blob: 9c1c479e8c624b99ec541f0a89dae6c20723581c [file] [log] [blame]
Rusty Russellf938d2c2007-07-26 10:41:02 -07001/*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 Russelld7e28ff2007-07-19 01:49:23 -070014#include <linux/uaccess.h>
Rusty Russellc18acd72007-10-22 11:03:35 +100015#include <linux/interrupt.h>
16#include <linux/module.h>
Rusty Russelld7e28ff2007-07-19 01:49:23 -070017#include "lg.h"
18
Rusty Russellc18acd72007-10-22 11:03:35 +100019/* Allow Guests to use a non-128 (ie. non-Linux) syscall trap. */
20static unsigned int syscall_vector = SYSCALL_VECTOR;
21module_param(syscall_vector, uint, 0444);
22
Rusty Russellbff672e2007-07-26 10:41:04 -070023/* The address of the interrupt handler is split into two bits: */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070024static unsigned long idt_address(u32 lo, u32 hi)
25{
26 return (lo & 0x0000FFFF) | (hi & 0xFFFF0000);
27}
28
Rusty Russellbff672e2007-07-26 10:41:04 -070029/* The "type" of the interrupt handler is a 4 bit field: we only support a
30 * couple of types. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070031static int idt_type(u32 lo, u32 hi)
32{
33 return (hi >> 8) & 0xF;
34}
35
Rusty Russellbff672e2007-07-26 10:41:04 -070036/* An IDT entry can't be used unless the "present" bit is set. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070037static int idt_present(u32 lo, u32 hi)
38{
39 return (hi & 0x8000);
40}
41
Rusty Russellbff672e2007-07-26 10:41:04 -070042/* 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. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070044static void push_guest_stack(struct lguest *lg, unsigned long *gstack, u32 val)
45{
Rusty Russellbff672e2007-07-26 10:41:04 -070046 /* Stack grows upwards: move stack then write value. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -070047 *gstack -= 4;
Rusty Russell2d37f942007-10-22 11:24:24 +100048 lgwrite(lg, *gstack, u32, val);
Rusty Russelld7e28ff2007-07-19 01:49:23 -070049}
50
Rusty Russellbff672e2007-07-26 10:41:04 -070051/*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). */
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -020063static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi, int has_err)
Rusty Russelld7e28ff2007-07-19 01:49:23 -070064{
Rusty Russell47436aa2007-10-22 11:03:36 +100065 unsigned long gstack, origstack;
Rusty Russelld7e28ff2007-07-19 01:49:23 -070066 u32 eflags, ss, irq_enable;
Rusty Russell47436aa2007-10-22 11:03:36 +100067 unsigned long virtstack;
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -020068 struct lguest *lg = cpu->lg;
Rusty Russelld7e28ff2007-07-19 01:49:23 -070069
Rusty Russellbff672e2007-07-26 10:41:04 -070070 /* 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 Costaa53a35a2008-01-07 11:05:32 -020073 if ((cpu->regs->ss&0x3) != GUEST_PL) {
Rusty Russellbff672e2007-07-26 10:41:04 -070074 /* The Guest told us their kernel stack with the SET_STACK
75 * hypercall: both the virtual address and the segment */
Rusty Russell47436aa2007-10-22 11:03:36 +100076 virtstack = lg->esp1;
Rusty Russelld7e28ff2007-07-19 01:49:23 -070077 ss = lg->ss1;
Rusty Russell47436aa2007-10-22 11:03:36 +100078
79 origstack = gstack = guest_pa(lg, virtstack);
Rusty Russellbff672e2007-07-26 10:41:04 -070080 /* 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 Costaa53a35a2008-01-07 11:05:32 -020084 push_guest_stack(lg, &gstack, cpu->regs->ss);
85 push_guest_stack(lg, &gstack, cpu->regs->esp);
Rusty Russelld7e28ff2007-07-19 01:49:23 -070086 } else {
Rusty Russellbff672e2007-07-26 10:41:04 -070087 /* We're staying on the same Guest (kernel) stack. */
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -020088 virtstack = cpu->regs->esp;
89 ss = cpu->regs->ss;
Rusty Russell47436aa2007-10-22 11:03:36 +100090
91 origstack = gstack = guest_pa(lg, virtstack);
Rusty Russelld7e28ff2007-07-19 01:49:23 -070092 }
93
Rusty Russellbff672e2007-07-26 10:41:04 -070094 /* 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 Russelle1e72962007-10-25 15:02:50 +100096 * Guest's "irq_enabled" field into the eflags word: we saw the Guest
97 * copy it back in "lguest_iret". */
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -020098 eflags = cpu->regs->eflags;
Rusty Russelle5faff42007-07-20 22:11:13 +100099 if (get_user(irq_enable, &lg->lguest_data->irq_enabled) == 0
100 && !(irq_enable & X86_EFLAGS_IF))
101 eflags &= ~X86_EFLAGS_IF;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700102
Rusty Russellbff672e2007-07-26 10:41:04 -0700103 /* 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. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700106 push_guest_stack(lg, &gstack, eflags);
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -0200107 push_guest_stack(lg, &gstack, cpu->regs->cs);
108 push_guest_stack(lg, &gstack, cpu->regs->eip);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700109
Rusty Russellbff672e2007-07-26 10:41:04 -0700110 /* For the six traps which supply an error code, we push that, too. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700111 if (has_err)
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -0200112 push_guest_stack(lg, &gstack, cpu->regs->errcode);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700113
Rusty Russellbff672e2007-07-26 10:41:04 -0700114 /* Now we've pushed all the old state, we change the stack, the code
115 * segment and the address to execute. */
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -0200116 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 Russelld7e28ff2007-07-19 01:49:23 -0700120
Rusty Russellbff672e2007-07-26 10:41:04 -0700121 /* There are two kinds of interrupt handlers: 0xE is an "interrupt
122 * gate" which expects interrupts to be disabled on entry. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700123 if (idt_type(lo, hi) == 0xE)
124 if (put_user(0, &lg->lguest_data->irq_enabled))
125 kill_guest(lg, "Disabling interrupts");
126}
127
Rusty Russelle1e72962007-10-25 15:02:50 +1000128/*H:205
Rusty Russellbff672e2007-07-26 10:41:04 -0700129 * Virtual Interrupts.
130 *
131 * maybe_do_interrupt() gets called before every entry to the Guest, to see if
132 * we should divert the Guest to running an interrupt handler. */
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -0200133void maybe_do_interrupt(struct lg_cpu *cpu)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700134{
135 unsigned int irq;
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -0200136 struct lguest *lg = cpu->lg;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700137 DECLARE_BITMAP(blk, LGUEST_IRQS);
138 struct desc_struct *idt;
139
Rusty Russellbff672e2007-07-26 10:41:04 -0700140 /* If the Guest hasn't even initialized yet, we can do nothing. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700141 if (!lg->lguest_data)
142 return;
143
Rusty Russellbff672e2007-07-26 10:41:04 -0700144 /* Take our "irqs_pending" array and remove any interrupts the Guest
145 * wants blocked: the result ends up in "blk". */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700146 if (copy_from_user(&blk, lg->lguest_data->blocked_interrupts,
147 sizeof(blk)))
148 return;
149
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -0200150 bitmap_andnot(blk, cpu->irqs_pending, blk, LGUEST_IRQS);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700151
Rusty Russellbff672e2007-07-26 10:41:04 -0700152 /* Find the first interrupt. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700153 irq = find_first_bit(blk, LGUEST_IRQS);
Rusty Russellbff672e2007-07-26 10:41:04 -0700154 /* None? Nothing to do */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700155 if (irq >= LGUEST_IRQS)
156 return;
157
Rusty Russellbff672e2007-07-26 10:41:04 -0700158 /* They may be in the middle of an iret, where they asked us never to
159 * deliver interrupts. */
Glauber de Oliveira Costaa53a35a2008-01-07 11:05:32 -0200160 if (cpu->regs->eip >= lg->noirq_start && cpu->regs->eip < lg->noirq_end)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700161 return;
162
Rusty Russellbff672e2007-07-26 10:41:04 -0700163 /* If they're halted, interrupts restart them. */
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -0200164 if (cpu->halted) {
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700165 /* Re-enable interrupts. */
166 if (put_user(X86_EFLAGS_IF, &lg->lguest_data->irq_enabled))
167 kill_guest(lg, "Re-enabling interrupts");
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -0200168 cpu->halted = 0;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700169 } else {
Rusty Russellbff672e2007-07-26 10:41:04 -0700170 /* Otherwise we check if they have interrupts disabled. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700171 u32 irq_enabled;
172 if (get_user(irq_enabled, &lg->lguest_data->irq_enabled))
173 irq_enabled = 0;
174 if (!irq_enabled)
175 return;
176 }
177
Rusty Russellbff672e2007-07-26 10:41:04 -0700178 /* Look at the IDT entry the Guest gave us for this interrupt. The
179 * first 32 (FIRST_EXTERNAL_VECTOR) entries are for traps, so we skip
180 * over them. */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200181 idt = &cpu->arch.idt[FIRST_EXTERNAL_VECTOR+irq];
Rusty Russellbff672e2007-07-26 10:41:04 -0700182 /* If they don't have a handler (yet?), we just ignore it */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700183 if (idt_present(idt->a, idt->b)) {
Rusty Russellbff672e2007-07-26 10:41:04 -0700184 /* OK, mark it no longer pending and deliver it. */
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -0200185 clear_bit(irq, cpu->irqs_pending);
Rusty Russellbff672e2007-07-26 10:41:04 -0700186 /* set_guest_interrupt() takes the interrupt descriptor and a
187 * flag to say whether this interrupt pushes an error code onto
188 * the stack as well: virtual interrupts never do. */
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -0200189 set_guest_interrupt(cpu, idt->a, idt->b, 0);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700190 }
Rusty Russell6c8dca52007-07-27 13:42:52 +1000191
192 /* Every time we deliver an interrupt, we update the timestamp in the
193 * Guest's lguest_data struct. It would be better for the Guest if we
194 * did this more often, but it can actually be quite slow: doing it
195 * here is a compromise which means at least it gets updated every
196 * timer interrupt. */
197 write_timestamp(lg);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700198}
Rusty Russellc18acd72007-10-22 11:03:35 +1000199/*:*/
200
201/* Linux uses trap 128 for system calls. Plan9 uses 64, and Ron Minnich sent
202 * me a patch, so we support that too. It'd be a big step for lguest if half
203 * the Plan 9 user base were to start using it.
204 *
205 * Actually now I think of it, it's possible that Ron *is* half the Plan 9
206 * userbase. Oh well. */
207static bool could_be_syscall(unsigned int num)
208{
209 /* Normal Linux SYSCALL_VECTOR or reserved vector? */
210 return num == SYSCALL_VECTOR || num == syscall_vector;
211}
212
213/* The syscall vector it wants must be unused by Host. */
214bool check_syscall_vector(struct lguest *lg)
215{
216 u32 vector;
217
218 if (get_user(vector, &lg->lguest_data->syscall_vec))
219 return false;
220
221 return could_be_syscall(vector);
222}
223
224int init_interrupts(void)
225{
226 /* If they want some strange system call vector, reserve it now */
227 if (syscall_vector != SYSCALL_VECTOR
228 && test_and_set_bit(syscall_vector, used_vectors)) {
229 printk("lg: couldn't reserve syscall %u\n", syscall_vector);
230 return -EBUSY;
231 }
232 return 0;
233}
234
235void free_interrupts(void)
236{
237 if (syscall_vector != SYSCALL_VECTOR)
238 clear_bit(syscall_vector, used_vectors);
239}
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700240
Rusty Russellbff672e2007-07-26 10:41:04 -0700241/*H:220 Now we've got the routines to deliver interrupts, delivering traps
242 * like page fault is easy. The only trick is that Intel decided that some
243 * traps should have error codes: */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700244static int has_err(unsigned int trap)
245{
246 return (trap == 8 || (trap >= 10 && trap <= 14) || trap == 17);
247}
248
Rusty Russellbff672e2007-07-26 10:41:04 -0700249/* deliver_trap() returns true if it could deliver the trap. */
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -0200250int deliver_trap(struct lg_cpu *cpu, unsigned int num)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700251{
Rusty Russell0d027c02007-08-09 20:57:13 +1000252 /* Trap numbers are always 8 bit, but we set an impossible trap number
253 * for traps inside the Switcher, so check that here. */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200254 if (num >= ARRAY_SIZE(cpu->arch.idt))
Rusty Russell0d027c02007-08-09 20:57:13 +1000255 return 0;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700256
Rusty Russellbff672e2007-07-26 10:41:04 -0700257 /* Early on the Guest hasn't set the IDT entries (or maybe it put a
258 * bogus one in): if we fail here, the Guest will be killed. */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200259 if (!idt_present(cpu->arch.idt[num].a, cpu->arch.idt[num].b))
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700260 return 0;
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200261 set_guest_interrupt(cpu, cpu->arch.idt[num].a,
262 cpu->arch.idt[num].b, has_err(num));
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700263 return 1;
264}
265
Rusty Russellbff672e2007-07-26 10:41:04 -0700266/*H:250 Here's the hard part: returning to the Host every time a trap happens
267 * and then calling deliver_trap() and re-entering the Guest is slow.
Rusty Russelle1e72962007-10-25 15:02:50 +1000268 * Particularly because Guest userspace system calls are traps (usually trap
269 * 128).
Rusty Russellbff672e2007-07-26 10:41:04 -0700270 *
271 * So we'd like to set up the IDT to tell the CPU to deliver traps directly
272 * into the Guest. This is possible, but the complexities cause the size of
273 * this file to double! However, 150 lines of code is worth writing for taking
274 * system calls down from 1750ns to 270ns. Plus, if lguest didn't do it, all
Rusty Russelle1e72962007-10-25 15:02:50 +1000275 * the other hypervisors would beat it up at lunchtime.
Rusty Russellbff672e2007-07-26 10:41:04 -0700276 *
Rusty Russell56adbe92007-10-22 11:03:28 +1000277 * This routine indicates if a particular trap number could be delivered
278 * directly. */
279static int direct_trap(unsigned int num)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700280{
Rusty Russellbff672e2007-07-26 10:41:04 -0700281 /* Hardware interrupts don't go to the Guest at all (except system
282 * call). */
Rusty Russellc18acd72007-10-22 11:03:35 +1000283 if (num >= FIRST_EXTERNAL_VECTOR && !could_be_syscall(num))
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700284 return 0;
285
Rusty Russellbff672e2007-07-26 10:41:04 -0700286 /* The Host needs to see page faults (for shadow paging and to save the
287 * fault address), general protection faults (in/out emulation) and
288 * device not available (TS handling), and of course, the hypercall
289 * trap. */
Rusty Russell56adbe92007-10-22 11:03:28 +1000290 return num != 14 && num != 13 && num != 7 && num != LGUEST_TRAP_ENTRY;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700291}
Rusty Russellf56a3842007-07-26 10:41:05 -0700292/*:*/
293
294/*M:005 The Guest has the ability to turn its interrupt gates into trap gates,
295 * if it is careful. The Host will let trap gates can go directly to the
296 * Guest, but the Guest needs the interrupts atomically disabled for an
297 * interrupt gate. It can do this by pointing the trap gate at instructions
298 * within noirq_start and noirq_end, where it can safely disable interrupts. */
299
300/*M:006 The Guests do not use the sysenter (fast system call) instruction,
301 * because it's hardcoded to enter privilege level 0 and so can't go direct.
302 * It's about twice as fast as the older "int 0x80" system call, so it might
303 * still be worthwhile to handle it in the Switcher and lcall down to the
304 * Guest. The sysenter semantics are hairy tho: search for that keyword in
305 * entry.S :*/
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700306
Rusty Russellbff672e2007-07-26 10:41:04 -0700307/*H:260 When we make traps go directly into the Guest, we need to make sure
308 * the kernel stack is valid (ie. mapped in the page tables). Otherwise, the
309 * CPU trying to deliver the trap will fault while trying to push the interrupt
310 * words on the stack: this is called a double fault, and it forces us to kill
311 * the Guest.
312 *
313 * Which is deeply unfair, because (literally!) it wasn't the Guests' fault. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700314void pin_stack_pages(struct lguest *lg)
315{
316 unsigned int i;
317
Rusty Russellbff672e2007-07-26 10:41:04 -0700318 /* Depending on the CONFIG_4KSTACKS option, the Guest can have one or
319 * two pages of stack space. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700320 for (i = 0; i < lg->stack_pages; i++)
Rusty Russell8057d762007-08-30 06:35:08 +1000321 /* The stack grows *upwards*, so the address we're given is the
322 * start of the page after the kernel stack. Subtract one to
323 * get back onto the first stack page, and keep subtracting to
324 * get to the rest of the stack pages. */
325 pin_page(lg, lg->esp1 - 1 - i * PAGE_SIZE);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700326}
327
Rusty Russellbff672e2007-07-26 10:41:04 -0700328/* Direct traps also mean that we need to know whenever the Guest wants to use
329 * a different kernel stack, so we can change the IDT entries to use that
330 * stack. The IDT entries expect a virtual address, so unlike most addresses
331 * the Guest gives us, the "esp" (stack pointer) value here is virtual, not
332 * physical.
333 *
334 * In Linux each process has its own kernel stack, so this happens a lot: we
335 * change stacks on each context switch. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700336void guest_set_stack(struct lguest *lg, u32 seg, u32 esp, unsigned int pages)
337{
Rusty Russelle1e72962007-10-25 15:02:50 +1000338 /* You are not allowed have a stack segment with privilege level 0: bad
Rusty Russellbff672e2007-07-26 10:41:04 -0700339 * Guest! */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700340 if ((seg & 0x3) != GUEST_PL)
341 kill_guest(lg, "bad stack segment %i", seg);
Rusty Russellbff672e2007-07-26 10:41:04 -0700342 /* We only expect one or two stack pages. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700343 if (pages > 2)
344 kill_guest(lg, "bad stack pages %u", pages);
Rusty Russellbff672e2007-07-26 10:41:04 -0700345 /* Save where the stack is, and how many pages */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700346 lg->ss1 = seg;
347 lg->esp1 = esp;
348 lg->stack_pages = pages;
Rusty Russellbff672e2007-07-26 10:41:04 -0700349 /* Make sure the new stack pages are mapped */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700350 pin_stack_pages(lg);
351}
352
Rusty Russellbff672e2007-07-26 10:41:04 -0700353/* All this reference to mapping stacks leads us neatly into the other complex
354 * part of the Host: page table handling. */
355
356/*H:235 This is the routine which actually checks the Guest's IDT entry and
Rusty Russelle1e72962007-10-25 15:02:50 +1000357 * transfers it into the entry in "struct lguest": */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700358static void set_trap(struct lguest *lg, struct desc_struct *trap,
359 unsigned int num, u32 lo, u32 hi)
360{
361 u8 type = idt_type(lo, hi);
362
Rusty Russellbff672e2007-07-26 10:41:04 -0700363 /* We zero-out a not-present entry */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700364 if (!idt_present(lo, hi)) {
365 trap->a = trap->b = 0;
366 return;
367 }
368
Rusty Russellbff672e2007-07-26 10:41:04 -0700369 /* We only support interrupt and trap gates. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700370 if (type != 0xE && type != 0xF)
371 kill_guest(lg, "bad IDT type %i", type);
372
Rusty Russellbff672e2007-07-26 10:41:04 -0700373 /* We only copy the handler address, present bit, privilege level and
374 * type. The privilege level controls where the trap can be triggered
375 * manually with an "int" instruction. This is usually GUEST_PL,
376 * except for system calls which userspace can use. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700377 trap->a = ((__KERNEL_CS|GUEST_PL)<<16) | (lo&0x0000FFFF);
378 trap->b = (hi&0xFFFFEF00);
379}
380
Rusty Russellbff672e2007-07-26 10:41:04 -0700381/*H:230 While we're here, dealing with delivering traps and interrupts to the
382 * Guest, we might as well complete the picture: how the Guest tells us where
383 * it wants them to go. This would be simple, except making traps fast
384 * requires some tricks.
385 *
386 * We saw the Guest setting Interrupt Descriptor Table (IDT) entries with the
387 * LHCALL_LOAD_IDT_ENTRY hypercall before: that comes here. */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200388void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int num, u32 lo, u32 hi)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700389{
Rusty Russellbff672e2007-07-26 10:41:04 -0700390 /* Guest never handles: NMI, doublefault, spurious interrupt or
391 * hypercall. We ignore when it tries to set them. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700392 if (num == 2 || num == 8 || num == 15 || num == LGUEST_TRAP_ENTRY)
393 return;
394
Rusty Russellbff672e2007-07-26 10:41:04 -0700395 /* Mark the IDT as changed: next time the Guest runs we'll know we have
396 * to copy this again. */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200397 cpu->lg->changed |= CHANGED_IDT;
Rusty Russellbff672e2007-07-26 10:41:04 -0700398
Rusty Russell56adbe92007-10-22 11:03:28 +1000399 /* Check that the Guest doesn't try to step outside the bounds. */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200400 if (num >= ARRAY_SIZE(cpu->arch.idt))
401 kill_guest(cpu->lg, "Setting idt entry %u", num);
Rusty Russell56adbe92007-10-22 11:03:28 +1000402 else
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200403 set_trap(cpu->lg, &cpu->arch.idt[num], num, lo, hi);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700404}
405
Rusty Russellbff672e2007-07-26 10:41:04 -0700406/* The default entry for each interrupt points into the Switcher routines which
407 * simply return to the Host. The run_guest() loop will then call
408 * deliver_trap() to bounce it back into the Guest. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700409static void default_idt_entry(struct desc_struct *idt,
410 int trap,
411 const unsigned long handler)
412{
Rusty Russellbff672e2007-07-26 10:41:04 -0700413 /* A present interrupt gate. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700414 u32 flags = 0x8e00;
415
Rusty Russellbff672e2007-07-26 10:41:04 -0700416 /* Set the privilege level on the entry for the hypercall: this allows
417 * the Guest to use the "int" instruction to trigger it. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700418 if (trap == LGUEST_TRAP_ENTRY)
419 flags |= (GUEST_PL << 13);
420
Rusty Russellbff672e2007-07-26 10:41:04 -0700421 /* Now pack it into the IDT entry in its weird format. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700422 idt->a = (LGUEST_CS<<16) | (handler&0x0000FFFF);
423 idt->b = (handler&0xFFFF0000) | flags;
424}
425
Rusty Russellbff672e2007-07-26 10:41:04 -0700426/* When the Guest first starts, we put default entries into the IDT. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700427void setup_default_idt_entries(struct lguest_ro_state *state,
428 const unsigned long *def)
429{
430 unsigned int i;
431
432 for (i = 0; i < ARRAY_SIZE(state->guest_idt); i++)
433 default_idt_entry(&state->guest_idt[i], i, def[i]);
434}
435
Rusty Russellbff672e2007-07-26 10:41:04 -0700436/*H:240 We don't use the IDT entries in the "struct lguest" directly, instead
437 * we copy them into the IDT which we've set up for Guests on this CPU, just
438 * before we run the Guest. This routine does that copy. */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200439void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt,
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700440 const unsigned long *def)
441{
442 unsigned int i;
443
Rusty Russellbff672e2007-07-26 10:41:04 -0700444 /* We can simply copy the direct traps, otherwise we use the default
445 * ones in the Switcher: they will return to the Host. */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200446 for (i = 0; i < ARRAY_SIZE(cpu->arch.idt); i++) {
Rusty Russell56adbe92007-10-22 11:03:28 +1000447 /* If no Guest can ever override this trap, leave it alone. */
448 if (!direct_trap(i))
449 continue;
450
451 /* Only trap gates (type 15) can go direct to the Guest.
452 * Interrupt gates (type 14) disable interrupts as they are
453 * entered, which we never let the Guest do. Not present
454 * entries (type 0x0) also can't go direct, of course. */
Glauber de Oliveira Costafc708b32008-01-07 11:05:33 -0200455 if (idt_type(cpu->arch.idt[i].a, cpu->arch.idt[i].b) == 0xF)
456 idt[i] = cpu->arch.idt[i];
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700457 else
Rusty Russell56adbe92007-10-22 11:03:28 +1000458 /* Reset it to the default. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700459 default_idt_entry(&idt[i], i, def[i]);
460 }
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700461}
462
Rusty Russelle1e72962007-10-25 15:02:50 +1000463/*H:200
464 * The Guest Clock.
465 *
466 * There are two sources of virtual interrupts. We saw one in lguest_user.c:
467 * the Launcher sending interrupts for virtual devices. The other is the Guest
468 * timer interrupt.
469 *
470 * The Guest uses the LHCALL_SET_CLOCKEVENT hypercall to tell us how long to
471 * the next timer interrupt (in nanoseconds). We use the high-resolution timer
472 * infrastructure to set a callback at that time.
473 *
474 * 0 means "turn off the clock". */
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200475void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700476{
477 ktime_t expires;
478
479 if (unlikely(delta == 0)) {
480 /* Clock event device is shutting down. */
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200481 hrtimer_cancel(&cpu->hrt);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700482 return;
483 }
484
Rusty Russelle1e72962007-10-25 15:02:50 +1000485 /* We use wallclock time here, so the Guest might not be running for
486 * all the time between now and the timer interrupt it asked for. This
487 * is almost always the right thing to do. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700488 expires = ktime_add_ns(ktime_get_real(), delta);
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200489 hrtimer_start(&cpu->hrt, expires, HRTIMER_MODE_ABS);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700490}
491
Rusty Russelle1e72962007-10-25 15:02:50 +1000492/* This is the function called when the Guest's timer expires. */
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700493static enum hrtimer_restart clockdev_fn(struct hrtimer *timer)
494{
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200495 struct lg_cpu *cpu = container_of(timer, struct lg_cpu, hrt);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700496
Rusty Russelle1e72962007-10-25 15:02:50 +1000497 /* Remember the first interrupt is the timer interrupt. */
Glauber de Oliveira Costa177e4492008-01-07 11:05:29 -0200498 set_bit(0, cpu->irqs_pending);
Rusty Russelle1e72962007-10-25 15:02:50 +1000499 /* If the Guest is actually stopped, we need to wake it up. */
Glauber de Oliveira Costa66686c22008-01-07 11:05:34 -0200500 if (cpu->halted)
501 wake_up_process(cpu->tsk);
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700502 return HRTIMER_NORESTART;
503}
504
Rusty Russelle1e72962007-10-25 15:02:50 +1000505/* This sets up the timer for this Guest. */
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200506void init_clockdev(struct lg_cpu *cpu)
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700507{
Glauber de Oliveira Costaad8d8f32008-01-07 11:05:28 -0200508 hrtimer_init(&cpu->hrt, CLOCK_REALTIME, HRTIMER_MODE_ABS);
509 cpu->hrt.function = clockdev_fn;
Rusty Russelld7e28ff2007-07-19 01:49:23 -0700510}