms of the GNU General Public License as published by |
| 7 | * the Free Software Foundation; either version 2 of the License, or |
| 8 | * (at your option) any later version. |
| 9 | * |
| 10 | * This program is distributed in the hope that it will be useful, but |
| 11 | * WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 | * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or |
| 13 | * NON INFRINGEMENT. See the GNU General Public License for more |
| 14 | * details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU General Public License |
| 17 | * along with this program; if not, write to the Free Software |
| 18 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
| 19 | */ |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 20 | /*P:450 |
| 21 | * This file contains the x86-specific lguest code. It used to be all |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 22 | * mixed in with drivers/lguest/core.c but several foolhardy code slashers |
| 23 | * wrestled most of the dependencies out to here in preparation for porting |
| 24 | * lguest to other architectures (see what I mean by foolhardy?). |
| 25 | * |
| 26 | * This also contains a couple of non-obvious setup and teardown pieces which |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 27 | * were implemented after days of debugging pain. |
| 28 | :*/ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 29 | #include <linux/kernel.h> |
| 30 | #include <linux/start_kernel.h> |
| 31 | #include <linux/string.h> |
| 32 | #include <linux/console.h> |
| 33 | #include <linux/screen_info.h> |
| 34 | #include <linux/irq.h> |
| 35 | #include <linux/interrupt.h> |
| 36 | #include <linux/clocksource.h> |
| 37 | #include <linux/clockchips.h> |
| 38 | #include <linux/cpu.h> |
| 39 | #include <linux/lguest.h> |
| 40 | #include <linux/lguest_launcher.h> |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 41 | #include <asm/paravirt.h> |
| 42 | #include <asm/param.h> |
| 43 | #include <asm/page.h> |
| 44 | #include <asm/pgtable.h> |
| 45 | #include <asm/desc.h> |
| 46 | #include <asm/setup.h> |
| 47 | #include <asm/lguest.h> |
| 48 | #include <asm/uaccess.h> |
| 49 | #include <asm/i387.h> |
| 50 | #include "../lg.h" |
| 51 | |
| 52 | static int cpu_had_pge; |
| 53 | |
| 54 | static struct { |
| 55 | unsigned long offset; |
| 56 | unsigned short segment; |
| 57 | } lguest_entry; |
| 58 | |
| 59 | /* Offset from where switcher.S was compiled to where we've copied it */ |
| 60 | static unsigned long switcher_offset(void) |
| 61 | { |
| 62 | return SWITCHER_ADDR - (unsigned long)start_switcher_text; |
| 63 | } |
| 64 | |
| 65 | /* This cpu's struct lguest_pages. */ |
| 66 | static struct lguest_pages *lguest_pages(unsigned int cpu) |
| 67 | { |
| 68 | return &(((struct lguest_pages *) |
| 69 | (SWITCHER_ADDR + SHARED_SWITCHER_PAGES*PAGE_SIZE))[cpu]); |
| 70 | } |
| 71 | |
Tejun Heo | 390dfd9 | 2009-10-29 22:34:14 +0900 | [diff] [blame] | 72 | static DEFINE_PER_CPU(struct lg_cpu *, lg_last_cpu); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 73 | |
| 74 | /*S:010 |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 75 | * We approach the Switcher. |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 76 | * |
| 77 | * Remember that each CPU has two pages which are visible to the Guest when it |
| 78 | * runs on that CPU. This has to contain the state for that Guest: we copy the |
| 79 | * state in just before we run the Guest. |
| 80 | * |
| 81 | * Each Guest has "changed" flags which indicate what has changed in the Guest |
| 82 | * since it last ran. We saw this set in interrupts_and_traps.c and |
| 83 | * segments.c. |
| 84 | */ |
Glauber de Oliveira Costa | d0953d4 | 2008-01-07 11:05:25 -0200 | [diff] [blame] | 85 | static void copy_in_guest_info(struct lg_cpu *cpu, struct lguest_pages *pages) |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 86 | { |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 87 | /* |
| 88 | * Copying all this data can be quite expensive. We usually run the |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 89 | * same Guest we ran last time (and that Guest hasn't run anywhere else |
| 90 | * meanwhile). If that's not the case, we pretend everything in the |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 91 | * Guest has changed. |
| 92 | */ |
Christoph Lameter | c9f2954 | 2010-11-30 13:07:21 -0600 | [diff] [blame] | 93 | if (__this_cpu_read(lg_last_cpu) != cpu || cpu->last_pages != pages) { |
Rusty Russell | ced05dd | 2011-01-20 21:37:29 -0600 | [diff] [blame] | 94 | __this_cpu_write(lg_last_cpu, cpu); |
Glauber de Oliveira Costa | f34f8c5 | 2008-01-17 19:13:26 -0200 | [diff] [blame] | 95 | cpu->last_pages = pages; |
Glauber de Oliveira Costa | ae3749d | 2008-01-17 19:14:46 -0200 | [diff] [blame] | 96 | cpu->changed = CHANGED_ALL; |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 97 | } |
| 98 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 99 | /* |
| 100 | * These copies are pretty cheap, so we do them unconditionally: */ |
| 101 | /* Save the current Host top-level page directory. |
| 102 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 103 | pages->state.host_cr3 = __pa(current->mm->pgd); |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 104 | /* |
| 105 | * Set up the Guest's page tables to see this CPU's pages (and no |
| 106 | * other CPU's pages). |
| 107 | */ |
Glauber de Oliveira Costa | 0c78441 | 2008-01-07 11:05:30 -0200 | [diff] [blame] | 108 | map_switcher_in_guest(cpu, pages); |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 109 | /* |
| 110 | * Set up the two "TSS" members which tell the CPU what stack to use |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 111 | * for traps which do directly into the Guest (ie. traps at privilege |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 112 | * level 1). |
| 113 | */ |
Rusty Russell | e95035c | 2008-01-31 18:00:47 +1100 | [diff] [blame] | 114 | pages->state.guest_tss.sp1 = cpu->esp1; |
Glauber de Oliveira Costa | 4665ac8 | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 115 | pages->state.guest_tss.ss1 = cpu->ss1; |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 116 | |
| 117 | /* Copy direct-to-Guest trap entries. */ |
Glauber de Oliveira Costa | ae3749d | 2008-01-17 19:14:46 -0200 | [diff] [blame] | 118 | if (cpu->changed & CHANGED_IDT) |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 119 | copy_traps(cpu, pages->state.guest_idt, default_idt_entries); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 120 | |
| 121 | /* Copy all GDT entries which the Guest can change. */ |
Glauber de Oliveira Costa | ae3749d | 2008-01-17 19:14:46 -0200 | [diff] [blame] | 122 | if (cpu->changed & CHANGED_GDT) |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 123 | copy_gdt(cpu, pages->state.guest_gdt); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 124 | /* If only the TLS entries have changed, copy them. */ |
Glauber de Oliveira Costa | ae3749d | 2008-01-17 19:14:46 -0200 | [diff] [blame] | 125 | else if (cpu->changed & CHANGED_GDT_TLS) |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 126 | copy_gdt_tls(cpu, pages->state.guest_gdt); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 127 | |
| 128 | /* Mark the Guest as unchanged for next time. */ |
Glauber de Oliveira Costa | ae3749d | 2008-01-17 19:14:46 -0200 | [diff] [blame] | 129 | cpu->changed = 0; |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 130 | } |
| 131 | |
| 132 | /* Finally: the code to actually call into the Switcher to run the Guest. */ |
Glauber de Oliveira Costa | d0953d4 | 2008-01-07 11:05:25 -0200 | [diff] [blame] | 133 | static void run_guest_once(struct lg_cpu *cpu, struct lguest_pages *pages) |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 134 | { |
| 135 | /* This is a dummy value we need for GCC's sake. */ |
| 136 | unsigned int clobber; |
| 137 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 138 | /* |
| 139 | * Copy the guest-specific information into this CPU's "struct |
| 140 | * lguest_pages". |
| 141 | */ |
Glauber de Oliveira Costa | d0953d4 | 2008-01-07 11:05:25 -0200 | [diff] [blame] | 142 | copy_in_guest_info(cpu, pages); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 143 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 144 | /* |
| 145 | * Set the trap number to 256 (impossible value). If we fault while |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 146 | * switching to the Guest (bad segment registers or bug), this will |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 147 | * cause us to abort the Guest. |
| 148 | */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 149 | cpu->regs->trapnum = 256; |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 150 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 151 | /* |
| 152 | * Now: we push the "eflags" register on the stack, then do an "lcall". |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 153 | * This is how we change from using the kernel code segment to using |
| 154 | * the dedicated lguest code segment, as well as jumping into the |
| 155 | * Switcher. |
| 156 | * |
| 157 | * The lcall also pushes the old code segment (KERNEL_CS) onto the |
| 158 | * stack, then the address of this call. This stack layout happens to |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 159 | * exactly match the stack layout created by an interrupt... |
| 160 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 161 | asm volatile("pushf; lcall *lguest_entry" |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 162 | /* |
| 163 | * This is how we tell GCC that %eax ("a") and %ebx ("b") |
| 164 | * are changed by this routine. The "=" means output. |
| 165 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 166 | : "=a"(clobber), "=b"(clobber) |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 167 | /* |
| 168 | * %eax contains the pages pointer. ("0" refers to the |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 169 | * 0-th argument above, ie "a"). %ebx contains the |
| 170 | * physical address of the Guest's top-level page |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 171 | * directory. |
| 172 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 173 | : "0"(pages), "1"(__pa(cpu->lg->pgdirs[cpu->cpu_pgd].pgdir)) |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 174 | /* |
| 175 | * We tell gcc that all these registers could change, |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 176 | * which means we don't have to save and restore them in |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 177 | * the Switcher. |
| 178 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 179 | : "memory", "%edx", "%ecx", "%edi", "%esi"); |
| 180 | } |
| 181 | /*:*/ |
| 182 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 183 | /*M:002 |
| 184 | * There are hooks in the scheduler which we can register to tell when we |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 185 | * get kicked off the CPU (preempt_notifier_register()). This would allow us |
| 186 | * to lazily disable SYSENTER which would regain some performance, and should |
| 187 | * also simplify copy_in_guest_info(). Note that we'd still need to restore |
| 188 | * things when we exit to Launcher userspace, but that's fairly easy. |
| 189 | * |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 190 | * We could also try using these hooks for PGE, but that might be too expensive. |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 191 | * |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 192 | * The hooks were designed for KVM, but we can also put them to good use. |
| 193 | :*/ |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 194 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 195 | /*H:040 |
| 196 | * This is the i386-specific code to setup and run the Guest. Interrupts |
| 197 | * are disabled: we own the CPU. |
| 198 | */ |
Glauber de Oliveira Costa | d0953d4 | 2008-01-07 11:05:25 -0200 | [diff] [blame] | 199 | void lguest_arch_run_guest(struct lg_cpu *cpu) |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 200 | { |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 201 | /* |
| 202 | * Remember the awfully-named TS bit? If the Guest has asked to set it |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 203 | * we set it now, so we can trap and pass that trap to the Guest if it |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 204 | * uses the FPU. |
| 205 | */ |
Glauber de Oliveira Costa | 4665ac8 | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 206 | if (cpu->ts) |
Suresh Siddha | 54481cf8 | 2008-06-19 09:41:22 -0700 | [diff] [blame] | 207 | unlazy_fpu(current); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 208 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 209 | /* |
| 210 | * SYSENTER is an optimized way of doing system calls. We can't allow |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 211 | * it because it always jumps to privilege level 0. A normal Guest |
| 212 | * won't try it because we don't advertise it in CPUID, but a malicious |
| 213 | * Guest (or malicious Guest userspace program) could, so we tell the |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 214 | * CPU to disable it before running the Guest. |
| 215 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 216 | if (boot_cpu_has(X86_FEATURE_SEP)) |
| 217 | wrmsr(MSR_IA32_SYSENTER_CS, 0, 0); |
| 218 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 219 | /* |
| 220 | * Now we actually run the Guest. It will return when something |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 221 | * interesting happens, and we can examine its registers to see what it |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 222 | * was doing. |
| 223 | */ |
Glauber de Oliveira Costa | d0953d4 | 2008-01-07 11:05:25 -0200 | [diff] [blame] | 224 | run_guest_once(cpu, lguest_pages(raw_smp_processor_id())); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 225 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 226 | /* |
| 227 | * Note that the "regs" structure contains two extra entries which are |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 228 | * not really registers: a trap number which says what interrupt or |
| 229 | * trap made the switcher code come back, and an error code which some |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 230 | * traps set. |
| 231 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 232 | |
Suresh Siddha | 54481cf8 | 2008-06-19 09:41:22 -0700 | [diff] [blame] | 233 | /* Restore SYSENTER if it's supposed to be on. */ |
| 234 | if (boot_cpu_has(X86_FEATURE_SEP)) |
| 235 | wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0); |
| 236 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 237 | /* |
| 238 | * If the Guest page faulted, then the cr2 register will tell us the |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 239 | * bad virtual address. We have to grab this now, because once we |
| 240 | * re-enable interrupts an interrupt could fault and thus overwrite |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 241 | * cr2, or we could even move off to a different CPU. |
| 242 | */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 243 | if (cpu->regs->trapnum == 14) |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 244 | cpu->arch.last_pagefault = read_cr2(); |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 245 | /* |
| 246 | * Similarly, if we took a trap because the Guest used the FPU, |
Suresh Siddha | 54481cf8 | 2008-06-19 09:41:22 -0700 | [diff] [blame] | 247 | * we have to restore the FPU it expects to see. |
| 248 | * math_state_restore() may sleep and we may even move off to |
| 249 | * a different CPU. So all the critical stuff should be done |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 250 | * before this. |
| 251 | */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 252 | else if (cpu->regs->trapnum == 7) |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 253 | math_state_restore(); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 254 | } |
| 255 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 256 | /*H:130 |
| 257 | * Now we've examined the hypercall code; our Guest can make requests. |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 258 | * Our Guest is usually so well behaved; it never tries to do things it isn't |
| 259 | * allowed to, and uses hypercalls instead. Unfortunately, Linux's paravirtual |
| 260 | * infrastructure isn't quite complete, because it doesn't contain replacements |
| 261 | * for the Intel I/O instructions. As a result, the Guest sometimes fumbles |
| 262 | * across one during the boot process as it probes for various things which are |
| 263 | * usually attached to a PC. |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 264 | * |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 265 | * When the Guest uses one of these instructions, we get a trap (General |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 266 | * Protection Fault) and come here. We see if it's one of those troublesome |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 267 | * instructions and skip over it. We return true if we did. |
| 268 | */ |
Glauber de Oliveira Costa | a3863f6 | 2008-01-07 11:05:31 -0200 | [diff] [blame] | 269 | static int emulate_insn(struct lg_cpu *cpu) |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 270 | { |
| 271 | u8 insn; |
Rusty Russell | 996ba96 | 2011-07-22 14:39:51 +0930 | [diff] [blame] | 272 | unsigned int insnlen = 0, in = 0, small_operand = 0; |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 273 | /* |
| 274 | * The eip contains the *virtual* address of the Guest's instruction: |
Rusty Russell | 9f54288 | 2011-07-22 14:39:50 +0930 | [diff] [blame] | 275 | * walk the Guest's page tables to find the "physical" address. |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 276 | */ |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 277 | unsigned long physaddr = guest_pa(cpu, cpu->regs->eip); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 278 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 279 | /* |
| 280 | * This must be the Guest kernel trying to do something, not userspace! |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 281 | * The bottom two bits of the CS segment register are the privilege |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 282 | * level. |
| 283 | */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 284 | if ((cpu->regs->cs & 3) != GUEST_PL) |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 285 | return 0; |
| 286 | |
| 287 | /* Decoding x86 instructions is icky. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 288 | insn = lgread(cpu, physaddr, u8); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 289 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 290 | /* |
Rusty Russell | 5094aea | 2010-04-14 21:43:53 -0600 | [diff] [blame] | 291 | * Around 2.6.33, the kernel started using an emulation for the |
| 292 | * cmpxchg8b instruction in early boot on many configurations. This |
| 293 | * code isn't paravirtualized, and it tries to disable interrupts. |
| 294 | * Ignore it, which will Mostly Work. |
| 295 | */ |
| 296 | if (insn == 0xfa) { |
| 297 | /* "cli", or Clear Interrupt Enable instruction. Skip it. */ |
| 298 | cpu->regs->eip++; |
| 299 | return 1; |
| 300 | } |
| 301 | |
| 302 | /* |
Rusty Russell | 996ba96 | 2011-07-22 14:39:51 +0930 | [diff] [blame] | 303 | * 0x66 is an "operand prefix". It means a 16, not 32 bit in/out. |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 304 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 305 | if (insn == 0x66) { |
Rusty Russell | 996ba96 | 2011-07-22 14:39:51 +0930 | [diff] [blame] | 306 | small_operand = 1; |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 307 | /* The instruction is 1 byte so far, read the next byte. */ |
| 308 | insnlen = 1; |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 309 | insn = lgread(cpu, physaddr + insnlen, u8); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 310 | } |
| 311 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 312 | /* |
| 313 | * We can ignore the lower bit for the moment and decode the 4 opcodes |
| 314 | * we need to emulate. |
| 315 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 316 | switch (insn & 0xFE) { |
| 317 | case 0xE4: /* in <next byte>,%al */ |
| 318 | insnlen += 2; |
| 319 | in = 1; |
| 320 | break; |
| 321 | case 0xEC: /* in (%dx),%al */ |
| 322 | insnlen += 1; |
| 323 | in = 1; |
| 324 | break; |
| 325 | case 0xE6: /* out %al,<next byte> */ |
| 326 | insnlen += 2; |
| 327 | break; |
| 328 | case 0xEE: /* out %al,(%dx) */ |
| 329 | insnlen += 1; |
| 330 | break; |
| 331 | default: |
| 332 | /* OK, we don't know what this is, can't emulate. */ |
| 333 | return 0; |
| 334 | } |
| 335 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 336 | /* |
| 337 | * If it was an "IN" instruction, they expect the result to be read |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 338 | * into %eax, so we change %eax. We always return all-ones, which |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 339 | * traditionally means "there's nothing there". |
| 340 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 341 | if (in) { |
Rusty Russell | 996ba96 | 2011-07-22 14:39:51 +0930 | [diff] [blame] | 342 | /* Lower bit tells means it's a 32/16 bit access */ |
| 343 | if (insn & 0x1) { |
| 344 | if (small_operand) |
| 345 | cpu->regs->eax |= 0xFFFF; |
| 346 | else |
| 347 | cpu->regs->eax = 0xFFFFFFFF; |
| 348 | } else |
| 349 | cpu->regs->eax |= 0xFF; |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 350 | } |
| 351 | /* Finally, we've "done" the instruction, so move past it. */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 352 | cpu->regs->eip += insnlen; |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 353 | /* Success! */ |
| 354 | return 1; |
| 355 | } |
| 356 | |
| 357 | /*H:050 Once we've re-enabled interrupts, we look at why the Guest exited. */ |
Glauber de Oliveira Costa | 73044f0 | 2008-01-07 11:05:27 -0200 | [diff] [blame] | 358 | void lguest_arch_handle_trap(struct lg_cpu *cpu) |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 359 | { |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 360 | switch (cpu->regs->trapnum) { |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 361 | case 13: /* We've intercepted a General Protection Fault. */ |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 362 | /* |
| 363 | * Check if this was one of those annoying IN or OUT |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 364 | * instructions which we need to emulate. If so, we just go |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 365 | * back into the Guest after we've done it. |
| 366 | */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 367 | if (cpu->regs->errcode == 0) { |
Glauber de Oliveira Costa | a3863f6 | 2008-01-07 11:05:31 -0200 | [diff] [blame] | 368 | if (emulate_insn(cpu)) |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 369 | return; |
| 370 | } |
| 371 | break; |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 372 | case 14: /* We've intercepted a Page Fault. */ |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 373 | /* |
| 374 | * The Guest accessed a virtual address that wasn't mapped. |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 375 | * This happens a lot: we don't actually set up most of the page |
| 376 | * tables for the Guest at all when we start: as it runs it asks |
| 377 | * for more and more, and we set them up as required. In this |
| 378 | * case, we don't even tell the Guest that the fault happened. |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 379 | * |
| 380 | * The errcode tells whether this was a read or a write, and |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 381 | * whether kernel or userspace code. |
| 382 | */ |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 383 | if (demand_page(cpu, cpu->arch.last_pagefault, |
| 384 | cpu->regs->errcode)) |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 385 | return; |
| 386 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 387 | /* |
| 388 | * OK, it's really not there (or not OK): the Guest needs to |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 389 | * know. We write out the cr2 value so it knows where the |
| 390 | * fault occurred. |
| 391 | * |
| 392 | * Note that if the Guest were really messed up, this could |
| 393 | * happen before it's done the LHCALL_LGUEST_INIT hypercall, so |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 394 | * lg->lguest_data could be NULL |
| 395 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 396 | if (cpu->lg->lguest_data && |
| 397 | put_user(cpu->arch.last_pagefault, |
| 398 | &cpu->lg->lguest_data->cr2)) |
| 399 | kill_guest(cpu, "Writing cr2"); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 400 | break; |
| 401 | case 7: /* We've intercepted a Device Not Available fault. */ |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 402 | /* |
| 403 | * If the Guest doesn't want to know, we already restored the |
| 404 | * Floating Point Unit, so we just continue without telling it. |
| 405 | */ |
Glauber de Oliveira Costa | 4665ac8 | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 406 | if (!cpu->ts) |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 407 | return; |
| 408 | break; |
| 409 | case 32 ... 255: |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 410 | /* |
| 411 | * These values mean a real interrupt occurred, in which case |
Matias Zabaljauregui | 4cd8b5e | 2009-03-14 13:37:52 -0200 | [diff] [blame] | 412 | * the Host handler has already been run. We just do a |
Rusty Russell | cc6d4fb | 2007-10-22 11:03:30 +1000 | [diff] [blame] | 413 | * friendly check if another process should now be run, then |
Rusty Russell | 9f54288 | 2011-07-22 14:39:50 +0930 | [diff] [blame] | 414 | * return to run the Guest again. |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 415 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 416 | cond_resched(); |
Rusty Russell | cc6d4fb | 2007-10-22 11:03:30 +1000 | [diff] [blame] | 417 | return; |
| 418 | case LGUEST_TRAP_ENTRY: |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 419 | /* |
| 420 | * Our 'struct hcall_args' maps directly over our regs: we set |
| 421 | * up the pointer now to indicate a hypercall is pending. |
| 422 | */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 423 | cpu->hcall = (struct hcall_args *)cpu->regs; |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 424 | return; |
| 425 | } |
| 426 | |
| 427 | /* We didn't handle the trap, so it needs to go to the Guest. */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 428 | if (!deliver_trap(cpu, cpu->regs->trapnum)) |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 429 | /* |
| 430 | * If the Guest doesn't have a handler (either it hasn't |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 431 | * registered any yet, or it's one of the faults we don't let |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 432 | * it handle), it dies with this cryptic error message. |
| 433 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 434 | kill_guest(cpu, "unhandled trap %li at %#lx (%#lx)", |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 435 | cpu->regs->trapnum, cpu->regs->eip, |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 436 | cpu->regs->trapnum == 14 ? cpu->arch.last_pagefault |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 437 | : cpu->regs->errcode); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 438 | } |
| 439 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 440 | /* |
| 441 | * Now we can look at each of the routines this calls, in increasing order of |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 442 | * complexity: do_hypercalls(), emulate_insn(), maybe_do_interrupt(), |
| 443 | * deliver_trap() and demand_page(). After all those, we'll be ready to |
| 444 | * examine the Switcher, and our philosophical understanding of the Host/Guest |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 445 | * duality will be complete. |
| 446 | :*/ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 447 | static void adjust_pge(void *on) |
| 448 | { |
| 449 | if (on) |
| 450 | write_cr4(read_cr4() | X86_CR4_PGE); |
| 451 | else |
| 452 | write_cr4(read_cr4() & ~X86_CR4_PGE); |
| 453 | } |
| 454 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 455 | /*H:020 |
| 456 | * Now the Switcher is mapped and every thing else is ready, we need to do |
| 457 | * some more i386-specific initialization. |
| 458 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 459 | void __init lguest_arch_host_init(void) |
| 460 | { |
| 461 | int i; |
| 462 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 463 | /* |
Rusty Russell | 9f54288 | 2011-07-22 14:39:50 +0930 | [diff] [blame] | 464 | * Most of the x86/switcher_32.S doesn't care that it's been moved; on |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 465 | * Intel, jumps are relative, and it doesn't access any references to |
| 466 | * external code or data. |
| 467 | * |
| 468 | * The only exception is the interrupt handlers in switcher.S: their |
| 469 | * addresses are placed in a table (default_idt_entries), so we need to |
| 470 | * update the table with the new addresses. switcher_offset() is a |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 471 | * convenience function which returns the distance between the |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 472 | * compiled-in switcher code and the high-mapped copy we just made. |
| 473 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 474 | for (i = 0; i < IDT_ENTRIES; i++) |
| 475 | default_idt_entries[i] += switcher_offset(); |
| 476 | |
| 477 | /* |
| 478 | * Set up the Switcher's per-cpu areas. |
| 479 | * |
| 480 | * Each CPU gets two pages of its own within the high-mapped region |
| 481 | * (aka. "struct lguest_pages"). Much of this can be initialized now, |
| 482 | * but some depends on what Guest we are running (which is set up in |
| 483 | * copy_in_guest_info()). |
| 484 | */ |
| 485 | for_each_possible_cpu(i) { |
| 486 | /* lguest_pages() returns this CPU's two pages. */ |
| 487 | struct lguest_pages *pages = lguest_pages(i); |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 488 | /* This is a convenience pointer to make the code neater. */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 489 | struct lguest_ro_state *state = &pages->state; |
| 490 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 491 | /* |
| 492 | * The Global Descriptor Table: the Host has a different one |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 493 | * for each CPU. We keep a descriptor for the GDT which says |
| 494 | * where it is and how big it is (the size is actually the last |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 495 | * byte, not the size, hence the "-1"). |
| 496 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 497 | state->host_gdt_desc.size = GDT_SIZE-1; |
| 498 | state->host_gdt_desc.address = (long)get_cpu_gdt_table(i); |
| 499 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 500 | /* |
| 501 | * All CPUs on the Host use the same Interrupt Descriptor |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 502 | * Table, so we just use store_idt(), which gets this CPU's IDT |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 503 | * descriptor. |
| 504 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 505 | store_idt(&state->host_idt_desc); |
| 506 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 507 | /* |
| 508 | * The descriptors for the Guest's GDT and IDT can be filled |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 509 | * out now, too. We copy the GDT & IDT into ->guest_gdt and |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 510 | * ->guest_idt before actually running the Guest. |
| 511 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 512 | state->guest_idt_desc.size = sizeof(state->guest_idt)-1; |
| 513 | state->guest_idt_desc.address = (long)&state->guest_idt; |
| 514 | state->guest_gdt_desc.size = sizeof(state->guest_gdt)-1; |
| 515 | state->guest_gdt_desc.address = (long)&state->guest_gdt; |
| 516 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 517 | /* |
| 518 | * We know where we want the stack to be when the Guest enters |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 519 | * the Switcher: in pages->regs. The stack grows upwards, so |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 520 | * we start it at the end of that structure. |
| 521 | */ |
H. Peter Anvin | faca622 | 2008-01-30 13:31:02 +0100 | [diff] [blame] | 522 | state->guest_tss.sp0 = (long)(&pages->regs + 1); |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 523 | /* |
| 524 | * And this is the GDT entry to use for the stack: we keep a |
| 525 | * couple of special LGUEST entries. |
| 526 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 527 | state->guest_tss.ss0 = LGUEST_DS; |
| 528 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 529 | /* |
| 530 | * x86 can have a finegrained bitmap which indicates what I/O |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 531 | * ports the process can use. We set it to the end of our |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 532 | * structure, meaning "none". |
| 533 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 534 | state->guest_tss.io_bitmap_base = sizeof(state->guest_tss); |
| 535 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 536 | /* |
| 537 | * Some GDT entries are the same across all Guests, so we can |
| 538 | * set them up now. |
| 539 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 540 | setup_default_gdt_entries(state); |
| 541 | /* Most IDT entries are the same for all Guests, too.*/ |
| 542 | setup_default_idt_entries(state, default_idt_entries); |
| 543 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 544 | /* |
| 545 | * The Host needs to be able to use the LGUEST segments on this |
| 546 | * CPU, too, so put them in the Host GDT. |
| 547 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 548 | get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT; |
| 549 | get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT; |
| 550 | } |
| 551 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 552 | /* |
| 553 | * In the Switcher, we want the %cs segment register to use the |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 554 | * LGUEST_CS GDT entry: we've put that in the Host and Guest GDTs, so |
| 555 | * it will be undisturbed when we switch. To change %cs and jump we |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 556 | * need this structure to feed to Intel's "lcall" instruction. |
| 557 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 558 | lguest_entry.offset = (long)switch_to_guest + switcher_offset(); |
| 559 | lguest_entry.segment = LGUEST_CS; |
| 560 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 561 | /* |
| 562 | * Finally, we need to turn off "Page Global Enable". PGE is an |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 563 | * optimization where page table entries are specially marked to show |
| 564 | * they never change. The Host kernel marks all the kernel pages this |
| 565 | * way because it's always present, even when userspace is running. |
| 566 | * |
| 567 | * Lguest breaks this: unbeknownst to the rest of the Host kernel, we |
| 568 | * switch to the Guest kernel. If you don't disable this on all CPUs, |
| 569 | * you'll get really weird bugs that you'll chase for two days. |
| 570 | * |
| 571 | * I used to turn PGE off every time we switched to the Guest and back |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 572 | * on when we return, but that slowed the Switcher down noticibly. |
| 573 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 574 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 575 | /* |
| 576 | * We don't need the complexity of CPUs coming and going while we're |
| 577 | * doing this. |
| 578 | */ |
Gautham R Shenoy | 86ef5c9 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 579 | get_online_cpus(); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 580 | if (cpu_has_pge) { /* We have a broader idea of "global". */ |
| 581 | /* Remember that this was originally set (for cleanup). */ |
| 582 | cpu_had_pge = 1; |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 583 | /* |
| 584 | * adjust_pge is a helper function which sets or unsets the PGE |
| 585 | * bit on its CPU, depending on the argument (0 == unset). |
| 586 | */ |
Jens Axboe | 15c8b6c | 2008-05-09 09:39:44 +0200 | [diff] [blame] | 587 | on_each_cpu(adjust_pge, (void *)0, 1); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 588 | /* Turn off the feature in the global feature set. */ |
Andrew Morton | cf485e5 | 2008-06-09 16:22:48 -0700 | [diff] [blame] | 589 | clear_cpu_cap(&boot_cpu_data, X86_FEATURE_PGE); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 590 | } |
Gautham R Shenoy | 86ef5c9 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 591 | put_online_cpus(); |
Rusty Russell | 9f54288 | 2011-07-22 14:39:50 +0930 | [diff] [blame] | 592 | } |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 593 | /*:*/ |
| 594 | |
| 595 | void __exit lguest_arch_host_fini(void) |
| 596 | { |
| 597 | /* If we had PGE before we started, turn it back on now. */ |
Gautham R Shenoy | 86ef5c9 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 598 | get_online_cpus(); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 599 | if (cpu_had_pge) { |
Andrew Morton | cf485e5 | 2008-06-09 16:22:48 -0700 | [diff] [blame] | 600 | set_cpu_cap(&boot_cpu_data, X86_FEATURE_PGE); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 601 | /* adjust_pge's argument "1" means set PGE. */ |
Jens Axboe | 15c8b6c | 2008-05-09 09:39:44 +0200 | [diff] [blame] | 602 | on_each_cpu(adjust_pge, (void *)1, 1); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 603 | } |
Gautham R Shenoy | 86ef5c9 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 604 | put_online_cpus(); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 605 | } |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 606 | |
| 607 | |
| 608 | /*H:122 The i386-specific hypercalls simply farm out to the right functions. */ |
Glauber de Oliveira Costa | 73044f0 | 2008-01-07 11:05:27 -0200 | [diff] [blame] | 609 | int lguest_arch_do_hcall(struct lg_cpu *cpu, struct hcall_args *args) |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 610 | { |
| 611 | switch (args->arg0) { |
Rusty Russell | a489f0b | 2009-04-19 23:14:00 -0600 | [diff] [blame] | 612 | case LHCALL_LOAD_GDT_ENTRY: |
| 613 | load_guest_gdt_entry(cpu, args->arg1, args->arg2, args->arg3); |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 614 | break; |
| 615 | case LHCALL_LOAD_IDT_ENTRY: |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 616 | load_guest_idt_entry(cpu, args->arg1, args->arg2, args->arg3); |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 617 | break; |
| 618 | case LHCALL_LOAD_TLS: |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 619 | guest_load_tls(cpu, args->arg1); |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 620 | break; |
| 621 | default: |
| 622 | /* Bad Guest. Bad! */ |
| 623 | return -EIO; |
| 624 | } |
| 625 | return 0; |
| 626 | } |
| 627 | |
| 628 | /*H:126 i386-specific hypercall initialization: */ |
Glauber de Oliveira Costa | 73044f0 | 2008-01-07 11:05:27 -0200 | [diff] [blame] | 629 | int lguest_arch_init_hypercalls(struct lg_cpu *cpu) |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 630 | { |
| 631 | u32 tsc_speed; |
| 632 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 633 | /* |
| 634 | * The pointer to the Guest's "struct lguest_data" is the only argument. |
| 635 | * We check that address now. |
| 636 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 637 | if (!lguest_address_ok(cpu->lg, cpu->hcall->arg1, |
| 638 | sizeof(*cpu->lg->lguest_data))) |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 639 | return -EFAULT; |
| 640 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 641 | /* |
| 642 | * Having checked it, we simply set lg->lguest_data to point straight |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 643 | * into the Launcher's memory at the right place and then use |
| 644 | * copy_to_user/from_user from now on, instead of lgread/write. I put |
| 645 | * this in to show that I'm not immune to writing stupid |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 646 | * optimizations. |
| 647 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 648 | cpu->lg->lguest_data = cpu->lg->mem_base + cpu->hcall->arg1; |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 649 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 650 | /* |
| 651 | * We insist that the Time Stamp Counter exist and doesn't change with |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 652 | * cpu frequency. Some devious chip manufacturers decided that TSC |
| 653 | * changes could be handled in software. I decided that time going |
| 654 | * backwards might be good for benchmarks, but it's bad for users. |
| 655 | * |
| 656 | * We also insist that the TSC be stable: the kernel detects unreliable |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 657 | * TSCs for its own purposes, and we use that here. |
| 658 | */ |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 659 | if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC) && !check_tsc_unstable()) |
| 660 | tsc_speed = tsc_khz; |
| 661 | else |
| 662 | tsc_speed = 0; |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 663 | if (put_user(tsc_speed, &cpu->lg->lguest_data->tsc_khz)) |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 664 | return -EFAULT; |
| 665 | |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 666 | /* The interrupt code might not like the system call vector. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 667 | if (!check_syscall_vector(cpu->lg)) |
| 668 | kill_guest(cpu, "bad syscall vector"); |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 669 | |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 670 | return 0; |
| 671 | } |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 672 | /*:*/ |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 673 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 674 | /*L:030 |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 675 | * Most of the Guest's registers are left alone: we used get_zeroed_page() to |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 676 | * allocate the structure, so they will be 0. |
| 677 | */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 678 | void lguest_arch_setup_regs(struct lg_cpu *cpu, unsigned long start) |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 679 | { |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 680 | struct lguest_regs *regs = cpu->regs; |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 681 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 682 | /* |
| 683 | * There are four "segment" registers which the Guest needs to boot: |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 684 | * The "code segment" register (cs) refers to the kernel code segment |
| 685 | * __KERNEL_CS, and the "data", "extra" and "stack" segment registers |
| 686 | * refer to the kernel data segment __KERNEL_DS. |
| 687 | * |
| 688 | * The privilege level is packed into the lower bits. The Guest runs |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 689 | * at privilege level 1 (GUEST_PL). |
| 690 | */ |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 691 | regs->ds = regs->es = regs->ss = __KERNEL_DS|GUEST_PL; |
| 692 | regs->cs = __KERNEL_CS|GUEST_PL; |
| 693 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 694 | /* |
| 695 | * The "eflags" register contains miscellaneous flags. Bit 1 (0x002) |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 696 | * is supposed to always be "1". Bit 9 (0x200) controls whether |
| 697 | * interrupts are enabled. We always leave interrupts enabled while |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 698 | * running the Guest. |
| 699 | */ |
Rusty Russell | 25c47bb | 2007-10-25 14:09:53 +1000 | [diff] [blame] | 700 | regs->eflags = X86_EFLAGS_IF | 0x2; |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 701 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 702 | /* |
| 703 | * The "Extended Instruction Pointer" register says where the Guest is |
| 704 | * running. |
| 705 | */ |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 706 | regs->eip = start; |
| 707 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 708 | /* |
| 709 | * %esi points to our boot information, at physical address 0, so don't |
| 710 | * touch it. |
| 711 | */ |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 712 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 713 | /* There are a couple of GDT entries the Guest expects at boot. */ |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 714 | setup_guest_gdt(cpu); |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 715 | } |