Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 1 | /*P:800 |
| 2 | * Interrupts (traps) are complicated enough to earn their own file. |
Rusty Russell | f938d2c | 2007-07-26 10:41:02 -0700 | [diff] [blame] | 3 | * There are three classes of interrupts: |
| 4 | * |
| 5 | * 1) Real hardware interrupts which occur while we're running the Guest, |
| 6 | * 2) Interrupts for virtual devices attached to the Guest, and |
| 7 | * 3) Traps and faults from the Guest. |
| 8 | * |
| 9 | * Real hardware interrupts must be delivered to the Host, not the Guest. |
| 10 | * Virtual interrupts must be delivered to the Guest, but we make them look |
| 11 | * just like real hardware would deliver them. Traps from the Guest can be set |
| 12 | * up to go directly back into the Guest, but sometimes the Host wants to see |
| 13 | * them first, so we also have a way of "reflecting" them into the Guest as if |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 14 | * they had been delivered to it directly. |
| 15 | :*/ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 16 | #include <linux/uaccess.h> |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 17 | #include <linux/interrupt.h> |
| 18 | #include <linux/module.h> |
Alexey Dobriyan | d43c36d | 2009-10-07 17:09:06 +0400 | [diff] [blame] | 19 | #include <linux/sched.h> |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 20 | #include "lg.h" |
| 21 | |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 22 | /* Allow Guests to use a non-128 (ie. non-Linux) syscall trap. */ |
Ingo Molnar | f435e68 | 2015-05-11 07:17:04 +0200 | [diff] [blame] | 23 | static unsigned int syscall_vector = IA32_SYSCALL_VECTOR; |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 24 | module_param(syscall_vector, uint, 0444); |
| 25 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 26 | /* The address of the interrupt handler is split into two bits: */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 27 | static unsigned long idt_address(u32 lo, u32 hi) |
| 28 | { |
| 29 | return (lo & 0x0000FFFF) | (hi & 0xFFFF0000); |
| 30 | } |
| 31 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 32 | /* |
| 33 | * The "type" of the interrupt handler is a 4 bit field: we only support a |
| 34 | * couple of types. |
| 35 | */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 36 | static int idt_type(u32 lo, u32 hi) |
| 37 | { |
| 38 | return (hi >> 8) & 0xF; |
| 39 | } |
| 40 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 41 | /* An IDT entry can't be used unless the "present" bit is set. */ |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 42 | static bool idt_present(u32 lo, u32 hi) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 43 | { |
| 44 | return (hi & 0x8000); |
| 45 | } |
| 46 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 47 | /* |
| 48 | * We need a helper to "push" a value onto the Guest's stack, since that's a |
| 49 | * big part of what delivering an interrupt does. |
| 50 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 51 | static void push_guest_stack(struct lg_cpu *cpu, unsigned long *gstack, u32 val) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 52 | { |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 53 | /* Stack grows upwards: move stack then write value. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 54 | *gstack -= 4; |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 55 | lgwrite(cpu, *gstack, u32, val); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 56 | } |
| 57 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 58 | /*H:210 |
Rusty Russell | 3eebd23 | 2015-04-01 13:32:20 +1030 | [diff] [blame] | 59 | * The push_guest_interrupt_stack() routine saves Guest state on the stack for |
| 60 | * an interrupt or trap. The mechanics of delivering traps and interrupts to |
| 61 | * the Guest are the same, except some traps have an "error code" which gets |
| 62 | * pushed onto the stack as well: the caller tells us if this is one. |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 63 | * |
| 64 | * We set up the stack just like the CPU does for a real interrupt, so it's |
| 65 | * identical for the Guest (and the standard "iret" instruction will undo |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 66 | * it). |
| 67 | */ |
Rusty Russell | 3eebd23 | 2015-04-01 13:32:20 +1030 | [diff] [blame] | 68 | static void push_guest_interrupt_stack(struct lg_cpu *cpu, bool has_err) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 69 | { |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 70 | unsigned long gstack, origstack; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 71 | u32 eflags, ss, irq_enable; |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 72 | unsigned long virtstack; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 73 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 74 | /* |
| 75 | * There are two cases for interrupts: one where the Guest is already |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 76 | * in the kernel, and a more complex one where the Guest is in |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 77 | * userspace. We check the privilege level to find out. |
| 78 | */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 79 | if ((cpu->regs->ss&0x3) != GUEST_PL) { |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 80 | /* |
| 81 | * The Guest told us their kernel stack with the SET_STACK |
| 82 | * hypercall: both the virtual address and the segment. |
| 83 | */ |
Glauber de Oliveira Costa | 4665ac8 | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 84 | virtstack = cpu->esp1; |
| 85 | ss = cpu->ss1; |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 86 | |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 87 | origstack = gstack = guest_pa(cpu, virtstack); |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 88 | /* |
| 89 | * We push the old stack segment and pointer onto the new |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 90 | * stack: when the Guest does an "iret" back from the interrupt |
| 91 | * handler the CPU will notice they're dropping privilege |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 92 | * levels and expect these here. |
| 93 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 94 | push_guest_stack(cpu, &gstack, cpu->regs->ss); |
| 95 | push_guest_stack(cpu, &gstack, cpu->regs->esp); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 96 | } else { |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 97 | /* We're staying on the same Guest (kernel) stack. */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 98 | virtstack = cpu->regs->esp; |
| 99 | ss = cpu->regs->ss; |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 100 | |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 101 | origstack = gstack = guest_pa(cpu, virtstack); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 102 | } |
| 103 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 104 | /* |
| 105 | * Remember that we never let the Guest actually disable interrupts, so |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 106 | * the "Interrupt Flag" bit is always set. We copy that bit from the |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 107 | * Guest's "irq_enabled" field into the eflags word: we saw the Guest |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 108 | * copy it back in "lguest_iret". |
| 109 | */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 110 | eflags = cpu->regs->eflags; |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 111 | if (get_user(irq_enable, &cpu->lg->lguest_data->irq_enabled) == 0 |
Rusty Russell | e5faff4 | 2007-07-20 22:11:13 +1000 | [diff] [blame] | 112 | && !(irq_enable & X86_EFLAGS_IF)) |
| 113 | eflags &= ~X86_EFLAGS_IF; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 114 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 115 | /* |
| 116 | * An interrupt is expected to push three things on the stack: the old |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 117 | * "eflags" word, the old code segment, and the old instruction |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 118 | * pointer. |
| 119 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 120 | push_guest_stack(cpu, &gstack, eflags); |
| 121 | push_guest_stack(cpu, &gstack, cpu->regs->cs); |
| 122 | push_guest_stack(cpu, &gstack, cpu->regs->eip); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 123 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 124 | /* For the six traps which supply an error code, we push that, too. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 125 | if (has_err) |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 126 | push_guest_stack(cpu, &gstack, cpu->regs->errcode); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 127 | |
Rusty Russell | 3eebd23 | 2015-04-01 13:32:20 +1030 | [diff] [blame] | 128 | /* Adjust the stack pointer and stack segment. */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 129 | cpu->regs->ss = ss; |
| 130 | cpu->regs->esp = virtstack + (gstack - origstack); |
Rusty Russell | 3eebd23 | 2015-04-01 13:32:20 +1030 | [diff] [blame] | 131 | } |
| 132 | |
| 133 | /* |
| 134 | * This actually makes the Guest start executing the given interrupt/trap |
| 135 | * handler. |
| 136 | * |
| 137 | * "lo" and "hi" are the two parts of the Interrupt Descriptor Table for this |
| 138 | * interrupt or trap. It's split into two parts for traditional reasons: gcc |
| 139 | * on i386 used to be frightened by 64 bit numbers. |
| 140 | */ |
| 141 | static void guest_run_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi) |
| 142 | { |
| 143 | /* If we're already in the kernel, we don't change stacks. */ |
| 144 | if ((cpu->regs->ss&0x3) != GUEST_PL) |
| 145 | cpu->regs->ss = cpu->esp1; |
| 146 | |
| 147 | /* |
| 148 | * Set the code segment and the address to execute. |
| 149 | */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 150 | cpu->regs->cs = (__KERNEL_CS|GUEST_PL); |
| 151 | cpu->regs->eip = idt_address(lo, hi); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 152 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 153 | /* |
Rusty Russell | 98fb4e5 | 2013-09-05 17:45:53 +0930 | [diff] [blame] | 154 | * Trapping always clears these flags: |
| 155 | * TF: Trap flag |
| 156 | * VM: Virtual 8086 mode |
| 157 | * RF: Resume |
| 158 | * NT: Nested task. |
| 159 | */ |
| 160 | cpu->regs->eflags &= |
| 161 | ~(X86_EFLAGS_TF|X86_EFLAGS_VM|X86_EFLAGS_RF|X86_EFLAGS_NT); |
| 162 | |
| 163 | /* |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 164 | * There are two kinds of interrupt handlers: 0xE is an "interrupt |
| 165 | * gate" which expects interrupts to be disabled on entry. |
| 166 | */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 167 | if (idt_type(lo, hi) == 0xE) |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 168 | if (put_user(0, &cpu->lg->lguest_data->irq_enabled)) |
| 169 | kill_guest(cpu, "Disabling interrupts"); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 170 | } |
| 171 | |
Rusty Russell | 3eebd23 | 2015-04-01 13:32:20 +1030 | [diff] [blame] | 172 | /* This restores the eflags word which was pushed on the stack by a trap */ |
| 173 | static void restore_eflags(struct lg_cpu *cpu) |
| 174 | { |
| 175 | /* This is the physical address of the stack. */ |
| 176 | unsigned long stack_pa = guest_pa(cpu, cpu->regs->esp); |
| 177 | |
| 178 | /* |
| 179 | * Stack looks like this: |
| 180 | * Address Contents |
| 181 | * esp EIP |
| 182 | * esp + 4 CS |
| 183 | * esp + 8 EFLAGS |
| 184 | */ |
| 185 | cpu->regs->eflags = lgread(cpu, stack_pa + 8, u32); |
| 186 | cpu->regs->eflags &= |
| 187 | ~(X86_EFLAGS_TF|X86_EFLAGS_VM|X86_EFLAGS_RF|X86_EFLAGS_NT); |
| 188 | } |
| 189 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 190 | /*H:205 |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 191 | * Virtual Interrupts. |
| 192 | * |
Rusty Russell | abd41f0 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 193 | * interrupt_pending() returns the first pending interrupt which isn't blocked |
| 194 | * by the Guest. It is called before every entry to the Guest, and just before |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 195 | * we go to sleep when the Guest has halted itself. |
| 196 | */ |
Rusty Russell | a32a8813 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 197 | unsigned int interrupt_pending(struct lg_cpu *cpu, bool *more) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 198 | { |
| 199 | unsigned int irq; |
| 200 | DECLARE_BITMAP(blk, LGUEST_IRQS); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 201 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 202 | /* If the Guest hasn't even initialized yet, we can do nothing. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 203 | if (!cpu->lg->lguest_data) |
Rusty Russell | abd41f0 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 204 | return LGUEST_IRQS; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 205 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 206 | /* |
| 207 | * Take our "irqs_pending" array and remove any interrupts the Guest |
| 208 | * wants blocked: the result ends up in "blk". |
| 209 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 210 | if (copy_from_user(&blk, cpu->lg->lguest_data->blocked_interrupts, |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 211 | sizeof(blk))) |
Rusty Russell | abd41f0 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 212 | return LGUEST_IRQS; |
Glauber de Oliveira Costa | 177e449 | 2008-01-07 11:05:29 -0200 | [diff] [blame] | 213 | bitmap_andnot(blk, cpu->irqs_pending, blk, LGUEST_IRQS); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 214 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 215 | /* Find the first interrupt. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 216 | irq = find_first_bit(blk, LGUEST_IRQS); |
Rusty Russell | a32a8813 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 217 | *more = find_next_bit(blk, LGUEST_IRQS, irq+1); |
Rusty Russell | abd41f0 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 218 | |
| 219 | return irq; |
| 220 | } |
| 221 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 222 | /* |
| 223 | * This actually diverts the Guest to running an interrupt handler, once an |
| 224 | * interrupt has been identified by interrupt_pending(). |
| 225 | */ |
Rusty Russell | a32a8813 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 226 | void try_deliver_interrupt(struct lg_cpu *cpu, unsigned int irq, bool more) |
Rusty Russell | abd41f0 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 227 | { |
| 228 | struct desc_struct *idt; |
| 229 | |
| 230 | BUG_ON(irq >= LGUEST_IRQS); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 231 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 232 | /* If they're halted, interrupts restart them. */ |
Glauber de Oliveira Costa | 66686c2 | 2008-01-07 11:05:34 -0200 | [diff] [blame] | 233 | if (cpu->halted) { |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 234 | /* Re-enable interrupts. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 235 | if (put_user(X86_EFLAGS_IF, &cpu->lg->lguest_data->irq_enabled)) |
| 236 | kill_guest(cpu, "Re-enabling interrupts"); |
Glauber de Oliveira Costa | 66686c2 | 2008-01-07 11:05:34 -0200 | [diff] [blame] | 237 | cpu->halted = 0; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 238 | } else { |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 239 | /* Otherwise we check if they have interrupts disabled. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 240 | u32 irq_enabled; |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 241 | if (get_user(irq_enabled, &cpu->lg->lguest_data->irq_enabled)) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 242 | irq_enabled = 0; |
Rusty Russell | a32a8813 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 243 | if (!irq_enabled) { |
| 244 | /* Make sure they know an IRQ is pending. */ |
| 245 | put_user(X86_EFLAGS_IF, |
| 246 | &cpu->lg->lguest_data->irq_pending); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 247 | return; |
Rusty Russell | a32a8813 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 248 | } |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 249 | } |
| 250 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 251 | /* |
| 252 | * Look at the IDT entry the Guest gave us for this interrupt. The |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 253 | * first 32 (FIRST_EXTERNAL_VECTOR) entries are for traps, so we skip |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 254 | * over them. |
| 255 | */ |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 256 | idt = &cpu->arch.idt[FIRST_EXTERNAL_VECTOR+irq]; |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 257 | /* If they don't have a handler (yet?), we just ignore it */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 258 | if (idt_present(idt->a, idt->b)) { |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 259 | /* OK, mark it no longer pending and deliver it. */ |
Glauber de Oliveira Costa | 177e449 | 2008-01-07 11:05:29 -0200 | [diff] [blame] | 260 | clear_bit(irq, cpu->irqs_pending); |
Rusty Russell | 3eebd23 | 2015-04-01 13:32:20 +1030 | [diff] [blame] | 261 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 262 | /* |
Rusty Russell | 3eebd23 | 2015-04-01 13:32:20 +1030 | [diff] [blame] | 263 | * They may be about to iret, where they asked us never to |
| 264 | * deliver interrupts. In this case, we can emulate that iret |
| 265 | * then immediately deliver the interrupt. This is basically |
| 266 | * a noop: the iret would pop the interrupt frame and restore |
| 267 | * eflags, and then we'd set it up again. So just restore the |
| 268 | * eflags word and jump straight to the handler in this case. |
| 269 | * |
| 270 | * Denys Vlasenko points out that this isn't quite right: if |
| 271 | * the iret was returning to userspace, then that interrupt |
| 272 | * would reset the stack pointer (which the Guest told us |
| 273 | * about via LHCALL_SET_STACK). But unless the Guest is being |
| 274 | * *really* weird, that will be the same as the current stack |
| 275 | * anyway. |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 276 | */ |
Rusty Russell | 3eebd23 | 2015-04-01 13:32:20 +1030 | [diff] [blame] | 277 | if (cpu->regs->eip == cpu->lg->noirq_iret) { |
| 278 | restore_eflags(cpu); |
| 279 | } else { |
| 280 | /* |
| 281 | * set_guest_interrupt() takes a flag to say whether |
| 282 | * this interrupt pushes an error code onto the stack |
| 283 | * as well: virtual interrupts never do. |
| 284 | */ |
| 285 | push_guest_interrupt_stack(cpu, false); |
| 286 | } |
| 287 | /* Actually make Guest cpu jump to handler. */ |
| 288 | guest_run_interrupt(cpu, idt->a, idt->b); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 289 | } |
Rusty Russell | 6c8dca5 | 2007-07-27 13:42:52 +1000 | [diff] [blame] | 290 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 291 | /* |
| 292 | * Every time we deliver an interrupt, we update the timestamp in the |
Rusty Russell | 6c8dca5 | 2007-07-27 13:42:52 +1000 | [diff] [blame] | 293 | * Guest's lguest_data struct. It would be better for the Guest if we |
| 294 | * did this more often, but it can actually be quite slow: doing it |
| 295 | * here is a compromise which means at least it gets updated every |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 296 | * timer interrupt. |
| 297 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 298 | write_timestamp(cpu); |
Rusty Russell | a32a8813 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 299 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 300 | /* |
| 301 | * If there are no other interrupts we want to deliver, clear |
| 302 | * the pending flag. |
| 303 | */ |
Rusty Russell | a32a8813 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 304 | if (!more) |
| 305 | put_user(0, &cpu->lg->lguest_data->irq_pending); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 306 | } |
Rusty Russell | 9f155a9 | 2009-06-12 22:27:08 -0600 | [diff] [blame] | 307 | |
| 308 | /* And this is the routine when we want to set an interrupt for the Guest. */ |
| 309 | void set_interrupt(struct lg_cpu *cpu, unsigned int irq) |
| 310 | { |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 311 | /* |
| 312 | * Next time the Guest runs, the core code will see if it can deliver |
| 313 | * this interrupt. |
| 314 | */ |
Rusty Russell | 9f155a9 | 2009-06-12 22:27:08 -0600 | [diff] [blame] | 315 | set_bit(irq, cpu->irqs_pending); |
| 316 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 317 | /* |
| 318 | * Make sure it sees it; it might be asleep (eg. halted), or running |
| 319 | * the Guest right now, in which case kick_process() will knock it out. |
| 320 | */ |
Rusty Russell | 9f155a9 | 2009-06-12 22:27:08 -0600 | [diff] [blame] | 321 | if (!wake_up_process(cpu->tsk)) |
| 322 | kick_process(cpu->tsk); |
| 323 | } |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 324 | /*:*/ |
| 325 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 326 | /* |
| 327 | * Linux uses trap 128 for system calls. Plan9 uses 64, and Ron Minnich sent |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 328 | * me a patch, so we support that too. It'd be a big step for lguest if half |
| 329 | * the Plan 9 user base were to start using it. |
| 330 | * |
| 331 | * Actually now I think of it, it's possible that Ron *is* half the Plan 9 |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 332 | * userbase. Oh well. |
| 333 | */ |
Rusty Russell | f87e043 | 2016-04-01 12:15:46 +1030 | [diff] [blame] | 334 | bool could_be_syscall(unsigned int num) |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 335 | { |
Ingo Molnar | f435e68 | 2015-05-11 07:17:04 +0200 | [diff] [blame] | 336 | /* Normal Linux IA32_SYSCALL_VECTOR or reserved vector? */ |
| 337 | return num == IA32_SYSCALL_VECTOR || num == syscall_vector; |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 338 | } |
| 339 | |
| 340 | /* The syscall vector it wants must be unused by Host. */ |
| 341 | bool check_syscall_vector(struct lguest *lg) |
| 342 | { |
| 343 | u32 vector; |
| 344 | |
| 345 | if (get_user(vector, &lg->lguest_data->syscall_vec)) |
| 346 | return false; |
| 347 | |
| 348 | return could_be_syscall(vector); |
| 349 | } |
| 350 | |
| 351 | int init_interrupts(void) |
| 352 | { |
| 353 | /* If they want some strange system call vector, reserve it now */ |
Ingo Molnar | f435e68 | 2015-05-11 07:17:04 +0200 | [diff] [blame] | 354 | if (syscall_vector != IA32_SYSCALL_VECTOR) { |
Yinghai Lu | b77b881 | 2008-12-19 15:23:44 -0800 | [diff] [blame] | 355 | if (test_bit(syscall_vector, used_vectors) || |
| 356 | vector_used_by_percpu_irq(syscall_vector)) { |
| 357 | printk(KERN_ERR "lg: couldn't reserve syscall %u\n", |
| 358 | syscall_vector); |
| 359 | return -EBUSY; |
| 360 | } |
| 361 | set_bit(syscall_vector, used_vectors); |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 362 | } |
Yinghai Lu | b77b881 | 2008-12-19 15:23:44 -0800 | [diff] [blame] | 363 | |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 364 | return 0; |
| 365 | } |
| 366 | |
| 367 | void free_interrupts(void) |
| 368 | { |
Ingo Molnar | f435e68 | 2015-05-11 07:17:04 +0200 | [diff] [blame] | 369 | if (syscall_vector != IA32_SYSCALL_VECTOR) |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 370 | clear_bit(syscall_vector, used_vectors); |
| 371 | } |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 372 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 373 | /*H:220 |
| 374 | * Now we've got the routines to deliver interrupts, delivering traps like |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 375 | * page fault is easy. The only trick is that Intel decided that some traps |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 376 | * should have error codes: |
| 377 | */ |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 378 | static bool has_err(unsigned int trap) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 379 | { |
| 380 | return (trap == 8 || (trap >= 10 && trap <= 14) || trap == 17); |
| 381 | } |
| 382 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 383 | /* deliver_trap() returns true if it could deliver the trap. */ |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 384 | bool deliver_trap(struct lg_cpu *cpu, unsigned int num) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 385 | { |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 386 | /* |
| 387 | * Trap numbers are always 8 bit, but we set an impossible trap number |
| 388 | * for traps inside the Switcher, so check that here. |
| 389 | */ |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 390 | if (num >= ARRAY_SIZE(cpu->arch.idt)) |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 391 | return false; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 392 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 393 | /* |
| 394 | * Early on the Guest hasn't set the IDT entries (or maybe it put a |
| 395 | * bogus one in): if we fail here, the Guest will be killed. |
| 396 | */ |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 397 | if (!idt_present(cpu->arch.idt[num].a, cpu->arch.idt[num].b)) |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 398 | return false; |
Rusty Russell | 3eebd23 | 2015-04-01 13:32:20 +1030 | [diff] [blame] | 399 | push_guest_interrupt_stack(cpu, has_err(num)); |
| 400 | guest_run_interrupt(cpu, cpu->arch.idt[num].a, |
| 401 | cpu->arch.idt[num].b); |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 402 | return true; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 403 | } |
| 404 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 405 | /*H:250 |
| 406 | * Here's the hard part: returning to the Host every time a trap happens |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 407 | * and then calling deliver_trap() and re-entering the Guest is slow. |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 408 | * Particularly because Guest userspace system calls are traps (usually trap |
| 409 | * 128). |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 410 | * |
| 411 | * So we'd like to set up the IDT to tell the CPU to deliver traps directly |
| 412 | * into the Guest. This is possible, but the complexities cause the size of |
| 413 | * this file to double! However, 150 lines of code is worth writing for taking |
| 414 | * system calls down from 1750ns to 270ns. Plus, if lguest didn't do it, all |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 415 | * the other hypervisors would beat it up at lunchtime. |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 416 | * |
Rusty Russell | 56adbe9 | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 417 | * This routine indicates if a particular trap number could be delivered |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 418 | * directly. |
Rusty Russell | f87e043 | 2016-04-01 12:15:46 +1030 | [diff] [blame] | 419 | * |
| 420 | * Unfortunately, Linux 4.6 started using an interrupt gate instead of a |
| 421 | * trap gate for syscalls, so this trick is ineffective. See Mastery for |
| 422 | * how we could do this anyway... |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 423 | */ |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 424 | static bool direct_trap(unsigned int num) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 425 | { |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 426 | /* |
| 427 | * Hardware interrupts don't go to the Guest at all (except system |
| 428 | * call). |
| 429 | */ |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 430 | if (num >= FIRST_EXTERNAL_VECTOR && !could_be_syscall(num)) |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 431 | return false; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 432 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 433 | /* |
| 434 | * The Host needs to see page faults (for shadow paging and to save the |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 435 | * fault address), general protection faults (in/out emulation) and |
Rusty Russell | 6d7a5d1 | 2011-07-22 14:39:49 +0930 | [diff] [blame] | 436 | * device not available (TS handling) and of course, the hypercall trap. |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 437 | */ |
Rusty Russell | 6d7a5d1 | 2011-07-22 14:39:49 +0930 | [diff] [blame] | 438 | return num != 14 && num != 13 && num != 7 && num != LGUEST_TRAP_ENTRY; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 439 | } |
Rusty Russell | f56a384 | 2007-07-26 10:41:05 -0700 | [diff] [blame] | 440 | /*:*/ |
| 441 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 442 | /*M:005 |
| 443 | * The Guest has the ability to turn its interrupt gates into trap gates, |
Rusty Russell | f56a384 | 2007-07-26 10:41:05 -0700 | [diff] [blame] | 444 | * if it is careful. The Host will let trap gates can go directly to the |
| 445 | * Guest, but the Guest needs the interrupts atomically disabled for an |
Rusty Russell | 2f921b5 | 2015-03-24 11:51:39 +1030 | [diff] [blame] | 446 | * interrupt gate. The Host could provide a mechanism to register more |
| 447 | * "no-interrupt" regions, and the Guest could point the trap gate at |
| 448 | * instructions within that region, where it can safely disable interrupts. |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 449 | */ |
Rusty Russell | f56a384 | 2007-07-26 10:41:05 -0700 | [diff] [blame] | 450 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 451 | /*M:006 |
| 452 | * The Guests do not use the sysenter (fast system call) instruction, |
Rusty Russell | f56a384 | 2007-07-26 10:41:05 -0700 | [diff] [blame] | 453 | * because it's hardcoded to enter privilege level 0 and so can't go direct. |
| 454 | * It's about twice as fast as the older "int 0x80" system call, so it might |
| 455 | * still be worthwhile to handle it in the Switcher and lcall down to the |
| 456 | * Guest. The sysenter semantics are hairy tho: search for that keyword in |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 457 | * entry.S |
| 458 | :*/ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 459 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 460 | /*H:260 |
| 461 | * When we make traps go directly into the Guest, we need to make sure |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 462 | * the kernel stack is valid (ie. mapped in the page tables). Otherwise, the |
| 463 | * CPU trying to deliver the trap will fault while trying to push the interrupt |
| 464 | * words on the stack: this is called a double fault, and it forces us to kill |
| 465 | * the Guest. |
| 466 | * |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 467 | * Which is deeply unfair, because (literally!) it wasn't the Guests' fault. |
| 468 | */ |
Glauber de Oliveira Costa | 4665ac8 | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 469 | void pin_stack_pages(struct lg_cpu *cpu) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 470 | { |
| 471 | unsigned int i; |
| 472 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 473 | /* |
| 474 | * Depending on the CONFIG_4KSTACKS option, the Guest can have one or |
| 475 | * two pages of stack space. |
| 476 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 477 | for (i = 0; i < cpu->lg->stack_pages; i++) |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 478 | /* |
| 479 | * The stack grows *upwards*, so the address we're given is the |
Rusty Russell | 8057d76 | 2007-08-30 06:35:08 +1000 | [diff] [blame] | 480 | * start of the page after the kernel stack. Subtract one to |
| 481 | * get back onto the first stack page, and keep subtracting to |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 482 | * get to the rest of the stack pages. |
| 483 | */ |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 484 | pin_page(cpu, cpu->esp1 - 1 - i * PAGE_SIZE); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 485 | } |
| 486 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 487 | /* |
| 488 | * Direct traps also mean that we need to know whenever the Guest wants to use |
Rusty Russell | 9f54288 | 2011-07-22 14:39:50 +0930 | [diff] [blame] | 489 | * a different kernel stack, so we can change the guest TSS to use that |
| 490 | * stack. The TSS entries expect a virtual address, so unlike most addresses |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 491 | * the Guest gives us, the "esp" (stack pointer) value here is virtual, not |
| 492 | * physical. |
| 493 | * |
| 494 | * In Linux each process has its own kernel stack, so this happens a lot: we |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 495 | * change stacks on each context switch. |
| 496 | */ |
Glauber de Oliveira Costa | 4665ac8 | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 497 | void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 498 | { |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 499 | /* |
| 500 | * You're not allowed a stack segment with privilege level 0: bad Guest! |
| 501 | */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 502 | if ((seg & 0x3) != GUEST_PL) |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 503 | kill_guest(cpu, "bad stack segment %i", seg); |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 504 | /* We only expect one or two stack pages. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 505 | if (pages > 2) |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 506 | kill_guest(cpu, "bad stack pages %u", pages); |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 507 | /* Save where the stack is, and how many pages */ |
Glauber de Oliveira Costa | 4665ac8 | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 508 | cpu->ss1 = seg; |
| 509 | cpu->esp1 = esp; |
| 510 | cpu->lg->stack_pages = pages; |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 511 | /* Make sure the new stack pages are mapped */ |
Glauber de Oliveira Costa | 4665ac8 | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 512 | pin_stack_pages(cpu); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 513 | } |
| 514 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 515 | /* |
| 516 | * All this reference to mapping stacks leads us neatly into the other complex |
| 517 | * part of the Host: page table handling. |
| 518 | */ |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 519 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 520 | /*H:235 |
| 521 | * This is the routine which actually checks the Guest's IDT entry and |
| 522 | * transfers it into the entry in "struct lguest": |
| 523 | */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 524 | static void set_trap(struct lg_cpu *cpu, struct desc_struct *trap, |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 525 | unsigned int num, u32 lo, u32 hi) |
| 526 | { |
| 527 | u8 type = idt_type(lo, hi); |
| 528 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 529 | /* We zero-out a not-present entry */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 530 | if (!idt_present(lo, hi)) { |
| 531 | trap->a = trap->b = 0; |
| 532 | return; |
| 533 | } |
| 534 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 535 | /* We only support interrupt and trap gates. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 536 | if (type != 0xE && type != 0xF) |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 537 | kill_guest(cpu, "bad IDT type %i", type); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 538 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 539 | /* |
| 540 | * We only copy the handler address, present bit, privilege level and |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 541 | * type. The privilege level controls where the trap can be triggered |
| 542 | * manually with an "int" instruction. This is usually GUEST_PL, |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 543 | * except for system calls which userspace can use. |
| 544 | */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 545 | trap->a = ((__KERNEL_CS|GUEST_PL)<<16) | (lo&0x0000FFFF); |
| 546 | trap->b = (hi&0xFFFFEF00); |
| 547 | } |
| 548 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 549 | /*H:230 |
| 550 | * While we're here, dealing with delivering traps and interrupts to the |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 551 | * Guest, we might as well complete the picture: how the Guest tells us where |
| 552 | * it wants them to go. This would be simple, except making traps fast |
| 553 | * requires some tricks. |
| 554 | * |
| 555 | * We saw the Guest setting Interrupt Descriptor Table (IDT) entries with the |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 556 | * LHCALL_LOAD_IDT_ENTRY hypercall before: that comes here. |
| 557 | */ |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 558 | void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int num, u32 lo, u32 hi) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 559 | { |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 560 | /* |
| 561 | * Guest never handles: NMI, doublefault, spurious interrupt or |
| 562 | * hypercall. We ignore when it tries to set them. |
| 563 | */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 564 | if (num == 2 || num == 8 || num == 15 || num == LGUEST_TRAP_ENTRY) |
| 565 | return; |
| 566 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 567 | /* |
| 568 | * Mark the IDT as changed: next time the Guest runs we'll know we have |
| 569 | * to copy this again. |
| 570 | */ |
Glauber de Oliveira Costa | ae3749d | 2008-01-17 19:14:46 -0200 | [diff] [blame] | 571 | cpu->changed |= CHANGED_IDT; |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 572 | |
Rusty Russell | 56adbe9 | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 573 | /* Check that the Guest doesn't try to step outside the bounds. */ |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 574 | if (num >= ARRAY_SIZE(cpu->arch.idt)) |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 575 | kill_guest(cpu, "Setting idt entry %u", num); |
Rusty Russell | 56adbe9 | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 576 | else |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 577 | set_trap(cpu, &cpu->arch.idt[num], num, lo, hi); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 578 | } |
| 579 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 580 | /* |
| 581 | * The default entry for each interrupt points into the Switcher routines which |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 582 | * simply return to the Host. The run_guest() loop will then call |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 583 | * deliver_trap() to bounce it back into the Guest. |
| 584 | */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 585 | static void default_idt_entry(struct desc_struct *idt, |
| 586 | int trap, |
Rusty Russell | 0c12091 | 2008-07-29 09:58:31 -0500 | [diff] [blame] | 587 | const unsigned long handler, |
| 588 | const struct desc_struct *base) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 589 | { |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 590 | /* A present interrupt gate. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 591 | u32 flags = 0x8e00; |
| 592 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 593 | /* |
| 594 | * Set the privilege level on the entry for the hypercall: this allows |
| 595 | * the Guest to use the "int" instruction to trigger it. |
| 596 | */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 597 | if (trap == LGUEST_TRAP_ENTRY) |
| 598 | flags |= (GUEST_PL << 13); |
Rusty Russell | 0c12091 | 2008-07-29 09:58:31 -0500 | [diff] [blame] | 599 | else if (base) |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 600 | /* |
| 601 | * Copy privilege level from what Guest asked for. This allows |
| 602 | * debug (int 3) traps from Guest userspace, for example. |
| 603 | */ |
Rusty Russell | 0c12091 | 2008-07-29 09:58:31 -0500 | [diff] [blame] | 604 | flags |= (base->b & 0x6000); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 605 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 606 | /* Now pack it into the IDT entry in its weird format. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 607 | idt->a = (LGUEST_CS<<16) | (handler&0x0000FFFF); |
| 608 | idt->b = (handler&0xFFFF0000) | flags; |
| 609 | } |
| 610 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 611 | /* When the Guest first starts, we put default entries into the IDT. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 612 | void setup_default_idt_entries(struct lguest_ro_state *state, |
| 613 | const unsigned long *def) |
| 614 | { |
| 615 | unsigned int i; |
| 616 | |
| 617 | for (i = 0; i < ARRAY_SIZE(state->guest_idt); i++) |
Rusty Russell | 0c12091 | 2008-07-29 09:58:31 -0500 | [diff] [blame] | 618 | default_idt_entry(&state->guest_idt[i], i, def[i], NULL); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 619 | } |
| 620 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 621 | /*H:240 |
| 622 | * We don't use the IDT entries in the "struct lguest" directly, instead |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 623 | * we copy them into the IDT which we've set up for Guests on this CPU, just |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 624 | * before we run the Guest. This routine does that copy. |
| 625 | */ |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 626 | void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt, |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 627 | const unsigned long *def) |
| 628 | { |
| 629 | unsigned int i; |
| 630 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 631 | /* |
| 632 | * We can simply copy the direct traps, otherwise we use the default |
| 633 | * ones in the Switcher: they will return to the Host. |
| 634 | */ |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 635 | for (i = 0; i < ARRAY_SIZE(cpu->arch.idt); i++) { |
Rusty Russell | 0c12091 | 2008-07-29 09:58:31 -0500 | [diff] [blame] | 636 | const struct desc_struct *gidt = &cpu->arch.idt[i]; |
| 637 | |
Rusty Russell | 56adbe9 | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 638 | /* If no Guest can ever override this trap, leave it alone. */ |
| 639 | if (!direct_trap(i)) |
| 640 | continue; |
| 641 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 642 | /* |
| 643 | * Only trap gates (type 15) can go direct to the Guest. |
Rusty Russell | 56adbe9 | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 644 | * Interrupt gates (type 14) disable interrupts as they are |
| 645 | * entered, which we never let the Guest do. Not present |
Rusty Russell | 0c12091 | 2008-07-29 09:58:31 -0500 | [diff] [blame] | 646 | * entries (type 0x0) also can't go direct, of course. |
| 647 | * |
| 648 | * If it can't go direct, we still need to copy the priv. level: |
| 649 | * they might want to give userspace access to a software |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 650 | * interrupt. |
| 651 | */ |
Rusty Russell | 0c12091 | 2008-07-29 09:58:31 -0500 | [diff] [blame] | 652 | if (idt_type(gidt->a, gidt->b) == 0xF) |
| 653 | idt[i] = *gidt; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 654 | else |
Rusty Russell | 0c12091 | 2008-07-29 09:58:31 -0500 | [diff] [blame] | 655 | default_idt_entry(&idt[i], i, def[i], gidt); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 656 | } |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 657 | } |
| 658 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 659 | /*H:200 |
| 660 | * The Guest Clock. |
| 661 | * |
| 662 | * There are two sources of virtual interrupts. We saw one in lguest_user.c: |
| 663 | * the Launcher sending interrupts for virtual devices. The other is the Guest |
| 664 | * timer interrupt. |
| 665 | * |
| 666 | * The Guest uses the LHCALL_SET_CLOCKEVENT hypercall to tell us how long to |
| 667 | * the next timer interrupt (in nanoseconds). We use the high-resolution timer |
| 668 | * infrastructure to set a callback at that time. |
| 669 | * |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 670 | * 0 means "turn off the clock". |
| 671 | */ |
Glauber de Oliveira Costa | ad8d8f3 | 2008-01-07 11:05:28 -0200 | [diff] [blame] | 672 | void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 673 | { |
| 674 | ktime_t expires; |
| 675 | |
| 676 | if (unlikely(delta == 0)) { |
| 677 | /* Clock event device is shutting down. */ |
Glauber de Oliveira Costa | ad8d8f3 | 2008-01-07 11:05:28 -0200 | [diff] [blame] | 678 | hrtimer_cancel(&cpu->hrt); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 679 | return; |
| 680 | } |
| 681 | |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 682 | /* |
| 683 | * We use wallclock time here, so the Guest might not be running for |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 684 | * all the time between now and the timer interrupt it asked for. This |
Rusty Russell | 2e04ef7 | 2009-07-30 16:03:45 -0600 | [diff] [blame] | 685 | * is almost always the right thing to do. |
| 686 | */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 687 | expires = ktime_add_ns(ktime_get_real(), delta); |
Glauber de Oliveira Costa | ad8d8f3 | 2008-01-07 11:05:28 -0200 | [diff] [blame] | 688 | hrtimer_start(&cpu->hrt, expires, HRTIMER_MODE_ABS); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 689 | } |
| 690 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 691 | /* This is the function called when the Guest's timer expires. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 692 | static enum hrtimer_restart clockdev_fn(struct hrtimer *timer) |
| 693 | { |
Glauber de Oliveira Costa | ad8d8f3 | 2008-01-07 11:05:28 -0200 | [diff] [blame] | 694 | struct lg_cpu *cpu = container_of(timer, struct lg_cpu, hrt); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 695 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 696 | /* Remember the first interrupt is the timer interrupt. */ |
Rusty Russell | 9f155a9 | 2009-06-12 22:27:08 -0600 | [diff] [blame] | 697 | set_interrupt(cpu, 0); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 698 | return HRTIMER_NORESTART; |
| 699 | } |
| 700 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 701 | /* This sets up the timer for this Guest. */ |
Glauber de Oliveira Costa | ad8d8f3 | 2008-01-07 11:05:28 -0200 | [diff] [blame] | 702 | void init_clockdev(struct lg_cpu *cpu) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 703 | { |
Glauber de Oliveira Costa | ad8d8f3 | 2008-01-07 11:05:28 -0200 | [diff] [blame] | 704 | hrtimer_init(&cpu->hrt, CLOCK_REALTIME, HRTIMER_MODE_ABS); |
| 705 | cpu->hrt.function = clockdev_fn; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 706 | } |