Rusty Russell | f938d2c | 2007-07-26 10:41:02 -0700 | [diff] [blame] | 1 | /*P:800 Interrupts (traps) are complicated enough to earn their own file. |
| 2 | * There are three classes of interrupts: |
| 3 | * |
| 4 | * 1) Real hardware interrupts which occur while we're running the Guest, |
| 5 | * 2) Interrupts for virtual devices attached to the Guest, and |
| 6 | * 3) Traps and faults from the Guest. |
| 7 | * |
| 8 | * Real hardware interrupts must be delivered to the Host, not the Guest. |
| 9 | * Virtual interrupts must be delivered to the Guest, but we make them look |
| 10 | * just like real hardware would deliver them. Traps from the Guest can be set |
| 11 | * up to go directly back into the Guest, but sometimes the Host wants to see |
| 12 | * them first, so we also have a way of "reflecting" them into the Guest as if |
| 13 | * they had been delivered to it directly. :*/ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 14 | #include <linux/uaccess.h> |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 15 | #include <linux/interrupt.h> |
| 16 | #include <linux/module.h> |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 17 | #include "lg.h" |
| 18 | |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 19 | /* Allow Guests to use a non-128 (ie. non-Linux) syscall trap. */ |
| 20 | static unsigned int syscall_vector = SYSCALL_VECTOR; |
| 21 | module_param(syscall_vector, uint, 0444); |
| 22 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 23 | /* The address of the interrupt handler is split into two bits: */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 24 | static unsigned long idt_address(u32 lo, u32 hi) |
| 25 | { |
| 26 | return (lo & 0x0000FFFF) | (hi & 0xFFFF0000); |
| 27 | } |
| 28 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 29 | /* The "type" of the interrupt handler is a 4 bit field: we only support a |
| 30 | * couple of types. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 31 | static int idt_type(u32 lo, u32 hi) |
| 32 | { |
| 33 | return (hi >> 8) & 0xF; |
| 34 | } |
| 35 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 36 | /* 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] | 37 | static bool idt_present(u32 lo, u32 hi) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 38 | { |
| 39 | return (hi & 0x8000); |
| 40 | } |
| 41 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 42 | /* We need a helper to "push" a value onto the Guest's stack, since that's a |
| 43 | * big part of what delivering an interrupt does. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 44 | 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] | 45 | { |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 46 | /* Stack grows upwards: move stack then write value. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 47 | *gstack -= 4; |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 48 | lgwrite(cpu, *gstack, u32, val); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 49 | } |
| 50 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 51 | /*H:210 The set_guest_interrupt() routine actually delivers the interrupt or |
| 52 | * trap. The mechanics of delivering traps and interrupts to the Guest are the |
| 53 | * same, except some traps have an "error code" which gets pushed onto the |
| 54 | * stack as well: the caller tells us if this is one. |
| 55 | * |
| 56 | * "lo" and "hi" are the two parts of the Interrupt Descriptor Table for this |
| 57 | * interrupt or trap. It's split into two parts for traditional reasons: gcc |
| 58 | * on i386 used to be frightened by 64 bit numbers. |
| 59 | * |
| 60 | * We set up the stack just like the CPU does for a real interrupt, so it's |
| 61 | * identical for the Guest (and the standard "iret" instruction will undo |
| 62 | * it). */ |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 63 | static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi, |
| 64 | bool has_err) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 65 | { |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 66 | unsigned long gstack, origstack; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 67 | u32 eflags, ss, irq_enable; |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 68 | unsigned long virtstack; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 69 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 70 | /* There are two cases for interrupts: one where the Guest is already |
| 71 | * in the kernel, and a more complex one where the Guest is in |
| 72 | * userspace. We check the privilege level to find out. */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 73 | if ((cpu->regs->ss&0x3) != GUEST_PL) { |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 74 | /* The Guest told us their kernel stack with the SET_STACK |
| 75 | * hypercall: both the virtual address and the segment */ |
Glauber de Oliveira Costa | 4665ac8e | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 76 | virtstack = cpu->esp1; |
| 77 | ss = cpu->ss1; |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 78 | |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 79 | origstack = gstack = guest_pa(cpu, virtstack); |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 80 | /* We push the old stack segment and pointer onto the new |
| 81 | * stack: when the Guest does an "iret" back from the interrupt |
| 82 | * handler the CPU will notice they're dropping privilege |
| 83 | * levels and expect these here. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 84 | push_guest_stack(cpu, &gstack, cpu->regs->ss); |
| 85 | push_guest_stack(cpu, &gstack, cpu->regs->esp); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 86 | } else { |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 87 | /* We're staying on the same Guest (kernel) stack. */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 88 | virtstack = cpu->regs->esp; |
| 89 | ss = cpu->regs->ss; |
Rusty Russell | 47436aa | 2007-10-22 11:03:36 +1000 | [diff] [blame] | 90 | |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 91 | origstack = gstack = guest_pa(cpu, virtstack); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 92 | } |
| 93 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 94 | /* Remember that we never let the Guest actually disable interrupts, so |
| 95 | * 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] | 96 | * Guest's "irq_enabled" field into the eflags word: we saw the Guest |
| 97 | * copy it back in "lguest_iret". */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 98 | eflags = cpu->regs->eflags; |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 99 | if (get_user(irq_enable, &cpu->lg->lguest_data->irq_enabled) == 0 |
Rusty Russell | e5faff4 | 2007-07-20 22:11:13 +1000 | [diff] [blame] | 100 | && !(irq_enable & X86_EFLAGS_IF)) |
| 101 | eflags &= ~X86_EFLAGS_IF; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 102 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 103 | /* An interrupt is expected to push three things on the stack: the old |
| 104 | * "eflags" word, the old code segment, and the old instruction |
| 105 | * pointer. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 106 | push_guest_stack(cpu, &gstack, eflags); |
| 107 | push_guest_stack(cpu, &gstack, cpu->regs->cs); |
| 108 | push_guest_stack(cpu, &gstack, cpu->regs->eip); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 109 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 110 | /* 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] | 111 | if (has_err) |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 112 | push_guest_stack(cpu, &gstack, cpu->regs->errcode); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 113 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 114 | /* Now we've pushed all the old state, we change the stack, the code |
| 115 | * segment and the address to execute. */ |
Glauber de Oliveira Costa | a53a35a | 2008-01-07 11:05:32 -0200 | [diff] [blame] | 116 | cpu->regs->ss = ss; |
| 117 | cpu->regs->esp = virtstack + (gstack - origstack); |
| 118 | cpu->regs->cs = (__KERNEL_CS|GUEST_PL); |
| 119 | cpu->regs->eip = idt_address(lo, hi); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 120 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 121 | /* There are two kinds of interrupt handlers: 0xE is an "interrupt |
| 122 | * gate" which expects interrupts to be disabled on entry. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 123 | if (idt_type(lo, hi) == 0xE) |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 124 | if (put_user(0, &cpu->lg->lguest_data->irq_enabled)) |
| 125 | kill_guest(cpu, "Disabling interrupts"); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 126 | } |
| 127 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 128 | /*H:205 |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 129 | * Virtual Interrupts. |
| 130 | * |
Rusty Russell | abd41f0 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 131 | * interrupt_pending() returns the first pending interrupt which isn't blocked |
| 132 | * by the Guest. It is called before every entry to the Guest, and just before |
| 133 | * we go to sleep when the Guest has halted itself. */ |
Rusty Russell | a32a8813 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 134 | unsigned int interrupt_pending(struct lg_cpu *cpu, bool *more) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 135 | { |
| 136 | unsigned int irq; |
| 137 | DECLARE_BITMAP(blk, LGUEST_IRQS); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 138 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 139 | /* 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] | 140 | if (!cpu->lg->lguest_data) |
Rusty Russell | abd41f0 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 141 | return LGUEST_IRQS; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 142 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 143 | /* Take our "irqs_pending" array and remove any interrupts the Guest |
| 144 | * wants blocked: the result ends up in "blk". */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 145 | if (copy_from_user(&blk, cpu->lg->lguest_data->blocked_interrupts, |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 146 | sizeof(blk))) |
Rusty Russell | abd41f0 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 147 | return LGUEST_IRQS; |
Glauber de Oliveira Costa | 177e449 | 2008-01-07 11:05:29 -0200 | [diff] [blame] | 148 | bitmap_andnot(blk, cpu->irqs_pending, blk, LGUEST_IRQS); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 149 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 150 | /* Find the first interrupt. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 151 | irq = find_first_bit(blk, LGUEST_IRQS); |
Rusty Russell | a32a8813 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 152 | *more = find_next_bit(blk, LGUEST_IRQS, irq+1); |
Rusty Russell | abd41f0 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 153 | |
| 154 | return irq; |
| 155 | } |
| 156 | |
| 157 | /* This actually diverts the Guest to running an interrupt handler, once an |
| 158 | * interrupt has been identified by interrupt_pending(). */ |
Rusty Russell | a32a8813 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 159 | 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] | 160 | { |
| 161 | struct desc_struct *idt; |
| 162 | |
| 163 | BUG_ON(irq >= LGUEST_IRQS); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 164 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 165 | /* They may be in the middle of an iret, where they asked us never to |
| 166 | * deliver interrupts. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 167 | if (cpu->regs->eip >= cpu->lg->noirq_start && |
| 168 | (cpu->regs->eip < cpu->lg->noirq_end)) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 169 | return; |
| 170 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 171 | /* If they're halted, interrupts restart them. */ |
Glauber de Oliveira Costa | 66686c2 | 2008-01-07 11:05:34 -0200 | [diff] [blame] | 172 | if (cpu->halted) { |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 173 | /* Re-enable interrupts. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 174 | if (put_user(X86_EFLAGS_IF, &cpu->lg->lguest_data->irq_enabled)) |
| 175 | kill_guest(cpu, "Re-enabling interrupts"); |
Glauber de Oliveira Costa | 66686c2 | 2008-01-07 11:05:34 -0200 | [diff] [blame] | 176 | cpu->halted = 0; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 177 | } else { |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 178 | /* Otherwise we check if they have interrupts disabled. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 179 | u32 irq_enabled; |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 180 | if (get_user(irq_enabled, &cpu->lg->lguest_data->irq_enabled)) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 181 | irq_enabled = 0; |
Rusty Russell | a32a8813 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 182 | if (!irq_enabled) { |
| 183 | /* Make sure they know an IRQ is pending. */ |
| 184 | put_user(X86_EFLAGS_IF, |
| 185 | &cpu->lg->lguest_data->irq_pending); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 186 | return; |
Rusty Russell | a32a8813 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 187 | } |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 188 | } |
| 189 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 190 | /* Look at the IDT entry the Guest gave us for this interrupt. The |
| 191 | * first 32 (FIRST_EXTERNAL_VECTOR) entries are for traps, so we skip |
| 192 | * over them. */ |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 193 | idt = &cpu->arch.idt[FIRST_EXTERNAL_VECTOR+irq]; |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 194 | /* If they don't have a handler (yet?), we just ignore it */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 195 | if (idt_present(idt->a, idt->b)) { |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 196 | /* OK, mark it no longer pending and deliver it. */ |
Glauber de Oliveira Costa | 177e449 | 2008-01-07 11:05:29 -0200 | [diff] [blame] | 197 | clear_bit(irq, cpu->irqs_pending); |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 198 | /* set_guest_interrupt() takes the interrupt descriptor and a |
| 199 | * flag to say whether this interrupt pushes an error code onto |
| 200 | * the stack as well: virtual interrupts never do. */ |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 201 | set_guest_interrupt(cpu, idt->a, idt->b, false); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 202 | } |
Rusty Russell | 6c8dca5 | 2007-07-27 13:42:52 +1000 | [diff] [blame] | 203 | |
| 204 | /* Every time we deliver an interrupt, we update the timestamp in the |
| 205 | * Guest's lguest_data struct. It would be better for the Guest if we |
| 206 | * did this more often, but it can actually be quite slow: doing it |
| 207 | * here is a compromise which means at least it gets updated every |
| 208 | * timer interrupt. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 209 | write_timestamp(cpu); |
Rusty Russell | a32a8813 | 2009-06-12 22:27:02 -0600 | [diff] [blame] | 210 | |
| 211 | /* If there are no other interrupts we want to deliver, clear |
| 212 | * the pending flag. */ |
| 213 | if (!more) |
| 214 | put_user(0, &cpu->lg->lguest_data->irq_pending); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 215 | } |
Rusty Russell | 9f155a9 | 2009-06-12 22:27:08 -0600 | [diff] [blame] | 216 | |
| 217 | /* And this is the routine when we want to set an interrupt for the Guest. */ |
| 218 | void set_interrupt(struct lg_cpu *cpu, unsigned int irq) |
| 219 | { |
| 220 | /* Next time the Guest runs, the core code will see if it can deliver |
| 221 | * this interrupt. */ |
| 222 | set_bit(irq, cpu->irqs_pending); |
| 223 | |
| 224 | /* Make sure it sees it; it might be asleep (eg. halted), or |
| 225 | * running the Guest right now, in which case kick_process() |
| 226 | * will knock it out. */ |
| 227 | if (!wake_up_process(cpu->tsk)) |
| 228 | kick_process(cpu->tsk); |
| 229 | } |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 230 | /*:*/ |
| 231 | |
| 232 | /* Linux uses trap 128 for system calls. Plan9 uses 64, and Ron Minnich sent |
| 233 | * me a patch, so we support that too. It'd be a big step for lguest if half |
| 234 | * the Plan 9 user base were to start using it. |
| 235 | * |
| 236 | * Actually now I think of it, it's possible that Ron *is* half the Plan 9 |
| 237 | * userbase. Oh well. */ |
| 238 | static bool could_be_syscall(unsigned int num) |
| 239 | { |
| 240 | /* Normal Linux SYSCALL_VECTOR or reserved vector? */ |
| 241 | return num == SYSCALL_VECTOR || num == syscall_vector; |
| 242 | } |
| 243 | |
| 244 | /* The syscall vector it wants must be unused by Host. */ |
| 245 | bool check_syscall_vector(struct lguest *lg) |
| 246 | { |
| 247 | u32 vector; |
| 248 | |
| 249 | if (get_user(vector, &lg->lguest_data->syscall_vec)) |
| 250 | return false; |
| 251 | |
| 252 | return could_be_syscall(vector); |
| 253 | } |
| 254 | |
| 255 | int init_interrupts(void) |
| 256 | { |
| 257 | /* If they want some strange system call vector, reserve it now */ |
Yinghai Lu | b77b881 | 2008-12-19 15:23:44 -0800 | [diff] [blame] | 258 | if (syscall_vector != SYSCALL_VECTOR) { |
| 259 | if (test_bit(syscall_vector, used_vectors) || |
| 260 | vector_used_by_percpu_irq(syscall_vector)) { |
| 261 | printk(KERN_ERR "lg: couldn't reserve syscall %u\n", |
| 262 | syscall_vector); |
| 263 | return -EBUSY; |
| 264 | } |
| 265 | set_bit(syscall_vector, used_vectors); |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 266 | } |
Yinghai Lu | b77b881 | 2008-12-19 15:23:44 -0800 | [diff] [blame] | 267 | |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 268 | return 0; |
| 269 | } |
| 270 | |
| 271 | void free_interrupts(void) |
| 272 | { |
| 273 | if (syscall_vector != SYSCALL_VECTOR) |
| 274 | clear_bit(syscall_vector, used_vectors); |
| 275 | } |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 276 | |
Rusty Russell | a6bd8e1 | 2008-03-28 11:05:53 -0500 | [diff] [blame] | 277 | /*H:220 Now we've got the routines to deliver interrupts, delivering traps like |
| 278 | * page fault is easy. The only trick is that Intel decided that some traps |
| 279 | * should have error codes: */ |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 280 | static bool has_err(unsigned int trap) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 281 | { |
| 282 | return (trap == 8 || (trap >= 10 && trap <= 14) || trap == 17); |
| 283 | } |
| 284 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 285 | /* deliver_trap() returns true if it could deliver the trap. */ |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 286 | bool deliver_trap(struct lg_cpu *cpu, unsigned int num) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 287 | { |
Rusty Russell | 0d027c0 | 2007-08-09 20:57:13 +1000 | [diff] [blame] | 288 | /* Trap numbers are always 8 bit, but we set an impossible trap number |
| 289 | * for traps inside the Switcher, so check that here. */ |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 290 | if (num >= ARRAY_SIZE(cpu->arch.idt)) |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 291 | return false; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 292 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 293 | /* Early on the Guest hasn't set the IDT entries (or maybe it put a |
| 294 | * bogus one in): if we fail here, the Guest will be killed. */ |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 295 | 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] | 296 | return false; |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 297 | set_guest_interrupt(cpu, cpu->arch.idt[num].a, |
| 298 | cpu->arch.idt[num].b, has_err(num)); |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 299 | return true; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 300 | } |
| 301 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 302 | /*H:250 Here's the hard part: returning to the Host every time a trap happens |
| 303 | * and then calling deliver_trap() and re-entering the Guest is slow. |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 304 | * Particularly because Guest userspace system calls are traps (usually trap |
| 305 | * 128). |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 306 | * |
| 307 | * So we'd like to set up the IDT to tell the CPU to deliver traps directly |
| 308 | * into the Guest. This is possible, but the complexities cause the size of |
| 309 | * this file to double! However, 150 lines of code is worth writing for taking |
| 310 | * 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] | 311 | * the other hypervisors would beat it up at lunchtime. |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 312 | * |
Rusty Russell | 56adbe9 | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 313 | * This routine indicates if a particular trap number could be delivered |
| 314 | * directly. */ |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 315 | static bool direct_trap(unsigned int num) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 316 | { |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 317 | /* Hardware interrupts don't go to the Guest at all (except system |
| 318 | * call). */ |
Rusty Russell | c18acd7 | 2007-10-22 11:03:35 +1000 | [diff] [blame] | 319 | if (num >= FIRST_EXTERNAL_VECTOR && !could_be_syscall(num)) |
Matias Zabaljauregui | df1693a | 2009-03-18 13:38:35 -0300 | [diff] [blame] | 320 | return false; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 321 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 322 | /* The Host needs to see page faults (for shadow paging and to save the |
| 323 | * fault address), general protection faults (in/out emulation) and |
Matias Zabaljauregui | 4cd8b5e | 2009-03-14 13:37:52 -0200 | [diff] [blame] | 324 | * device not available (TS handling), invalid opcode fault (kvm hcall), |
| 325 | * and of course, the hypercall trap. */ |
| 326 | return num != 14 && num != 13 && num != 7 && |
| 327 | num != 6 && num != LGUEST_TRAP_ENTRY; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 328 | } |
Rusty Russell | f56a384 | 2007-07-26 10:41:05 -0700 | [diff] [blame] | 329 | /*:*/ |
| 330 | |
| 331 | /*M:005 The Guest has the ability to turn its interrupt gates into trap gates, |
| 332 | * if it is careful. The Host will let trap gates can go directly to the |
| 333 | * Guest, but the Guest needs the interrupts atomically disabled for an |
| 334 | * interrupt gate. It can do this by pointing the trap gate at instructions |
| 335 | * within noirq_start and noirq_end, where it can safely disable interrupts. */ |
| 336 | |
| 337 | /*M:006 The Guests do not use the sysenter (fast system call) instruction, |
| 338 | * because it's hardcoded to enter privilege level 0 and so can't go direct. |
| 339 | * It's about twice as fast as the older "int 0x80" system call, so it might |
| 340 | * still be worthwhile to handle it in the Switcher and lcall down to the |
| 341 | * Guest. The sysenter semantics are hairy tho: search for that keyword in |
| 342 | * entry.S :*/ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 343 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 344 | /*H:260 When we make traps go directly into the Guest, we need to make sure |
| 345 | * the kernel stack is valid (ie. mapped in the page tables). Otherwise, the |
| 346 | * CPU trying to deliver the trap will fault while trying to push the interrupt |
| 347 | * words on the stack: this is called a double fault, and it forces us to kill |
| 348 | * the Guest. |
| 349 | * |
| 350 | * Which is deeply unfair, because (literally!) it wasn't the Guests' fault. */ |
Glauber de Oliveira Costa | 4665ac8e | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 351 | void pin_stack_pages(struct lg_cpu *cpu) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 352 | { |
| 353 | unsigned int i; |
| 354 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 355 | /* Depending on the CONFIG_4KSTACKS option, the Guest can have one or |
| 356 | * two pages of stack space. */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 357 | for (i = 0; i < cpu->lg->stack_pages; i++) |
Rusty Russell | 8057d76 | 2007-08-30 06:35:08 +1000 | [diff] [blame] | 358 | /* The stack grows *upwards*, so the address we're given is the |
| 359 | * start of the page after the kernel stack. Subtract one to |
| 360 | * get back onto the first stack page, and keep subtracting to |
| 361 | * get to the rest of the stack pages. */ |
Glauber de Oliveira Costa | 1713608 | 2008-01-07 11:05:37 -0200 | [diff] [blame] | 362 | pin_page(cpu, cpu->esp1 - 1 - i * PAGE_SIZE); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 363 | } |
| 364 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 365 | /* Direct traps also mean that we need to know whenever the Guest wants to use |
| 366 | * a different kernel stack, so we can change the IDT entries to use that |
| 367 | * stack. The IDT entries expect a virtual address, so unlike most addresses |
| 368 | * the Guest gives us, the "esp" (stack pointer) value here is virtual, not |
| 369 | * physical. |
| 370 | * |
| 371 | * In Linux each process has its own kernel stack, so this happens a lot: we |
| 372 | * change stacks on each context switch. */ |
Glauber de Oliveira Costa | 4665ac8e | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 373 | 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] | 374 | { |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 375 | /* You are not allowed have a stack segment with privilege level 0: bad |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 376 | * Guest! */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 377 | if ((seg & 0x3) != GUEST_PL) |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 378 | kill_guest(cpu, "bad stack segment %i", seg); |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 379 | /* We only expect one or two stack pages. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 380 | if (pages > 2) |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 381 | kill_guest(cpu, "bad stack pages %u", pages); |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 382 | /* Save where the stack is, and how many pages */ |
Glauber de Oliveira Costa | 4665ac8e | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 383 | cpu->ss1 = seg; |
| 384 | cpu->esp1 = esp; |
| 385 | cpu->lg->stack_pages = pages; |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 386 | /* Make sure the new stack pages are mapped */ |
Glauber de Oliveira Costa | 4665ac8e | 2008-01-07 11:05:35 -0200 | [diff] [blame] | 387 | pin_stack_pages(cpu); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 388 | } |
| 389 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 390 | /* All this reference to mapping stacks leads us neatly into the other complex |
| 391 | * part of the Host: page table handling. */ |
| 392 | |
| 393 | /*H:235 This is the routine which actually checks the Guest's IDT entry and |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 394 | * transfers it into the entry in "struct lguest": */ |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 395 | static void set_trap(struct lg_cpu *cpu, struct desc_struct *trap, |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 396 | unsigned int num, u32 lo, u32 hi) |
| 397 | { |
| 398 | u8 type = idt_type(lo, hi); |
| 399 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 400 | /* We zero-out a not-present entry */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 401 | if (!idt_present(lo, hi)) { |
| 402 | trap->a = trap->b = 0; |
| 403 | return; |
| 404 | } |
| 405 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 406 | /* We only support interrupt and trap gates. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 407 | if (type != 0xE && type != 0xF) |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 408 | kill_guest(cpu, "bad IDT type %i", type); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 409 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 410 | /* We only copy the handler address, present bit, privilege level and |
| 411 | * type. The privilege level controls where the trap can be triggered |
| 412 | * manually with an "int" instruction. This is usually GUEST_PL, |
| 413 | * except for system calls which userspace can use. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 414 | trap->a = ((__KERNEL_CS|GUEST_PL)<<16) | (lo&0x0000FFFF); |
| 415 | trap->b = (hi&0xFFFFEF00); |
| 416 | } |
| 417 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 418 | /*H:230 While we're here, dealing with delivering traps and interrupts to the |
| 419 | * Guest, we might as well complete the picture: how the Guest tells us where |
| 420 | * it wants them to go. This would be simple, except making traps fast |
| 421 | * requires some tricks. |
| 422 | * |
| 423 | * We saw the Guest setting Interrupt Descriptor Table (IDT) entries with the |
| 424 | * LHCALL_LOAD_IDT_ENTRY hypercall before: that comes here. */ |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 425 | 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] | 426 | { |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 427 | /* Guest never handles: NMI, doublefault, spurious interrupt or |
| 428 | * hypercall. We ignore when it tries to set them. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 429 | if (num == 2 || num == 8 || num == 15 || num == LGUEST_TRAP_ENTRY) |
| 430 | return; |
| 431 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 432 | /* Mark the IDT as changed: next time the Guest runs we'll know we have |
| 433 | * to copy this again. */ |
Glauber de Oliveira Costa | ae3749d | 2008-01-17 19:14:46 -0200 | [diff] [blame] | 434 | cpu->changed |= CHANGED_IDT; |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 435 | |
Rusty Russell | 56adbe9 | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 436 | /* 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] | 437 | if (num >= ARRAY_SIZE(cpu->arch.idt)) |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 438 | kill_guest(cpu, "Setting idt entry %u", num); |
Rusty Russell | 56adbe9 | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 439 | else |
Glauber de Oliveira Costa | 382ac6b | 2008-01-17 19:19:42 -0200 | [diff] [blame] | 440 | set_trap(cpu, &cpu->arch.idt[num], num, lo, hi); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 441 | } |
| 442 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 443 | /* The default entry for each interrupt points into the Switcher routines which |
| 444 | * simply return to the Host. The run_guest() loop will then call |
| 445 | * deliver_trap() to bounce it back into the Guest. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 446 | static void default_idt_entry(struct desc_struct *idt, |
| 447 | int trap, |
Rusty Russell | 0c12091 | 2008-07-29 09:58:31 -0500 | [diff] [blame] | 448 | const unsigned long handler, |
| 449 | const struct desc_struct *base) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 450 | { |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 451 | /* A present interrupt gate. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 452 | u32 flags = 0x8e00; |
| 453 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 454 | /* Set the privilege level on the entry for the hypercall: this allows |
| 455 | * the Guest to use the "int" instruction to trigger it. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 456 | if (trap == LGUEST_TRAP_ENTRY) |
| 457 | flags |= (GUEST_PL << 13); |
Rusty Russell | 0c12091 | 2008-07-29 09:58:31 -0500 | [diff] [blame] | 458 | else if (base) |
| 459 | /* Copy priv. level from what Guest asked for. This allows |
| 460 | * debug (int 3) traps from Guest userspace, for example. */ |
| 461 | flags |= (base->b & 0x6000); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 462 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 463 | /* Now pack it into the IDT entry in its weird format. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 464 | idt->a = (LGUEST_CS<<16) | (handler&0x0000FFFF); |
| 465 | idt->b = (handler&0xFFFF0000) | flags; |
| 466 | } |
| 467 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 468 | /* When the Guest first starts, we put default entries into the IDT. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 469 | void setup_default_idt_entries(struct lguest_ro_state *state, |
| 470 | const unsigned long *def) |
| 471 | { |
| 472 | unsigned int i; |
| 473 | |
| 474 | for (i = 0; i < ARRAY_SIZE(state->guest_idt); i++) |
Rusty Russell | 0c12091 | 2008-07-29 09:58:31 -0500 | [diff] [blame] | 475 | default_idt_entry(&state->guest_idt[i], i, def[i], NULL); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 476 | } |
| 477 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 478 | /*H:240 We don't use the IDT entries in the "struct lguest" directly, instead |
| 479 | * we copy them into the IDT which we've set up for Guests on this CPU, just |
| 480 | * before we run the Guest. This routine does that copy. */ |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 481 | void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt, |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 482 | const unsigned long *def) |
| 483 | { |
| 484 | unsigned int i; |
| 485 | |
Rusty Russell | bff672e | 2007-07-26 10:41:04 -0700 | [diff] [blame] | 486 | /* We can simply copy the direct traps, otherwise we use the default |
| 487 | * ones in the Switcher: they will return to the Host. */ |
Glauber de Oliveira Costa | fc708b3 | 2008-01-07 11:05:33 -0200 | [diff] [blame] | 488 | for (i = 0; i < ARRAY_SIZE(cpu->arch.idt); i++) { |
Rusty Russell | 0c12091 | 2008-07-29 09:58:31 -0500 | [diff] [blame] | 489 | const struct desc_struct *gidt = &cpu->arch.idt[i]; |
| 490 | |
Rusty Russell | 56adbe9 | 2007-10-22 11:03:28 +1000 | [diff] [blame] | 491 | /* If no Guest can ever override this trap, leave it alone. */ |
| 492 | if (!direct_trap(i)) |
| 493 | continue; |
| 494 | |
| 495 | /* Only trap gates (type 15) can go direct to the Guest. |
| 496 | * Interrupt gates (type 14) disable interrupts as they are |
| 497 | * entered, which we never let the Guest do. Not present |
Rusty Russell | 0c12091 | 2008-07-29 09:58:31 -0500 | [diff] [blame] | 498 | * entries (type 0x0) also can't go direct, of course. |
| 499 | * |
| 500 | * If it can't go direct, we still need to copy the priv. level: |
| 501 | * they might want to give userspace access to a software |
| 502 | * interrupt. */ |
| 503 | if (idt_type(gidt->a, gidt->b) == 0xF) |
| 504 | idt[i] = *gidt; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 505 | else |
Rusty Russell | 0c12091 | 2008-07-29 09:58:31 -0500 | [diff] [blame] | 506 | default_idt_entry(&idt[i], i, def[i], gidt); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 507 | } |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 508 | } |
| 509 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 510 | /*H:200 |
| 511 | * The Guest Clock. |
| 512 | * |
| 513 | * There are two sources of virtual interrupts. We saw one in lguest_user.c: |
| 514 | * the Launcher sending interrupts for virtual devices. The other is the Guest |
| 515 | * timer interrupt. |
| 516 | * |
| 517 | * The Guest uses the LHCALL_SET_CLOCKEVENT hypercall to tell us how long to |
| 518 | * the next timer interrupt (in nanoseconds). We use the high-resolution timer |
| 519 | * infrastructure to set a callback at that time. |
| 520 | * |
| 521 | * 0 means "turn off the clock". */ |
Glauber de Oliveira Costa | ad8d8f3 | 2008-01-07 11:05:28 -0200 | [diff] [blame] | 522 | void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 523 | { |
| 524 | ktime_t expires; |
| 525 | |
| 526 | if (unlikely(delta == 0)) { |
| 527 | /* Clock event device is shutting down. */ |
Glauber de Oliveira Costa | ad8d8f3 | 2008-01-07 11:05:28 -0200 | [diff] [blame] | 528 | hrtimer_cancel(&cpu->hrt); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 529 | return; |
| 530 | } |
| 531 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 532 | /* We use wallclock time here, so the Guest might not be running for |
| 533 | * all the time between now and the timer interrupt it asked for. This |
| 534 | * is almost always the right thing to do. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 535 | expires = ktime_add_ns(ktime_get_real(), delta); |
Glauber de Oliveira Costa | ad8d8f3 | 2008-01-07 11:05:28 -0200 | [diff] [blame] | 536 | hrtimer_start(&cpu->hrt, expires, HRTIMER_MODE_ABS); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 537 | } |
| 538 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 539 | /* This is the function called when the Guest's timer expires. */ |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 540 | static enum hrtimer_restart clockdev_fn(struct hrtimer *timer) |
| 541 | { |
Glauber de Oliveira Costa | ad8d8f3 | 2008-01-07 11:05:28 -0200 | [diff] [blame] | 542 | struct lg_cpu *cpu = container_of(timer, struct lg_cpu, hrt); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 543 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 544 | /* Remember the first interrupt is the timer interrupt. */ |
Rusty Russell | 9f155a9 | 2009-06-12 22:27:08 -0600 | [diff] [blame] | 545 | set_interrupt(cpu, 0); |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 546 | return HRTIMER_NORESTART; |
| 547 | } |
| 548 | |
Rusty Russell | e1e7296 | 2007-10-25 15:02:50 +1000 | [diff] [blame] | 549 | /* This sets up the timer for this Guest. */ |
Glauber de Oliveira Costa | ad8d8f3 | 2008-01-07 11:05:28 -0200 | [diff] [blame] | 550 | void init_clockdev(struct lg_cpu *cpu) |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 551 | { |
Glauber de Oliveira Costa | ad8d8f3 | 2008-01-07 11:05:28 -0200 | [diff] [blame] | 552 | hrtimer_init(&cpu->hrt, CLOCK_REALTIME, HRTIMER_MODE_ABS); |
| 553 | cpu->hrt.function = clockdev_fn; |
Rusty Russell | d7e28ff | 2007-07-19 01:49:23 -0700 | [diff] [blame] | 554 | } |