blob: eb934b0242e0e17652b7fc17c3255b5e70e40d15 [file] [log] [blame]
Rusty Russell2e04ef72009-07-30 16:03:45 -06001/*P:800
2 * Interrupts (traps) are complicated enough to earn their own file.
Rusty Russellf938d2c2007-07-26 10:41:02 -07003 * 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 Russell2e04ef72009-07-30 16:03:45 -060014 * they had been delivered to it directly.
15:*/
Rusty Russelld7e28ff2007-07-19 01:49:23 -070016#include <linux/uaccess.h>
Rusty Russellc18acd72007-10-22 11:03:35 +100017#include <linux/interrupt.h>
18#include <linux/module.h>
Alexey Dobriyand43c36d2009-10-07 17:09:06 +040019#include <linux/sched.h>
Rusty Russelld7e28ff2007-07-19 01:49:23 -070020#include "lg.h"
21
Rusty Russellc18acd72007-10-22 11:03:35 +100022/* Allow Guests to use a non-128 (ie. non-Linux) syscall trap. */
Ingo Molnarf435e682015-05-11 07:17:04 +020023static unsigned int syscall_vector = IA32_SYSCALL_VECTOR;
Rusty Russellc18acd72007-10-22 11:03:35 +100024module_param(syscall_vector, uint, 0444);
25
Rusty Russellbff672e2007-07-26 10:41:04 -070026/* The address of the interrupt handler is split into two bits: */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070027static unsigned long idt_address(u32 lo, u32 hi)
28{
29 return (lo & 0x0000FFFF) | (hi & 0xFFFF0000);
30}
31
Rusty Russell2e04ef72009-07-30 16:03:45 -060032/*
33 * The "type" of the interrupt handler is a 4 bit field: we only support a
34 * couple of types.
35 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070036static int idt_type(u32 lo, u32 hi)
37{
38 return (hi >> 8) & 0xF;
39}
40
Rusty Russellbff672e2007-07-26 10:41:04 -070041/* An IDT entry can't be used unless the "present" bit is set. */
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -030042static bool idt_present(u32 lo, u32 hi)
Rusty Russelld7e28ff2007-07-19 01:49:23 -070043{
44 return (hi & 0x8000);
45}
46
Rusty Russell2e04ef72009-07-30 16:03:45 -060047/*
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 Costa382ac6b2008-01-17 19:19:42 -020051static void push_guest_stack(struct lg_cpu *cpu, unsigned long *gstack, u32 val)
Rusty Russelld7e28ff2007-07-19 01:49:23 -070052{
Rusty Russellbff672e2007-07-26 10:41:04 -070053 /* Stack grows upwards: move stack then write value. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070054 *gstack -= 4;
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -020055 lgwrite(cpu, *gstack, u32, val);
Rusty Russelld7e28ff2007-07-19 01:49:23 -070056}
57
Rusty Russell2e04ef72009-07-30 16:03:45 -060058/*H:210
Rusty Russell3eebd232015-04-01 13:32:20 +103059 * 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 Russellbff672e2007-07-26 10:41:04 -070063 *
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 Russell2e04ef72009-07-30 16:03:45 -060066 * it).
67 */
Rusty Russell3eebd232015-04-01 13:32:20 +103068static void push_guest_interrupt_stack(struct lg_cpu *cpu, bool has_err)
Rusty Russelld7e28ff2007-07-19 01:49:23 -070069{
Rusty Russell47436aa2007-10-22 11:03:36 +100070 unsigned long gstack, origstack;
Rusty Russelld7e28ff2007-07-19 01:49:23 -070071 u32 eflags, ss, irq_enable;
Rusty Russell47436aa2007-10-22 11:03:36 +100072 unsigned long virtstack;
Rusty Russelld7e28ff2007-07-19 01:49:23 -070073
Rusty Russell2e04ef72009-07-30 16:03:45 -060074 /*
75 * There are two cases for interrupts: one where the Guest is already
Rusty Russellbff672e2007-07-26 10:41:04 -070076 * in the kernel, and a more complex one where the Guest is in
Rusty Russell2e04ef72009-07-30 16:03:45 -060077 * userspace. We check the privilege level to find out.
78 */
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -020079 if ((cpu->regs->ss&0x3) != GUEST_PL) {
Rusty Russell2e04ef72009-07-30 16:03:45 -060080 /*
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 Costa4665ac8e2008-01-07 11:05:35 -020084 virtstack = cpu->esp1;
85 ss = cpu->ss1;
Rusty Russell47436aa2007-10-22 11:03:36 +100086
Glauber de Oliveira Costa17136082008-01-07 11:05:37 -020087 origstack = gstack = guest_pa(cpu, virtstack);
Rusty Russell2e04ef72009-07-30 16:03:45 -060088 /*
89 * We push the old stack segment and pointer onto the new
Rusty Russellbff672e2007-07-26 10:41:04 -070090 * stack: when the Guest does an "iret" back from the interrupt
91 * handler the CPU will notice they're dropping privilege
Rusty Russell2e04ef72009-07-30 16:03:45 -060092 * levels and expect these here.
93 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -020094 push_guest_stack(cpu, &gstack, cpu->regs->ss);
95 push_guest_stack(cpu, &gstack, cpu->regs->esp);
Rusty Russelld7e28ff2007-07-19 01:49:23 -070096 } else {
Rusty Russellbff672e2007-07-26 10:41:04 -070097 /* We're staying on the same Guest (kernel) stack. */
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -020098 virtstack = cpu->regs->esp;
99 ss = cpu->regs->ss;
Rusty Russell47436aa2007-10-22 11:03:36 +1000100
Glauber de Oliveira Costa17136082008-01-07 11:05:37 -0200101 origstack = gstack = guest_pa(cpu, virtstack);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700102 }
103
Rusty Russell2e04ef72009-07-30 16:03:45 -0600104 /*
105 * Remember that we never let the Guest actually disable interrupts, so
Rusty Russellbff672e2007-07-26 10:41:04 -0700106 * the "Interrupt Flag" bit is always set. We copy that bit from the
Rusty Russelle1e72962007-10-25 15:02:50 +1000107 * Guest's "irq_enabled" field into the eflags word: we saw the Guest
Rusty Russell2e04ef72009-07-30 16:03:45 -0600108 * copy it back in "lguest_iret".
109 */
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -0200110 eflags = cpu->regs->eflags;
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200111 if (get_user(irq_enable, &cpu->lg->lguest_data->irq_enabled) == 0
Rusty Russelle5faff42007-07-20 22:11:13 +1000112 && !(irq_enable & X86_EFLAGS_IF))
113 eflags &= ~X86_EFLAGS_IF;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700114
Rusty Russell2e04ef72009-07-30 16:03:45 -0600115 /*
116 * An interrupt is expected to push three things on the stack: the old
Rusty Russellbff672e2007-07-26 10:41:04 -0700117 * "eflags" word, the old code segment, and the old instruction
Rusty Russell2e04ef72009-07-30 16:03:45 -0600118 * pointer.
119 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200120 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 Russelld7e28ff2007-07-19 01:49:23 -0700123
Rusty Russellbff672e2007-07-26 10:41:04 -0700124 /* For the six traps which supply an error code, we push that, too. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700125 if (has_err)
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200126 push_guest_stack(cpu, &gstack, cpu->regs->errcode);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700127
Rusty Russell3eebd232015-04-01 13:32:20 +1030128 /* Adjust the stack pointer and stack segment. */
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -0200129 cpu->regs->ss = ss;
130 cpu->regs->esp = virtstack + (gstack - origstack);
Rusty Russell3eebd232015-04-01 13:32:20 +1030131}
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 */
141static 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 Costaa53a35a2008-01-07 11:05:32 -0200150 cpu->regs->cs = (__KERNEL_CS|GUEST_PL);
151 cpu->regs->eip = idt_address(lo, hi);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700152
Rusty Russell2e04ef72009-07-30 16:03:45 -0600153 /*
Rusty Russell98fb4e52013-09-05 17:45:53 +0930154 * 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 Russell2e04ef72009-07-30 16:03:45 -0600164 * There are two kinds of interrupt handlers: 0xE is an "interrupt
165 * gate" which expects interrupts to be disabled on entry.
166 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700167 if (idt_type(lo, hi) == 0xE)
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200168 if (put_user(0, &cpu->lg->lguest_data->irq_enabled))
169 kill_guest(cpu, "Disabling interrupts");
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700170}
171
Rusty Russell3eebd232015-04-01 13:32:20 +1030172/* This restores the eflags word which was pushed on the stack by a trap */
173static 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 Russelle1e72962007-10-25 15:02:50 +1000190/*H:205
Rusty Russellbff672e2007-07-26 10:41:04 -0700191 * Virtual Interrupts.
192 *
Rusty Russellabd41f02009-06-12 22:27:02 -0600193 * 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 Russell2e04ef72009-07-30 16:03:45 -0600195 * we go to sleep when the Guest has halted itself.
196 */
Rusty Russella32a88132009-06-12 22:27:02 -0600197unsigned int interrupt_pending(struct lg_cpu *cpu, bool *more)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700198{
199 unsigned int irq;
200 DECLARE_BITMAP(blk, LGUEST_IRQS);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700201
Rusty Russellbff672e2007-07-26 10:41:04 -0700202 /* If the Guest hasn't even initialized yet, we can do nothing. */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200203 if (!cpu->lg->lguest_data)
Rusty Russellabd41f02009-06-12 22:27:02 -0600204 return LGUEST_IRQS;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700205
Rusty Russell2e04ef72009-07-30 16:03:45 -0600206 /*
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 Costa382ac6b2008-01-17 19:19:42 -0200210 if (copy_from_user(&blk, cpu->lg->lguest_data->blocked_interrupts,
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700211 sizeof(blk)))
Rusty Russellabd41f02009-06-12 22:27:02 -0600212 return LGUEST_IRQS;
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -0200213 bitmap_andnot(blk, cpu->irqs_pending, blk, LGUEST_IRQS);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700214
Rusty Russellbff672e2007-07-26 10:41:04 -0700215 /* Find the first interrupt. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700216 irq = find_first_bit(blk, LGUEST_IRQS);
Rusty Russella32a88132009-06-12 22:27:02 -0600217 *more = find_next_bit(blk, LGUEST_IRQS, irq+1);
Rusty Russellabd41f02009-06-12 22:27:02 -0600218
219 return irq;
220}
221
Rusty Russell2e04ef72009-07-30 16:03:45 -0600222/*
223 * This actually diverts the Guest to running an interrupt handler, once an
224 * interrupt has been identified by interrupt_pending().
225 */
Rusty Russella32a88132009-06-12 22:27:02 -0600226void try_deliver_interrupt(struct lg_cpu *cpu, unsigned int irq, bool more)
Rusty Russellabd41f02009-06-12 22:27:02 -0600227{
228 struct desc_struct *idt;
229
230 BUG_ON(irq >= LGUEST_IRQS);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700231
Rusty Russellbff672e2007-07-26 10:41:04 -0700232 /* If they're halted, interrupts restart them. */
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -0200233 if (cpu->halted) {
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700234 /* Re-enable interrupts. */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200235 if (put_user(X86_EFLAGS_IF, &cpu->lg->lguest_data->irq_enabled))
236 kill_guest(cpu, "Re-enabling interrupts");
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -0200237 cpu->halted = 0;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700238 } else {
Rusty Russellbff672e2007-07-26 10:41:04 -0700239 /* Otherwise we check if they have interrupts disabled. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700240 u32 irq_enabled;
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200241 if (get_user(irq_enabled, &cpu->lg->lguest_data->irq_enabled))
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700242 irq_enabled = 0;
Rusty Russella32a88132009-06-12 22:27:02 -0600243 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 Russelld7e28ff2007-07-19 01:49:23 -0700247 return;
Rusty Russella32a88132009-06-12 22:27:02 -0600248 }
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700249 }
250
Rusty Russell2e04ef72009-07-30 16:03:45 -0600251 /*
252 * Look at the IDT entry the Guest gave us for this interrupt. The
Rusty Russellbff672e2007-07-26 10:41:04 -0700253 * first 32 (FIRST_EXTERNAL_VECTOR) entries are for traps, so we skip
Rusty Russell2e04ef72009-07-30 16:03:45 -0600254 * over them.
255 */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200256 idt = &cpu->arch.idt[FIRST_EXTERNAL_VECTOR+irq];
Rusty Russellbff672e2007-07-26 10:41:04 -0700257 /* If they don't have a handler (yet?), we just ignore it */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700258 if (idt_present(idt->a, idt->b)) {
Rusty Russellbff672e2007-07-26 10:41:04 -0700259 /* OK, mark it no longer pending and deliver it. */
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -0200260 clear_bit(irq, cpu->irqs_pending);
Rusty Russell3eebd232015-04-01 13:32:20 +1030261
Rusty Russell2e04ef72009-07-30 16:03:45 -0600262 /*
Rusty Russell3eebd232015-04-01 13:32:20 +1030263 * 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 Russell2e04ef72009-07-30 16:03:45 -0600276 */
Rusty Russell3eebd232015-04-01 13:32:20 +1030277 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 Russelld7e28ff2007-07-19 01:49:23 -0700289 }
Rusty Russell6c8dca52007-07-27 13:42:52 +1000290
Rusty Russell2e04ef72009-07-30 16:03:45 -0600291 /*
292 * Every time we deliver an interrupt, we update the timestamp in the
Rusty Russell6c8dca52007-07-27 13:42:52 +1000293 * 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 Russell2e04ef72009-07-30 16:03:45 -0600296 * timer interrupt.
297 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200298 write_timestamp(cpu);
Rusty Russella32a88132009-06-12 22:27:02 -0600299
Rusty Russell2e04ef72009-07-30 16:03:45 -0600300 /*
301 * If there are no other interrupts we want to deliver, clear
302 * the pending flag.
303 */
Rusty Russella32a88132009-06-12 22:27:02 -0600304 if (!more)
305 put_user(0, &cpu->lg->lguest_data->irq_pending);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700306}
Rusty Russell9f155a92009-06-12 22:27:08 -0600307
308/* And this is the routine when we want to set an interrupt for the Guest. */
309void set_interrupt(struct lg_cpu *cpu, unsigned int irq)
310{
Rusty Russell2e04ef72009-07-30 16:03:45 -0600311 /*
312 * Next time the Guest runs, the core code will see if it can deliver
313 * this interrupt.
314 */
Rusty Russell9f155a92009-06-12 22:27:08 -0600315 set_bit(irq, cpu->irqs_pending);
316
Rusty Russell2e04ef72009-07-30 16:03:45 -0600317 /*
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 Russell9f155a92009-06-12 22:27:08 -0600321 if (!wake_up_process(cpu->tsk))
322 kick_process(cpu->tsk);
323}
Rusty Russellc18acd72007-10-22 11:03:35 +1000324/*:*/
325
Rusty Russell2e04ef72009-07-30 16:03:45 -0600326/*
327 * Linux uses trap 128 for system calls. Plan9 uses 64, and Ron Minnich sent
Rusty Russellc18acd72007-10-22 11:03:35 +1000328 * 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 Russell2e04ef72009-07-30 16:03:45 -0600332 * userbase. Oh well.
333 */
Rusty Russellc18acd72007-10-22 11:03:35 +1000334static bool could_be_syscall(unsigned int num)
335{
Ingo Molnarf435e682015-05-11 07:17:04 +0200336 /* Normal Linux IA32_SYSCALL_VECTOR or reserved vector? */
337 return num == IA32_SYSCALL_VECTOR || num == syscall_vector;
Rusty Russellc18acd72007-10-22 11:03:35 +1000338}
339
340/* The syscall vector it wants must be unused by Host. */
341bool 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
351int init_interrupts(void)
352{
353 /* If they want some strange system call vector, reserve it now */
Ingo Molnarf435e682015-05-11 07:17:04 +0200354 if (syscall_vector != IA32_SYSCALL_VECTOR) {
Yinghai Lub77b8812008-12-19 15:23:44 -0800355 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 Russellc18acd72007-10-22 11:03:35 +1000362 }
Yinghai Lub77b8812008-12-19 15:23:44 -0800363
Rusty Russellc18acd72007-10-22 11:03:35 +1000364 return 0;
365}
366
367void free_interrupts(void)
368{
Ingo Molnarf435e682015-05-11 07:17:04 +0200369 if (syscall_vector != IA32_SYSCALL_VECTOR)
Rusty Russellc18acd72007-10-22 11:03:35 +1000370 clear_bit(syscall_vector, used_vectors);
371}
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700372
Rusty Russell2e04ef72009-07-30 16:03:45 -0600373/*H:220
374 * Now we've got the routines to deliver interrupts, delivering traps like
Rusty Russella6bd8e12008-03-28 11:05:53 -0500375 * page fault is easy. The only trick is that Intel decided that some traps
Rusty Russell2e04ef72009-07-30 16:03:45 -0600376 * should have error codes:
377 */
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300378static bool has_err(unsigned int trap)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700379{
380 return (trap == 8 || (trap >= 10 && trap <= 14) || trap == 17);
381}
382
Rusty Russellbff672e2007-07-26 10:41:04 -0700383/* deliver_trap() returns true if it could deliver the trap. */
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300384bool deliver_trap(struct lg_cpu *cpu, unsigned int num)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700385{
Rusty Russell2e04ef72009-07-30 16:03:45 -0600386 /*
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 Costafc708b32008-01-07 11:05:33 -0200390 if (num >= ARRAY_SIZE(cpu->arch.idt))
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300391 return false;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700392
Rusty Russell2e04ef72009-07-30 16:03:45 -0600393 /*
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 Costafc708b32008-01-07 11:05:33 -0200397 if (!idt_present(cpu->arch.idt[num].a, cpu->arch.idt[num].b))
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300398 return false;
Rusty Russell3eebd232015-04-01 13:32:20 +1030399 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 Zabaljaureguidf1693a2009-03-18 13:38:35 -0300402 return true;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700403}
404
Rusty Russell2e04ef72009-07-30 16:03:45 -0600405/*H:250
406 * Here's the hard part: returning to the Host every time a trap happens
Rusty Russellbff672e2007-07-26 10:41:04 -0700407 * and then calling deliver_trap() and re-entering the Guest is slow.
Rusty Russelle1e72962007-10-25 15:02:50 +1000408 * Particularly because Guest userspace system calls are traps (usually trap
409 * 128).
Rusty Russellbff672e2007-07-26 10:41:04 -0700410 *
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 Russelle1e72962007-10-25 15:02:50 +1000415 * the other hypervisors would beat it up at lunchtime.
Rusty Russellbff672e2007-07-26 10:41:04 -0700416 *
Rusty Russell56adbe92007-10-22 11:03:28 +1000417 * This routine indicates if a particular trap number could be delivered
Rusty Russell2e04ef72009-07-30 16:03:45 -0600418 * directly.
419 */
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300420static bool direct_trap(unsigned int num)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700421{
Rusty Russell2e04ef72009-07-30 16:03:45 -0600422 /*
423 * Hardware interrupts don't go to the Guest at all (except system
424 * call).
425 */
Rusty Russellc18acd72007-10-22 11:03:35 +1000426 if (num >= FIRST_EXTERNAL_VECTOR && !could_be_syscall(num))
Matias Zabaljaureguidf1693a2009-03-18 13:38:35 -0300427 return false;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700428
Rusty Russell2e04ef72009-07-30 16:03:45 -0600429 /*
430 * The Host needs to see page faults (for shadow paging and to save the
Rusty Russellbff672e2007-07-26 10:41:04 -0700431 * fault address), general protection faults (in/out emulation) and
Rusty Russell6d7a5d12011-07-22 14:39:49 +0930432 * device not available (TS handling) and of course, the hypercall trap.
Rusty Russell2e04ef72009-07-30 16:03:45 -0600433 */
Rusty Russell6d7a5d12011-07-22 14:39:49 +0930434 return num != 14 && num != 13 && num != 7 && num != LGUEST_TRAP_ENTRY;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700435}
Rusty Russellf56a3842007-07-26 10:41:05 -0700436/*:*/
437
Rusty Russell2e04ef72009-07-30 16:03:45 -0600438/*M:005
439 * The Guest has the ability to turn its interrupt gates into trap gates,
Rusty Russellf56a3842007-07-26 10:41:05 -0700440 * if it is careful. The Host will let trap gates can go directly to the
441 * Guest, but the Guest needs the interrupts atomically disabled for an
Rusty Russell2f921b52015-03-24 11:51:39 +1030442 * interrupt gate. The Host could provide a mechanism to register more
443 * "no-interrupt" regions, and the Guest could point the trap gate at
444 * instructions within that region, where it can safely disable interrupts.
Rusty Russell2e04ef72009-07-30 16:03:45 -0600445 */
Rusty Russellf56a3842007-07-26 10:41:05 -0700446
Rusty Russell2e04ef72009-07-30 16:03:45 -0600447/*M:006
448 * The Guests do not use the sysenter (fast system call) instruction,
Rusty Russellf56a3842007-07-26 10:41:05 -0700449 * because it's hardcoded to enter privilege level 0 and so can't go direct.
450 * It's about twice as fast as the older "int 0x80" system call, so it might
451 * still be worthwhile to handle it in the Switcher and lcall down to the
452 * Guest. The sysenter semantics are hairy tho: search for that keyword in
Rusty Russell2e04ef72009-07-30 16:03:45 -0600453 * entry.S
454:*/
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700455
Rusty Russell2e04ef72009-07-30 16:03:45 -0600456/*H:260
457 * When we make traps go directly into the Guest, we need to make sure
Rusty Russellbff672e2007-07-26 10:41:04 -0700458 * the kernel stack is valid (ie. mapped in the page tables). Otherwise, the
459 * CPU trying to deliver the trap will fault while trying to push the interrupt
460 * words on the stack: this is called a double fault, and it forces us to kill
461 * the Guest.
462 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600463 * Which is deeply unfair, because (literally!) it wasn't the Guests' fault.
464 */
Glauber de Oliveira Costa4665ac8e2008-01-07 11:05:35 -0200465void pin_stack_pages(struct lg_cpu *cpu)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700466{
467 unsigned int i;
468
Rusty Russell2e04ef72009-07-30 16:03:45 -0600469 /*
470 * Depending on the CONFIG_4KSTACKS option, the Guest can have one or
471 * two pages of stack space.
472 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200473 for (i = 0; i < cpu->lg->stack_pages; i++)
Rusty Russell2e04ef72009-07-30 16:03:45 -0600474 /*
475 * The stack grows *upwards*, so the address we're given is the
Rusty Russell8057d762007-08-30 06:35:08 +1000476 * start of the page after the kernel stack. Subtract one to
477 * get back onto the first stack page, and keep subtracting to
Rusty Russell2e04ef72009-07-30 16:03:45 -0600478 * get to the rest of the stack pages.
479 */
Glauber de Oliveira Costa17136082008-01-07 11:05:37 -0200480 pin_page(cpu, cpu->esp1 - 1 - i * PAGE_SIZE);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700481}
482
Rusty Russell2e04ef72009-07-30 16:03:45 -0600483/*
484 * Direct traps also mean that we need to know whenever the Guest wants to use
Rusty Russell9f542882011-07-22 14:39:50 +0930485 * a different kernel stack, so we can change the guest TSS to use that
486 * stack. The TSS entries expect a virtual address, so unlike most addresses
Rusty Russellbff672e2007-07-26 10:41:04 -0700487 * the Guest gives us, the "esp" (stack pointer) value here is virtual, not
488 * physical.
489 *
490 * In Linux each process has its own kernel stack, so this happens a lot: we
Rusty Russell2e04ef72009-07-30 16:03:45 -0600491 * change stacks on each context switch.
492 */
Glauber de Oliveira Costa4665ac8e2008-01-07 11:05:35 -0200493void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700494{
Rusty Russell2e04ef72009-07-30 16:03:45 -0600495 /*
496 * You're not allowed a stack segment with privilege level 0: bad Guest!
497 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700498 if ((seg & 0x3) != GUEST_PL)
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200499 kill_guest(cpu, "bad stack segment %i", seg);
Rusty Russellbff672e2007-07-26 10:41:04 -0700500 /* We only expect one or two stack pages. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700501 if (pages > 2)
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200502 kill_guest(cpu, "bad stack pages %u", pages);
Rusty Russellbff672e2007-07-26 10:41:04 -0700503 /* Save where the stack is, and how many pages */
Glauber de Oliveira Costa4665ac8e2008-01-07 11:05:35 -0200504 cpu->ss1 = seg;
505 cpu->esp1 = esp;
506 cpu->lg->stack_pages = pages;
Rusty Russellbff672e2007-07-26 10:41:04 -0700507 /* Make sure the new stack pages are mapped */
Glauber de Oliveira Costa4665ac8e2008-01-07 11:05:35 -0200508 pin_stack_pages(cpu);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700509}
510
Rusty Russell2e04ef72009-07-30 16:03:45 -0600511/*
512 * All this reference to mapping stacks leads us neatly into the other complex
513 * part of the Host: page table handling.
514 */
Rusty Russellbff672e2007-07-26 10:41:04 -0700515
Rusty Russell2e04ef72009-07-30 16:03:45 -0600516/*H:235
517 * This is the routine which actually checks the Guest's IDT entry and
518 * transfers it into the entry in "struct lguest":
519 */
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200520static void set_trap(struct lg_cpu *cpu, struct desc_struct *trap,
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700521 unsigned int num, u32 lo, u32 hi)
522{
523 u8 type = idt_type(lo, hi);
524
Rusty Russellbff672e2007-07-26 10:41:04 -0700525 /* We zero-out a not-present entry */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700526 if (!idt_present(lo, hi)) {
527 trap->a = trap->b = 0;
528 return;
529 }
530
Rusty Russellbff672e2007-07-26 10:41:04 -0700531 /* We only support interrupt and trap gates. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700532 if (type != 0xE && type != 0xF)
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200533 kill_guest(cpu, "bad IDT type %i", type);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700534
Rusty Russell2e04ef72009-07-30 16:03:45 -0600535 /*
536 * We only copy the handler address, present bit, privilege level and
Rusty Russellbff672e2007-07-26 10:41:04 -0700537 * type. The privilege level controls where the trap can be triggered
538 * manually with an "int" instruction. This is usually GUEST_PL,
Rusty Russell2e04ef72009-07-30 16:03:45 -0600539 * except for system calls which userspace can use.
540 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700541 trap->a = ((__KERNEL_CS|GUEST_PL)<<16) | (lo&0x0000FFFF);
542 trap->b = (hi&0xFFFFEF00);
543}
544
Rusty Russell2e04ef72009-07-30 16:03:45 -0600545/*H:230
546 * While we're here, dealing with delivering traps and interrupts to the
Rusty Russellbff672e2007-07-26 10:41:04 -0700547 * Guest, we might as well complete the picture: how the Guest tells us where
548 * it wants them to go. This would be simple, except making traps fast
549 * requires some tricks.
550 *
551 * We saw the Guest setting Interrupt Descriptor Table (IDT) entries with the
Rusty Russell2e04ef72009-07-30 16:03:45 -0600552 * LHCALL_LOAD_IDT_ENTRY hypercall before: that comes here.
553 */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200554void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int num, u32 lo, u32 hi)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700555{
Rusty Russell2e04ef72009-07-30 16:03:45 -0600556 /*
557 * Guest never handles: NMI, doublefault, spurious interrupt or
558 * hypercall. We ignore when it tries to set them.
559 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700560 if (num == 2 || num == 8 || num == 15 || num == LGUEST_TRAP_ENTRY)
561 return;
562
Rusty Russell2e04ef72009-07-30 16:03:45 -0600563 /*
564 * Mark the IDT as changed: next time the Guest runs we'll know we have
565 * to copy this again.
566 */
Glauber de Oliveira Costaae3749d2008-01-17 19:14:46 -0200567 cpu->changed |= CHANGED_IDT;
Rusty Russellbff672e2007-07-26 10:41:04 -0700568
Rusty Russell56adbe92007-10-22 11:03:28 +1000569 /* Check that the Guest doesn't try to step outside the bounds. */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200570 if (num >= ARRAY_SIZE(cpu->arch.idt))
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200571 kill_guest(cpu, "Setting idt entry %u", num);
Rusty Russell56adbe92007-10-22 11:03:28 +1000572 else
Glauber de Oliveira Costa382ac6b2008-01-17 19:19:42 -0200573 set_trap(cpu, &cpu->arch.idt[num], num, lo, hi);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700574}
575
Rusty Russell2e04ef72009-07-30 16:03:45 -0600576/*
577 * The default entry for each interrupt points into the Switcher routines which
Rusty Russellbff672e2007-07-26 10:41:04 -0700578 * simply return to the Host. The run_guest() loop will then call
Rusty Russell2e04ef72009-07-30 16:03:45 -0600579 * deliver_trap() to bounce it back into the Guest.
580 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700581static void default_idt_entry(struct desc_struct *idt,
582 int trap,
Rusty Russell0c120912008-07-29 09:58:31 -0500583 const unsigned long handler,
584 const struct desc_struct *base)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700585{
Rusty Russellbff672e2007-07-26 10:41:04 -0700586 /* A present interrupt gate. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700587 u32 flags = 0x8e00;
588
Rusty Russell2e04ef72009-07-30 16:03:45 -0600589 /*
590 * Set the privilege level on the entry for the hypercall: this allows
591 * the Guest to use the "int" instruction to trigger it.
592 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700593 if (trap == LGUEST_TRAP_ENTRY)
594 flags |= (GUEST_PL << 13);
Rusty Russell0c120912008-07-29 09:58:31 -0500595 else if (base)
Rusty Russell2e04ef72009-07-30 16:03:45 -0600596 /*
597 * Copy privilege level from what Guest asked for. This allows
598 * debug (int 3) traps from Guest userspace, for example.
599 */
Rusty Russell0c120912008-07-29 09:58:31 -0500600 flags |= (base->b & 0x6000);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700601
Rusty Russellbff672e2007-07-26 10:41:04 -0700602 /* Now pack it into the IDT entry in its weird format. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700603 idt->a = (LGUEST_CS<<16) | (handler&0x0000FFFF);
604 idt->b = (handler&0xFFFF0000) | flags;
605}
606
Rusty Russellbff672e2007-07-26 10:41:04 -0700607/* When the Guest first starts, we put default entries into the IDT. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700608void setup_default_idt_entries(struct lguest_ro_state *state,
609 const unsigned long *def)
610{
611 unsigned int i;
612
613 for (i = 0; i < ARRAY_SIZE(state->guest_idt); i++)
Rusty Russell0c120912008-07-29 09:58:31 -0500614 default_idt_entry(&state->guest_idt[i], i, def[i], NULL);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700615}
616
Rusty Russell2e04ef72009-07-30 16:03:45 -0600617/*H:240
618 * We don't use the IDT entries in the "struct lguest" directly, instead
Rusty Russellbff672e2007-07-26 10:41:04 -0700619 * we copy them into the IDT which we've set up for Guests on this CPU, just
Rusty Russell2e04ef72009-07-30 16:03:45 -0600620 * before we run the Guest. This routine does that copy.
621 */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200622void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt,
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700623 const unsigned long *def)
624{
625 unsigned int i;
626
Rusty Russell2e04ef72009-07-30 16:03:45 -0600627 /*
628 * We can simply copy the direct traps, otherwise we use the default
629 * ones in the Switcher: they will return to the Host.
630 */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200631 for (i = 0; i < ARRAY_SIZE(cpu->arch.idt); i++) {
Rusty Russell0c120912008-07-29 09:58:31 -0500632 const struct desc_struct *gidt = &cpu->arch.idt[i];
633
Rusty Russell56adbe92007-10-22 11:03:28 +1000634 /* If no Guest can ever override this trap, leave it alone. */
635 if (!direct_trap(i))
636 continue;
637
Rusty Russell2e04ef72009-07-30 16:03:45 -0600638 /*
639 * Only trap gates (type 15) can go direct to the Guest.
Rusty Russell56adbe92007-10-22 11:03:28 +1000640 * Interrupt gates (type 14) disable interrupts as they are
641 * entered, which we never let the Guest do. Not present
Rusty Russell0c120912008-07-29 09:58:31 -0500642 * entries (type 0x0) also can't go direct, of course.
643 *
644 * If it can't go direct, we still need to copy the priv. level:
645 * they might want to give userspace access to a software
Rusty Russell2e04ef72009-07-30 16:03:45 -0600646 * interrupt.
647 */
Rusty Russell0c120912008-07-29 09:58:31 -0500648 if (idt_type(gidt->a, gidt->b) == 0xF)
649 idt[i] = *gidt;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700650 else
Rusty Russell0c120912008-07-29 09:58:31 -0500651 default_idt_entry(&idt[i], i, def[i], gidt);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700652 }
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700653}
654
Rusty Russelle1e72962007-10-25 15:02:50 +1000655/*H:200
656 * The Guest Clock.
657 *
658 * There are two sources of virtual interrupts. We saw one in lguest_user.c:
659 * the Launcher sending interrupts for virtual devices. The other is the Guest
660 * timer interrupt.
661 *
662 * The Guest uses the LHCALL_SET_CLOCKEVENT hypercall to tell us how long to
663 * the next timer interrupt (in nanoseconds). We use the high-resolution timer
664 * infrastructure to set a callback at that time.
665 *
Rusty Russell2e04ef72009-07-30 16:03:45 -0600666 * 0 means "turn off the clock".
667 */
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200668void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700669{
670 ktime_t expires;
671
672 if (unlikely(delta == 0)) {
673 /* Clock event device is shutting down. */
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200674 hrtimer_cancel(&cpu->hrt);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700675 return;
676 }
677
Rusty Russell2e04ef72009-07-30 16:03:45 -0600678 /*
679 * We use wallclock time here, so the Guest might not be running for
Rusty Russelle1e72962007-10-25 15:02:50 +1000680 * all the time between now and the timer interrupt it asked for. This
Rusty Russell2e04ef72009-07-30 16:03:45 -0600681 * is almost always the right thing to do.
682 */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700683 expires = ktime_add_ns(ktime_get_real(), delta);
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200684 hrtimer_start(&cpu->hrt, expires, HRTIMER_MODE_ABS);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700685}
686
Rusty Russelle1e72962007-10-25 15:02:50 +1000687/* This is the function called when the Guest's timer expires. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700688static enum hrtimer_restart clockdev_fn(struct hrtimer *timer)
689{
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200690 struct lg_cpu *cpu = container_of(timer, struct lg_cpu, hrt);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700691
Rusty Russelle1e72962007-10-25 15:02:50 +1000692 /* Remember the first interrupt is the timer interrupt. */
Rusty Russell9f155a92009-06-12 22:27:08 -0600693 set_interrupt(cpu, 0);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700694 return HRTIMER_NORESTART;
695}
696
Rusty Russelle1e72962007-10-25 15:02:50 +1000697/* This sets up the timer for this Guest. */
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200698void init_clockdev(struct lg_cpu *cpu)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700699{
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200700 hrtimer_init(&cpu->hrt, CLOCK_REALTIME, HRTIMER_MODE_ABS);
701 cpu->hrt.function = clockdev_fn;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700702}