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> |
Ingo Molnar | 952f07e | 2015-04-26 16:56:05 +0200 | [diff] [blame] | 49 | #include <asm/fpu/internal.h> |
Andy Lutomirski | 375074c | 2014-10-24 15:58:07 -0700 | [diff] [blame] | 50 | #include <asm/tlbflush.h> |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 51 | #include "../lg.h" |
| 52 | |
| 53 | static int cpu_had_pge; |
| 54 | |
| 55 | static struct { |
| 56 | unsigned long offset; |
| 57 | unsigned short segment; |
| 58 | } lguest_entry; |
| 59 | |
| 60 | /* Offset from where switcher.S was compiled to where we've copied it */ |
| 61 | static unsigned long switcher_offset(void) |
| 62 | { |
Rusty Russell | 406a590b | 2013-04-22 14:10:37 +0930 | [diff] [blame] | 63 | return switcher_addr - (unsigned long)start_switcher_text; |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 64 | } |
| 65 | |
Rusty Russell | 93a2cdf | 2013-04-22 14:10:38 +0930 | [diff] [blame] | 66 | /* This cpu's struct lguest_pages (after the Switcher text page) */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 67 | static struct lguest_pages *lguest_pages(unsigned int cpu) |
| 68 | { |
Rusty Russell | 93a2cdf | 2013-04-22 14:10:38 +0930 | [diff] [blame] | 69 | return &(((struct lguest_pages *)(switcher_addr + PAGE_SIZE))[cpu]); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 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 | */ |
Andi Kleen | cdd77e8 | 2013-11-05 21:22:28 +1030 | [diff] [blame] | 161 | asm volatile("pushf; lcall *%4" |
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 | */ |
Andi Kleen | cdd77e8 | 2013-11-05 21:22:28 +1030 | [diff] [blame] | 173 | : "0"(pages), |
| 174 | "1"(__pa(cpu->lg->pgdirs[cpu->cpu_pgd].pgdir)), |
| 175 | "m"(lguest_entry) |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 176 | /* |
| 177 | * We tell gcc that all these registers could change, |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 178 | * which means we don't have to save and restore them in |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 179 | * the Switcher. |
| 180 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 181 | : "memory", "%edx", "%ecx", "%edi", "%esi"); |
| 182 | } |
| 183 | /*:*/ |
| 184 | |
Rusty Russell | 18c1373 | 2015-02-11 15:15:09 +1030 | [diff] [blame] | 185 | unsigned long *lguest_arch_regptr(struct lg_cpu *cpu, size_t reg_off, bool any) |
| 186 | { |
| 187 | switch (reg_off) { |
| 188 | case offsetof(struct pt_regs, bx): |
| 189 | return &cpu->regs->ebx; |
| 190 | case offsetof(struct pt_regs, cx): |
| 191 | return &cpu->regs->ecx; |
| 192 | case offsetof(struct pt_regs, dx): |
| 193 | return &cpu->regs->edx; |
| 194 | case offsetof(struct pt_regs, si): |
| 195 | return &cpu->regs->esi; |
| 196 | case offsetof(struct pt_regs, di): |
| 197 | return &cpu->regs->edi; |
| 198 | case offsetof(struct pt_regs, bp): |
| 199 | return &cpu->regs->ebp; |
| 200 | case offsetof(struct pt_regs, ax): |
| 201 | return &cpu->regs->eax; |
| 202 | case offsetof(struct pt_regs, ip): |
| 203 | return &cpu->regs->eip; |
| 204 | case offsetof(struct pt_regs, sp): |
| 205 | return &cpu->regs->esp; |
| 206 | } |
| 207 | |
| 208 | /* Launcher can read these, but we don't allow any setting. */ |
| 209 | if (any) { |
| 210 | switch (reg_off) { |
| 211 | case offsetof(struct pt_regs, ds): |
| 212 | return &cpu->regs->ds; |
| 213 | case offsetof(struct pt_regs, es): |
| 214 | return &cpu->regs->es; |
| 215 | case offsetof(struct pt_regs, fs): |
| 216 | return &cpu->regs->fs; |
| 217 | case offsetof(struct pt_regs, gs): |
| 218 | return &cpu->regs->gs; |
| 219 | case offsetof(struct pt_regs, cs): |
| 220 | return &cpu->regs->cs; |
| 221 | case offsetof(struct pt_regs, flags): |
| 222 | return &cpu->regs->eflags; |
| 223 | case offsetof(struct pt_regs, ss): |
| 224 | return &cpu->regs->ss; |
| 225 | } |
| 226 | } |
| 227 | |
| 228 | return NULL; |
| 229 | } |
| 230 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 231 | /*M:002 |
| 232 | * 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] | 233 | * get kicked off the CPU (preempt_notifier_register()). This would allow us |
| 234 | * to lazily disable SYSENTER which would regain some performance, and should |
| 235 | * also simplify copy_in_guest_info(). Note that we'd still need to restore |
| 236 | * things when we exit to Launcher userspace, but that's fairly easy. |
| 237 | * |
Rusty Russell | a91d74a | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 238 | * 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] | 239 | * |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 240 | * The hooks were designed for KVM, but we can also put them to good use. |
| 241 | :*/ |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 242 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 243 | /*H:040 |
| 244 | * This is the i386-specific code to setup and run the Guest. Interrupts |
| 245 | * are disabled: we own the CPU. |
| 246 | */ |
Glauber de Oliveira Costa | d0953d4 | 2008-01-07 11:05:25 -0200 | [diff] [blame] | 247 | void lguest_arch_run_guest(struct lg_cpu *cpu) |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 248 | { |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 249 | /* |
| 250 | * 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] | 251 | * 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] | 252 | * uses the FPU. |
| 253 | */ |
Ingo Molnar | 3c6dffa | 2015-04-28 12:28:08 +0200 | [diff] [blame] | 254 | if (cpu->ts && fpregs_active()) |
Suresh Siddha | 9c6ff8bb | 2012-08-24 14:13:01 -0700 | [diff] [blame] | 255 | stts(); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 256 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 257 | /* |
| 258 | * 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] | 259 | * it because it always jumps to privilege level 0. A normal Guest |
| 260 | * won't try it because we don't advertise it in CPUID, but a malicious |
| 261 | * Guest (or malicious Guest userspace program) could, so we tell the |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 262 | * CPU to disable it before running the Guest. |
| 263 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 264 | if (boot_cpu_has(X86_FEATURE_SEP)) |
| 265 | wrmsr(MSR_IA32_SYSENTER_CS, 0, 0); |
| 266 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 267 | /* |
| 268 | * Now we actually run the Guest. It will return when something |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 269 | * interesting happens, and we can examine its registers to see what it |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 270 | * was doing. |
| 271 | */ |
Glauber de Oliveira Costa | d0953d4 | 2008-01-07 11:05:25 -0200 | [diff] [blame] | 272 | run_guest_once(cpu, lguest_pages(raw_smp_processor_id())); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 273 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 274 | /* |
| 275 | * Note that the "regs" structure contains two extra entries which are |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 276 | * not really registers: a trap number which says what interrupt or |
| 277 | * 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] | 278 | * traps set. |
| 279 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 280 | |
Suresh Siddha | 54481cf8 | 2008-06-19 09:41:22 -0700 | [diff] [blame] | 281 | /* Restore SYSENTER if it's supposed to be on. */ |
| 282 | if (boot_cpu_has(X86_FEATURE_SEP)) |
| 283 | wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0); |
| 284 | |
Suresh Siddha | 9c6ff8bb | 2012-08-24 14:13:01 -0700 | [diff] [blame] | 285 | /* Clear the host TS bit if it was set above. */ |
Ingo Molnar | 3c6dffa | 2015-04-28 12:28:08 +0200 | [diff] [blame] | 286 | if (cpu->ts && fpregs_active()) |
Suresh Siddha | 9c6ff8bb | 2012-08-24 14:13:01 -0700 | [diff] [blame] | 287 | clts(); |
| 288 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 289 | /* |
| 290 | * 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] | 291 | * bad virtual address. We have to grab this now, because once we |
| 292 | * re-enable interrupts an interrupt could fault and thus overwrite |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 293 | * cr2, or we could even move off to a different CPU. |
| 294 | */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 295 | if (cpu->regs->trapnum == 14) |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 296 | cpu->arch.last_pagefault = read_cr2(); |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 297 | /* |
| 298 | * Similarly, if we took a trap because the Guest used the FPU, |
Suresh Siddha | 54481cf8 | 2008-06-19 09:41:22 -0700 | [diff] [blame] | 299 | * we have to restore the FPU it expects to see. |
Ingo Molnar | 3a0aee4 | 2015-04-22 13:16:47 +0200 | [diff] [blame] | 300 | * fpu__restore() may sleep and we may even move off to |
Suresh Siddha | 54481cf8 | 2008-06-19 09:41:22 -0700 | [diff] [blame] | 301 | * a different CPU. So all the critical stuff should be done |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 302 | * before this. |
| 303 | */ |
Ingo Molnar | 3c6dffa | 2015-04-28 12:28:08 +0200 | [diff] [blame] | 304 | else if (cpu->regs->trapnum == 7 && !fpregs_active()) |
Ingo Molnar | e1884d6 | 2015-05-04 11:49:58 +0200 | [diff] [blame] | 305 | fpu__restore(¤t->thread.fpu); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 306 | } |
| 307 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 308 | /*H:130 |
| 309 | * Now we've examined the hypercall code; our Guest can make requests. |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 310 | * Our Guest is usually so well behaved; it never tries to do things it isn't |
| 311 | * allowed to, and uses hypercalls instead. Unfortunately, Linux's paravirtual |
| 312 | * infrastructure isn't quite complete, because it doesn't contain replacements |
| 313 | * for the Intel I/O instructions. As a result, the Guest sometimes fumbles |
| 314 | * across one during the boot process as it probes for various things which are |
| 315 | * usually attached to a PC. |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 316 | * |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 317 | * When the Guest uses one of these instructions, we get a trap (General |
Rusty Russell | c565650b | 2015-02-11 15:15:10 +1030 | [diff] [blame] | 318 | * Protection Fault) and come here. We queue this to be sent out to the |
| 319 | * Launcher to handle. |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 320 | */ |
Rusty Russell | c565650b | 2015-02-11 15:15:10 +1030 | [diff] [blame] | 321 | |
| 322 | /* |
| 323 | * The eip contains the *virtual* address of the Guest's instruction: |
| 324 | * we copy the instruction here so the Launcher doesn't have to walk |
| 325 | * the page tables to decode it. We handle the case (eg. in a kernel |
| 326 | * module) where the instruction is over two pages, and the pages are |
| 327 | * virtually but not physically contiguous. |
| 328 | * |
| 329 | * The longest possible x86 instruction is 15 bytes, but we don't handle |
| 330 | * anything that strange. |
| 331 | */ |
| 332 | static void copy_from_guest(struct lg_cpu *cpu, |
| 333 | void *dst, unsigned long vaddr, size_t len) |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 334 | { |
Rusty Russell | c565650b | 2015-02-11 15:15:10 +1030 | [diff] [blame] | 335 | size_t to_page_end = PAGE_SIZE - (vaddr % PAGE_SIZE); |
| 336 | unsigned long paddr; |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 337 | |
Rusty Russell | c565650b | 2015-02-11 15:15:10 +1030 | [diff] [blame] | 338 | BUG_ON(len > PAGE_SIZE); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 339 | |
Rusty Russell | c565650b | 2015-02-11 15:15:10 +1030 | [diff] [blame] | 340 | /* If it goes over a page, copy in two parts. */ |
| 341 | if (len > to_page_end) { |
| 342 | /* But make sure the next page is mapped! */ |
| 343 | if (__guest_pa(cpu, vaddr + to_page_end, &paddr)) |
| 344 | copy_from_guest(cpu, dst + to_page_end, |
| 345 | vaddr + to_page_end, |
| 346 | len - to_page_end); |
| 347 | else |
| 348 | /* Otherwise fill with zeroes. */ |
| 349 | memset(dst + to_page_end, 0, len - to_page_end); |
| 350 | len = to_page_end; |
Rusty Russell | 5094aea | 2010-04-14 21:43:53 -0600 | [diff] [blame] | 351 | } |
| 352 | |
Rusty Russell | c565650b | 2015-02-11 15:15:10 +1030 | [diff] [blame] | 353 | /* This will kill the guest if it isn't mapped, but that |
| 354 | * shouldn't happen. */ |
| 355 | __lgread(cpu, dst, guest_pa(cpu, vaddr), len); |
| 356 | } |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 357 | |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 358 | |
Rusty Russell | c565650b | 2015-02-11 15:15:10 +1030 | [diff] [blame] | 359 | static void setup_emulate_insn(struct lg_cpu *cpu) |
| 360 | { |
| 361 | cpu->pending.trap = 13; |
| 362 | copy_from_guest(cpu, cpu->pending.insn, cpu->regs->eip, |
| 363 | sizeof(cpu->pending.insn)); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 364 | } |
| 365 | |
Rusty Russell | 7313d52 | 2015-02-11 15:15:10 +1030 | [diff] [blame] | 366 | static void setup_iomem_insn(struct lg_cpu *cpu, unsigned long iomem_addr) |
| 367 | { |
| 368 | cpu->pending.trap = 14; |
| 369 | cpu->pending.addr = iomem_addr; |
| 370 | copy_from_guest(cpu, cpu->pending.insn, cpu->regs->eip, |
| 371 | sizeof(cpu->pending.insn)); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 372 | } |
| 373 | |
| 374 | /*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] | 375 | void lguest_arch_handle_trap(struct lg_cpu *cpu) |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 376 | { |
Rusty Russell | 7313d52 | 2015-02-11 15:15:10 +1030 | [diff] [blame] | 377 | unsigned long iomem_addr; |
| 378 | |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 379 | switch (cpu->regs->trapnum) { |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 380 | case 13: /* We've intercepted a General Protection Fault. */ |
Rusty Russell | c565650b | 2015-02-11 15:15:10 +1030 | [diff] [blame] | 381 | /* Hand to Launcher to emulate those pesky IN and OUT insns */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 382 | if (cpu->regs->errcode == 0) { |
Rusty Russell | c565650b | 2015-02-11 15:15:10 +1030 | [diff] [blame] | 383 | setup_emulate_insn(cpu); |
| 384 | return; |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 385 | } |
| 386 | break; |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 387 | case 14: /* We've intercepted a Page Fault. */ |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 388 | /* |
| 389 | * The Guest accessed a virtual address that wasn't mapped. |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 390 | * This happens a lot: we don't actually set up most of the page |
| 391 | * tables for the Guest at all when we start: as it runs it asks |
| 392 | * for more and more, and we set them up as required. In this |
| 393 | * case, we don't even tell the Guest that the fault happened. |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 394 | * |
| 395 | * The errcode tells whether this was a read or a write, and |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 396 | * whether kernel or userspace code. |
| 397 | */ |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 398 | if (demand_page(cpu, cpu->arch.last_pagefault, |
Rusty Russell | 7313d52 | 2015-02-11 15:15:10 +1030 | [diff] [blame] | 399 | cpu->regs->errcode, &iomem_addr)) |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 400 | return; |
| 401 | |
Rusty Russell | 7313d52 | 2015-02-11 15:15:10 +1030 | [diff] [blame] | 402 | /* Was this an access to memory mapped IO? */ |
| 403 | if (iomem_addr) { |
| 404 | /* Tell Launcher, let it handle it. */ |
| 405 | setup_iomem_insn(cpu, iomem_addr); |
| 406 | return; |
| 407 | } |
| 408 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 409 | /* |
| 410 | * 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] | 411 | * know. We write out the cr2 value so it knows where the |
| 412 | * fault occurred. |
| 413 | * |
| 414 | * Note that if the Guest were really messed up, this could |
| 415 | * happen before it's done the LHCALL_LGUEST_INIT hypercall, so |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 416 | * lg->lguest_data could be NULL |
| 417 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 418 | if (cpu->lg->lguest_data && |
| 419 | put_user(cpu->arch.last_pagefault, |
| 420 | &cpu->lg->lguest_data->cr2)) |
| 421 | kill_guest(cpu, "Writing cr2"); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 422 | break; |
| 423 | case 7: /* We've intercepted a Device Not Available fault. */ |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 424 | /* |
| 425 | * If the Guest doesn't want to know, we already restored the |
| 426 | * Floating Point Unit, so we just continue without telling it. |
| 427 | */ |
Glauber de Oliveira Costa | 4665ac8 | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 428 | if (!cpu->ts) |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 429 | return; |
| 430 | break; |
| 431 | case 32 ... 255: |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 432 | /* |
| 433 | * These values mean a real interrupt occurred, in which case |
Matias Zabaljauregui | 4cd8b5e | 2009-03-14 13:37:52 -0200 | [diff] [blame] | 434 | * the Host handler has already been run. We just do a |
Rusty Russell | cc6d4fb | 2007-10-22 11:03:30 +1000 | [diff] [blame] | 435 | * friendly check if another process should now be run, then |
Rusty Russell | 9f54288 | 2011-07-22 14:39:50 +0930 | [diff] [blame] | 436 | * return to run the Guest again. |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 437 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 438 | cond_resched(); |
Rusty Russell | cc6d4fb | 2007-10-22 11:03:30 +1000 | [diff] [blame] | 439 | return; |
| 440 | case LGUEST_TRAP_ENTRY: |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 441 | /* |
| 442 | * Our 'struct hcall_args' maps directly over our regs: we set |
| 443 | * up the pointer now to indicate a hypercall is pending. |
| 444 | */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 445 | cpu->hcall = (struct hcall_args *)cpu->regs; |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 446 | return; |
| 447 | } |
| 448 | |
| 449 | /* 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] | 450 | if (!deliver_trap(cpu, cpu->regs->trapnum)) |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 451 | /* |
| 452 | * If the Guest doesn't have a handler (either it hasn't |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 453 | * 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] | 454 | * it handle), it dies with this cryptic error message. |
| 455 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 456 | kill_guest(cpu, "unhandled trap %li at %#lx (%#lx)", |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 457 | cpu->regs->trapnum, cpu->regs->eip, |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 458 | cpu->regs->trapnum == 14 ? cpu->arch.last_pagefault |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 459 | : cpu->regs->errcode); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 460 | } |
| 461 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 462 | /* |
| 463 | * 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] | 464 | * complexity: do_hypercalls(), emulate_insn(), maybe_do_interrupt(), |
| 465 | * deliver_trap() and demand_page(). After all those, we'll be ready to |
| 466 | * examine the Switcher, and our philosophical understanding of the Host/Guest |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 467 | * duality will be complete. |
| 468 | :*/ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 469 | static void adjust_pge(void *on) |
| 470 | { |
| 471 | if (on) |
Andy Lutomirski | 375074c | 2014-10-24 15:58:07 -0700 | [diff] [blame] | 472 | cr4_set_bits(X86_CR4_PGE); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 473 | else |
Andy Lutomirski | 375074c | 2014-10-24 15:58:07 -0700 | [diff] [blame] | 474 | cr4_clear_bits(X86_CR4_PGE); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 475 | } |
| 476 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 477 | /*H:020 |
| 478 | * Now the Switcher is mapped and every thing else is ready, we need to do |
| 479 | * some more i386-specific initialization. |
| 480 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 481 | void __init lguest_arch_host_init(void) |
| 482 | { |
| 483 | int i; |
| 484 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 485 | /* |
Rusty Russell | 9f54288 | 2011-07-22 14:39:50 +0930 | [diff] [blame] | 486 | * 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] | 487 | * Intel, jumps are relative, and it doesn't access any references to |
| 488 | * external code or data. |
| 489 | * |
| 490 | * The only exception is the interrupt handlers in switcher.S: their |
| 491 | * addresses are placed in a table (default_idt_entries), so we need to |
| 492 | * update the table with the new addresses. switcher_offset() is a |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 493 | * convenience function which returns the distance between the |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 494 | * compiled-in switcher code and the high-mapped copy we just made. |
| 495 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 496 | for (i = 0; i < IDT_ENTRIES; i++) |
| 497 | default_idt_entries[i] += switcher_offset(); |
| 498 | |
| 499 | /* |
| 500 | * Set up the Switcher's per-cpu areas. |
| 501 | * |
| 502 | * Each CPU gets two pages of its own within the high-mapped region |
| 503 | * (aka. "struct lguest_pages"). Much of this can be initialized now, |
| 504 | * but some depends on what Guest we are running (which is set up in |
| 505 | * copy_in_guest_info()). |
| 506 | */ |
| 507 | for_each_possible_cpu(i) { |
| 508 | /* lguest_pages() returns this CPU's two pages. */ |
| 509 | struct lguest_pages *pages = lguest_pages(i); |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 510 | /* This is a convenience pointer to make the code neater. */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 511 | struct lguest_ro_state *state = &pages->state; |
| 512 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 513 | /* |
| 514 | * The Global Descriptor Table: the Host has a different one |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 515 | * for each CPU. We keep a descriptor for the GDT which says |
| 516 | * 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] | 517 | * byte, not the size, hence the "-1"). |
| 518 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 519 | state->host_gdt_desc.size = GDT_SIZE-1; |
| 520 | state->host_gdt_desc.address = (long)get_cpu_gdt_table(i); |
| 521 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 522 | /* |
| 523 | * All CPUs on the Host use the same Interrupt Descriptor |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 524 | * 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] | 525 | * descriptor. |
| 526 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 527 | store_idt(&state->host_idt_desc); |
| 528 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 529 | /* |
| 530 | * The descriptors for the Guest's GDT and IDT can be filled |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 531 | * out now, too. We copy the GDT & IDT into ->guest_gdt and |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 532 | * ->guest_idt before actually running the Guest. |
| 533 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 534 | state->guest_idt_desc.size = sizeof(state->guest_idt)-1; |
| 535 | state->guest_idt_desc.address = (long)&state->guest_idt; |
| 536 | state->guest_gdt_desc.size = sizeof(state->guest_gdt)-1; |
| 537 | state->guest_gdt_desc.address = (long)&state->guest_gdt; |
| 538 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 539 | /* |
| 540 | * 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] | 541 | * the Switcher: in pages->regs. The stack grows upwards, so |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 542 | * we start it at the end of that structure. |
| 543 | */ |
H. Peter Anvin | faca622 | 2008-01-30 13:31:02 +0100 | [diff] [blame] | 544 | state->guest_tss.sp0 = (long)(&pages->regs + 1); |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 545 | /* |
| 546 | * And this is the GDT entry to use for the stack: we keep a |
| 547 | * couple of special LGUEST entries. |
| 548 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 549 | state->guest_tss.ss0 = LGUEST_DS; |
| 550 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 551 | /* |
| 552 | * x86 can have a finegrained bitmap which indicates what I/O |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 553 | * 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] | 554 | * structure, meaning "none". |
| 555 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 556 | state->guest_tss.io_bitmap_base = sizeof(state->guest_tss); |
| 557 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 558 | /* |
| 559 | * Some GDT entries are the same across all Guests, so we can |
| 560 | * set them up now. |
| 561 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 562 | setup_default_gdt_entries(state); |
| 563 | /* Most IDT entries are the same for all Guests, too.*/ |
| 564 | setup_default_idt_entries(state, default_idt_entries); |
| 565 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 566 | /* |
| 567 | * The Host needs to be able to use the LGUEST segments on this |
| 568 | * CPU, too, so put them in the Host GDT. |
| 569 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 570 | get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT; |
| 571 | get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT; |
| 572 | } |
| 573 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 574 | /* |
| 575 | * In the Switcher, we want the %cs segment register to use the |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 576 | * LGUEST_CS GDT entry: we've put that in the Host and Guest GDTs, so |
| 577 | * 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] | 578 | * need this structure to feed to Intel's "lcall" instruction. |
| 579 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 580 | lguest_entry.offset = (long)switch_to_guest + switcher_offset(); |
| 581 | lguest_entry.segment = LGUEST_CS; |
| 582 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 583 | /* |
| 584 | * Finally, we need to turn off "Page Global Enable". PGE is an |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 585 | * optimization where page table entries are specially marked to show |
| 586 | * they never change. The Host kernel marks all the kernel pages this |
| 587 | * way because it's always present, even when userspace is running. |
| 588 | * |
| 589 | * Lguest breaks this: unbeknownst to the rest of the Host kernel, we |
| 590 | * switch to the Guest kernel. If you don't disable this on all CPUs, |
| 591 | * you'll get really weird bugs that you'll chase for two days. |
| 592 | * |
| 593 | * 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] | 594 | * on when we return, but that slowed the Switcher down noticibly. |
| 595 | */ |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 596 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 597 | /* |
| 598 | * We don't need the complexity of CPUs coming and going while we're |
| 599 | * doing this. |
| 600 | */ |
Gautham R Shenoy | 86ef5c9 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 601 | get_online_cpus(); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 602 | if (cpu_has_pge) { /* We have a broader idea of "global". */ |
| 603 | /* Remember that this was originally set (for cleanup). */ |
| 604 | cpu_had_pge = 1; |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 605 | /* |
| 606 | * adjust_pge is a helper function which sets or unsets the PGE |
| 607 | * bit on its CPU, depending on the argument (0 == unset). |
| 608 | */ |
Jens Axboe | 15c8b6c | 2008-05-09 09:39:44 +0200 | [diff] [blame] | 609 | on_each_cpu(adjust_pge, (void *)0, 1); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 610 | /* Turn off the feature in the global feature set. */ |
Andrew Morton | cf485e5 | 2008-06-09 16:22:48 -0700 | [diff] [blame] | 611 | clear_cpu_cap(&boot_cpu_data, X86_FEATURE_PGE); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 612 | } |
Gautham R Shenoy | 86ef5c9 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 613 | put_online_cpus(); |
Rusty Russell | 9f54288 | 2011-07-22 14:39:50 +0930 | [diff] [blame] | 614 | } |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 615 | /*:*/ |
| 616 | |
| 617 | void __exit lguest_arch_host_fini(void) |
| 618 | { |
| 619 | /* 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] | 620 | get_online_cpus(); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 621 | if (cpu_had_pge) { |
Andrew Morton | cf485e5 | 2008-06-09 16:22:48 -0700 | [diff] [blame] | 622 | set_cpu_cap(&boot_cpu_data, X86_FEATURE_PGE); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 623 | /* adjust_pge's argument "1" means set PGE. */ |
Jens Axboe | 15c8b6c | 2008-05-09 09:39:44 +0200 | [diff] [blame] | 624 | on_each_cpu(adjust_pge, (void *)1, 1); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 625 | } |
Gautham R Shenoy | 86ef5c9 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 626 | put_online_cpus(); |
Jes Sorensen | 625efab | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 627 | } |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 628 | |
| 629 | |
| 630 | /*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] | 631 | 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] | 632 | { |
| 633 | switch (args->arg0) { |
Rusty Russell | a489f0b | 2009-04-19 23:14:00 -0600 | [diff] [blame] | 634 | case LHCALL_LOAD_GDT_ENTRY: |
| 635 | load_guest_gdt_entry(cpu, args->arg1, args->arg2, args->arg3); |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 636 | break; |
| 637 | case LHCALL_LOAD_IDT_ENTRY: |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 638 | load_guest_idt_entry(cpu, args->arg1, args->arg2, args->arg3); |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 639 | break; |
| 640 | case LHCALL_LOAD_TLS: |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 641 | guest_load_tls(cpu, args->arg1); |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 642 | break; |
| 643 | default: |
| 644 | /* Bad Guest. Bad! */ |
| 645 | return -EIO; |
| 646 | } |
| 647 | return 0; |
| 648 | } |
| 649 | |
| 650 | /*H:126 i386-specific hypercall initialization: */ |
Glauber de Oliveira Costa | 73044f0 | 2008-01-07 11:05:27 -0200 | [diff] [blame] | 651 | int lguest_arch_init_hypercalls(struct lg_cpu *cpu) |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 652 | { |
| 653 | u32 tsc_speed; |
| 654 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 655 | /* |
| 656 | * The pointer to the Guest's "struct lguest_data" is the only argument. |
| 657 | * We check that address now. |
| 658 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 659 | if (!lguest_address_ok(cpu->lg, cpu->hcall->arg1, |
| 660 | sizeof(*cpu->lg->lguest_data))) |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 661 | return -EFAULT; |
| 662 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 663 | /* |
| 664 | * Having checked it, we simply set lg->lguest_data to point straight |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 665 | * into the Launcher's memory at the right place and then use |
| 666 | * copy_to_user/from_user from now on, instead of lgread/write. I put |
| 667 | * this in to show that I'm not immune to writing stupid |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 668 | * optimizations. |
| 669 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 670 | cpu->lg->lguest_data = cpu->lg->mem_base + cpu->hcall->arg1; |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 671 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 672 | /* |
| 673 | * 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] | 674 | * cpu frequency. Some devious chip manufacturers decided that TSC |
| 675 | * changes could be handled in software. I decided that time going |
| 676 | * backwards might be good for benchmarks, but it's bad for users. |
| 677 | * |
| 678 | * We also insist that the TSC be stable: the kernel detects unreliable |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 679 | * TSCs for its own purposes, and we use that here. |
| 680 | */ |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 681 | if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC) && !check_tsc_unstable()) |
| 682 | tsc_speed = tsc_khz; |
| 683 | else |
| 684 | tsc_speed = 0; |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 685 | if (put_user(tsc_speed, &cpu->lg->lguest_data->tsc_khz)) |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 686 | return -EFAULT; |
| 687 | |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 688 | /* The interrupt code might not like the system call vector. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 689 | if (!check_syscall_vector(cpu->lg)) |
| 690 | kill_guest(cpu, "bad syscall vector"); |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 691 | |
Jes Sorensen | b410e7b | 2007-10-22 11:03:31 +1000 | [diff] [blame] | 692 | return 0; |
| 693 | } |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 694 | /*:*/ |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 695 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 696 | /*L:030 |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 697 | * 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] | 698 | * allocate the structure, so they will be 0. |
| 699 | */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 700 | void lguest_arch_setup_regs(struct lg_cpu *cpu, unsigned long start) |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 701 | { |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 702 | struct lguest_regs *regs = cpu->regs; |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 703 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 704 | /* |
| 705 | * There are four "segment" registers which the Guest needs to boot: |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 706 | * The "code segment" register (cs) refers to the kernel code segment |
| 707 | * __KERNEL_CS, and the "data", "extra" and "stack" segment registers |
| 708 | * refer to the kernel data segment __KERNEL_DS. |
| 709 | * |
| 710 | * The privilege level is packed into the lower bits. The Guest runs |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 711 | * at privilege level 1 (GUEST_PL). |
| 712 | */ |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 713 | regs->ds = regs->es = regs->ss = __KERNEL_DS|GUEST_PL; |
| 714 | regs->cs = __KERNEL_CS|GUEST_PL; |
| 715 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 716 | /* |
| 717 | * The "eflags" register contains miscellaneous flags. Bit 1 (0x002) |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 718 | * is supposed to always be "1". Bit 9 (0x200) controls whether |
| 719 | * interrupts are enabled. We always leave interrupts enabled while |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 720 | * running the Guest. |
| 721 | */ |
H. Peter Anvin | 1adfa76 | 2013-04-27 16:10:11 -0700 | [diff] [blame] | 722 | regs->eflags = X86_EFLAGS_IF | X86_EFLAGS_FIXED; |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 723 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 724 | /* |
| 725 | * The "Extended Instruction Pointer" register says where the Guest is |
| 726 | * running. |
| 727 | */ |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 728 | regs->eip = start; |
| 729 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 730 | /* |
| 731 | * %esi points to our boot information, at physical address 0, so don't |
| 732 | * touch it. |
| 733 | */ |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 734 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 735 | /* 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] | 736 | setup_guest_gdt(cpu); |
Jes Sorensen | d612cde | 2007-10-22 11:03:32 +1000 | [diff] [blame] | 737 | } |