Commit | Line | Data |
---|---|---|
f938d2c8 RR |
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. :*/ | |
d7e28ffe | 14 | #include <linux/uaccess.h> |
c18acd73 RR |
15 | #include <linux/interrupt.h> |
16 | #include <linux/module.h> | |
d7e28ffe RR |
17 | #include "lg.h" |
18 | ||
c18acd73 RR |
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 | ||
bff672e6 | 23 | /* The address of the interrupt handler is split into two bits: */ |
d7e28ffe RR |
24 | static unsigned long idt_address(u32 lo, u32 hi) |
25 | { | |
26 | return (lo & 0x0000FFFF) | (hi & 0xFFFF0000); | |
27 | } | |
28 | ||
bff672e6 RR |
29 | /* The "type" of the interrupt handler is a 4 bit field: we only support a |
30 | * couple of types. */ | |
d7e28ffe RR |
31 | static int idt_type(u32 lo, u32 hi) |
32 | { | |
33 | return (hi >> 8) & 0xF; | |
34 | } | |
35 | ||
bff672e6 | 36 | /* An IDT entry can't be used unless the "present" bit is set. */ |
d7e28ffe RR |
37 | static int idt_present(u32 lo, u32 hi) |
38 | { | |
39 | return (hi & 0x8000); | |
40 | } | |
41 | ||
bff672e6 RR |
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. */ | |
382ac6b3 | 44 | static void push_guest_stack(struct lg_cpu *cpu, unsigned long *gstack, u32 val) |
d7e28ffe | 45 | { |
bff672e6 | 46 | /* Stack grows upwards: move stack then write value. */ |
d7e28ffe | 47 | *gstack -= 4; |
382ac6b3 | 48 | lgwrite(cpu, *gstack, u32, val); |
d7e28ffe RR |
49 | } |
50 | ||
bff672e6 RR |
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). */ | |
177e449d | 63 | static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi, int has_err) |
d7e28ffe | 64 | { |
47436aa4 | 65 | unsigned long gstack, origstack; |
d7e28ffe | 66 | u32 eflags, ss, irq_enable; |
47436aa4 | 67 | unsigned long virtstack; |
d7e28ffe | 68 | |
bff672e6 RR |
69 | /* There are two cases for interrupts: one where the Guest is already |
70 | * in the kernel, and a more complex one where the Guest is in | |
71 | * userspace. We check the privilege level to find out. */ | |
a53a35a8 | 72 | if ((cpu->regs->ss&0x3) != GUEST_PL) { |
bff672e6 RR |
73 | /* The Guest told us their kernel stack with the SET_STACK |
74 | * hypercall: both the virtual address and the segment */ | |
4665ac8e GOC |
75 | virtstack = cpu->esp1; |
76 | ss = cpu->ss1; | |
47436aa4 | 77 | |
1713608f | 78 | origstack = gstack = guest_pa(cpu, virtstack); |
bff672e6 RR |
79 | /* We push the old stack segment and pointer onto the new |
80 | * stack: when the Guest does an "iret" back from the interrupt | |
81 | * handler the CPU will notice they're dropping privilege | |
82 | * levels and expect these here. */ | |
382ac6b3 GOC |
83 | push_guest_stack(cpu, &gstack, cpu->regs->ss); |
84 | push_guest_stack(cpu, &gstack, cpu->regs->esp); | |
d7e28ffe | 85 | } else { |
bff672e6 | 86 | /* We're staying on the same Guest (kernel) stack. */ |
a53a35a8 GOC |
87 | virtstack = cpu->regs->esp; |
88 | ss = cpu->regs->ss; | |
47436aa4 | 89 | |
1713608f | 90 | origstack = gstack = guest_pa(cpu, virtstack); |
d7e28ffe RR |
91 | } |
92 | ||
bff672e6 RR |
93 | /* Remember that we never let the Guest actually disable interrupts, so |
94 | * the "Interrupt Flag" bit is always set. We copy that bit from the | |
e1e72965 RR |
95 | * Guest's "irq_enabled" field into the eflags word: we saw the Guest |
96 | * copy it back in "lguest_iret". */ | |
a53a35a8 | 97 | eflags = cpu->regs->eflags; |
382ac6b3 | 98 | if (get_user(irq_enable, &cpu->lg->lguest_data->irq_enabled) == 0 |
e5faff45 RR |
99 | && !(irq_enable & X86_EFLAGS_IF)) |
100 | eflags &= ~X86_EFLAGS_IF; | |
d7e28ffe | 101 | |
bff672e6 RR |
102 | /* An interrupt is expected to push three things on the stack: the old |
103 | * "eflags" word, the old code segment, and the old instruction | |
104 | * pointer. */ | |
382ac6b3 GOC |
105 | push_guest_stack(cpu, &gstack, eflags); |
106 | push_guest_stack(cpu, &gstack, cpu->regs->cs); | |
107 | push_guest_stack(cpu, &gstack, cpu->regs->eip); | |
d7e28ffe | 108 | |
bff672e6 | 109 | /* For the six traps which supply an error code, we push that, too. */ |
d7e28ffe | 110 | if (has_err) |
382ac6b3 | 111 | push_guest_stack(cpu, &gstack, cpu->regs->errcode); |
d7e28ffe | 112 | |
bff672e6 RR |
113 | /* Now we've pushed all the old state, we change the stack, the code |
114 | * segment and the address to execute. */ | |
a53a35a8 GOC |
115 | cpu->regs->ss = ss; |
116 | cpu->regs->esp = virtstack + (gstack - origstack); | |
117 | cpu->regs->cs = (__KERNEL_CS|GUEST_PL); | |
118 | cpu->regs->eip = idt_address(lo, hi); | |
d7e28ffe | 119 | |
bff672e6 RR |
120 | /* There are two kinds of interrupt handlers: 0xE is an "interrupt |
121 | * gate" which expects interrupts to be disabled on entry. */ | |
d7e28ffe | 122 | if (idt_type(lo, hi) == 0xE) |
382ac6b3 GOC |
123 | if (put_user(0, &cpu->lg->lguest_data->irq_enabled)) |
124 | kill_guest(cpu, "Disabling interrupts"); | |
d7e28ffe RR |
125 | } |
126 | ||
e1e72965 | 127 | /*H:205 |
bff672e6 RR |
128 | * Virtual Interrupts. |
129 | * | |
130 | * maybe_do_interrupt() gets called before every entry to the Guest, to see if | |
131 | * we should divert the Guest to running an interrupt handler. */ | |
177e449d | 132 | void maybe_do_interrupt(struct lg_cpu *cpu) |
d7e28ffe RR |
133 | { |
134 | unsigned int irq; | |
135 | DECLARE_BITMAP(blk, LGUEST_IRQS); | |
136 | struct desc_struct *idt; | |
137 | ||
bff672e6 | 138 | /* If the Guest hasn't even initialized yet, we can do nothing. */ |
382ac6b3 | 139 | if (!cpu->lg->lguest_data) |
d7e28ffe RR |
140 | return; |
141 | ||
bff672e6 RR |
142 | /* Take our "irqs_pending" array and remove any interrupts the Guest |
143 | * wants blocked: the result ends up in "blk". */ | |
382ac6b3 | 144 | if (copy_from_user(&blk, cpu->lg->lguest_data->blocked_interrupts, |
d7e28ffe RR |
145 | sizeof(blk))) |
146 | return; | |
147 | ||
177e449d | 148 | bitmap_andnot(blk, cpu->irqs_pending, blk, LGUEST_IRQS); |
d7e28ffe | 149 | |
bff672e6 | 150 | /* Find the first interrupt. */ |
d7e28ffe | 151 | irq = find_first_bit(blk, LGUEST_IRQS); |
bff672e6 | 152 | /* None? Nothing to do */ |
d7e28ffe RR |
153 | if (irq >= LGUEST_IRQS) |
154 | return; | |
155 | ||
bff672e6 RR |
156 | /* They may be in the middle of an iret, where they asked us never to |
157 | * deliver interrupts. */ | |
382ac6b3 GOC |
158 | if (cpu->regs->eip >= cpu->lg->noirq_start && |
159 | (cpu->regs->eip < cpu->lg->noirq_end)) | |
d7e28ffe RR |
160 | return; |
161 | ||
bff672e6 | 162 | /* If they're halted, interrupts restart them. */ |
66686c2a | 163 | if (cpu->halted) { |
d7e28ffe | 164 | /* Re-enable interrupts. */ |
382ac6b3 GOC |
165 | if (put_user(X86_EFLAGS_IF, &cpu->lg->lguest_data->irq_enabled)) |
166 | kill_guest(cpu, "Re-enabling interrupts"); | |
66686c2a | 167 | cpu->halted = 0; |
d7e28ffe | 168 | } else { |
bff672e6 | 169 | /* Otherwise we check if they have interrupts disabled. */ |
d7e28ffe | 170 | u32 irq_enabled; |
382ac6b3 | 171 | if (get_user(irq_enabled, &cpu->lg->lguest_data->irq_enabled)) |
d7e28ffe RR |
172 | irq_enabled = 0; |
173 | if (!irq_enabled) | |
174 | return; | |
175 | } | |
176 | ||
bff672e6 RR |
177 | /* Look at the IDT entry the Guest gave us for this interrupt. The |
178 | * first 32 (FIRST_EXTERNAL_VECTOR) entries are for traps, so we skip | |
179 | * over them. */ | |
fc708b3e | 180 | idt = &cpu->arch.idt[FIRST_EXTERNAL_VECTOR+irq]; |
bff672e6 | 181 | /* If they don't have a handler (yet?), we just ignore it */ |
d7e28ffe | 182 | if (idt_present(idt->a, idt->b)) { |
bff672e6 | 183 | /* OK, mark it no longer pending and deliver it. */ |
177e449d | 184 | clear_bit(irq, cpu->irqs_pending); |
bff672e6 RR |
185 | /* set_guest_interrupt() takes the interrupt descriptor and a |
186 | * flag to say whether this interrupt pushes an error code onto | |
187 | * the stack as well: virtual interrupts never do. */ | |
177e449d | 188 | set_guest_interrupt(cpu, idt->a, idt->b, 0); |
d7e28ffe | 189 | } |
6c8dca5d RR |
190 | |
191 | /* Every time we deliver an interrupt, we update the timestamp in the | |
192 | * Guest's lguest_data struct. It would be better for the Guest if we | |
193 | * did this more often, but it can actually be quite slow: doing it | |
194 | * here is a compromise which means at least it gets updated every | |
195 | * timer interrupt. */ | |
382ac6b3 | 196 | write_timestamp(cpu); |
d7e28ffe | 197 | } |
c18acd73 RR |
198 | /*:*/ |
199 | ||
200 | /* Linux uses trap 128 for system calls. Plan9 uses 64, and Ron Minnich sent | |
201 | * me a patch, so we support that too. It'd be a big step for lguest if half | |
202 | * the Plan 9 user base were to start using it. | |
203 | * | |
204 | * Actually now I think of it, it's possible that Ron *is* half the Plan 9 | |
205 | * userbase. Oh well. */ | |
206 | static bool could_be_syscall(unsigned int num) | |
207 | { | |
208 | /* Normal Linux SYSCALL_VECTOR or reserved vector? */ | |
209 | return num == SYSCALL_VECTOR || num == syscall_vector; | |
210 | } | |
211 | ||
212 | /* The syscall vector it wants must be unused by Host. */ | |
213 | bool check_syscall_vector(struct lguest *lg) | |
214 | { | |
215 | u32 vector; | |
216 | ||
217 | if (get_user(vector, &lg->lguest_data->syscall_vec)) | |
218 | return false; | |
219 | ||
220 | return could_be_syscall(vector); | |
221 | } | |
222 | ||
223 | int init_interrupts(void) | |
224 | { | |
225 | /* If they want some strange system call vector, reserve it now */ | |
226 | if (syscall_vector != SYSCALL_VECTOR | |
227 | && test_and_set_bit(syscall_vector, used_vectors)) { | |
228 | printk("lg: couldn't reserve syscall %u\n", syscall_vector); | |
229 | return -EBUSY; | |
230 | } | |
231 | return 0; | |
232 | } | |
233 | ||
234 | void free_interrupts(void) | |
235 | { | |
236 | if (syscall_vector != SYSCALL_VECTOR) | |
237 | clear_bit(syscall_vector, used_vectors); | |
238 | } | |
d7e28ffe | 239 | |
bff672e6 RR |
240 | /*H:220 Now we've got the routines to deliver interrupts, delivering traps |
241 | * like page fault is easy. The only trick is that Intel decided that some | |
242 | * traps should have error codes: */ | |
d7e28ffe RR |
243 | static int has_err(unsigned int trap) |
244 | { | |
245 | return (trap == 8 || (trap >= 10 && trap <= 14) || trap == 17); | |
246 | } | |
247 | ||
bff672e6 | 248 | /* deliver_trap() returns true if it could deliver the trap. */ |
177e449d | 249 | int deliver_trap(struct lg_cpu *cpu, unsigned int num) |
d7e28ffe | 250 | { |
0d027c01 RR |
251 | /* Trap numbers are always 8 bit, but we set an impossible trap number |
252 | * for traps inside the Switcher, so check that here. */ | |
fc708b3e | 253 | if (num >= ARRAY_SIZE(cpu->arch.idt)) |
0d027c01 | 254 | return 0; |
d7e28ffe | 255 | |
bff672e6 RR |
256 | /* Early on the Guest hasn't set the IDT entries (or maybe it put a |
257 | * bogus one in): if we fail here, the Guest will be killed. */ | |
fc708b3e | 258 | if (!idt_present(cpu->arch.idt[num].a, cpu->arch.idt[num].b)) |
d7e28ffe | 259 | return 0; |
fc708b3e GOC |
260 | set_guest_interrupt(cpu, cpu->arch.idt[num].a, |
261 | cpu->arch.idt[num].b, has_err(num)); | |
d7e28ffe RR |
262 | return 1; |
263 | } | |
264 | ||
bff672e6 RR |
265 | /*H:250 Here's the hard part: returning to the Host every time a trap happens |
266 | * and then calling deliver_trap() and re-entering the Guest is slow. | |
e1e72965 RR |
267 | * Particularly because Guest userspace system calls are traps (usually trap |
268 | * 128). | |
bff672e6 RR |
269 | * |
270 | * So we'd like to set up the IDT to tell the CPU to deliver traps directly | |
271 | * into the Guest. This is possible, but the complexities cause the size of | |
272 | * this file to double! However, 150 lines of code is worth writing for taking | |
273 | * system calls down from 1750ns to 270ns. Plus, if lguest didn't do it, all | |
e1e72965 | 274 | * the other hypervisors would beat it up at lunchtime. |
bff672e6 | 275 | * |
56adbe9d RR |
276 | * This routine indicates if a particular trap number could be delivered |
277 | * directly. */ | |
278 | static int direct_trap(unsigned int num) | |
d7e28ffe | 279 | { |
bff672e6 RR |
280 | /* Hardware interrupts don't go to the Guest at all (except system |
281 | * call). */ | |
c18acd73 | 282 | if (num >= FIRST_EXTERNAL_VECTOR && !could_be_syscall(num)) |
d7e28ffe RR |
283 | return 0; |
284 | ||
bff672e6 RR |
285 | /* The Host needs to see page faults (for shadow paging and to save the |
286 | * fault address), general protection faults (in/out emulation) and | |
287 | * device not available (TS handling), and of course, the hypercall | |
288 | * trap. */ | |
56adbe9d | 289 | return num != 14 && num != 13 && num != 7 && num != LGUEST_TRAP_ENTRY; |
d7e28ffe | 290 | } |
f56a384e RR |
291 | /*:*/ |
292 | ||
293 | /*M:005 The Guest has the ability to turn its interrupt gates into trap gates, | |
294 | * if it is careful. The Host will let trap gates can go directly to the | |
295 | * Guest, but the Guest needs the interrupts atomically disabled for an | |
296 | * interrupt gate. It can do this by pointing the trap gate at instructions | |
297 | * within noirq_start and noirq_end, where it can safely disable interrupts. */ | |
298 | ||
299 | /*M:006 The Guests do not use the sysenter (fast system call) instruction, | |
300 | * because it's hardcoded to enter privilege level 0 and so can't go direct. | |
301 | * It's about twice as fast as the older "int 0x80" system call, so it might | |
302 | * still be worthwhile to handle it in the Switcher and lcall down to the | |
303 | * Guest. The sysenter semantics are hairy tho: search for that keyword in | |
304 | * entry.S :*/ | |
d7e28ffe | 305 | |
bff672e6 RR |
306 | /*H:260 When we make traps go directly into the Guest, we need to make sure |
307 | * the kernel stack is valid (ie. mapped in the page tables). Otherwise, the | |
308 | * CPU trying to deliver the trap will fault while trying to push the interrupt | |
309 | * words on the stack: this is called a double fault, and it forces us to kill | |
310 | * the Guest. | |
311 | * | |
312 | * Which is deeply unfair, because (literally!) it wasn't the Guests' fault. */ | |
4665ac8e | 313 | void pin_stack_pages(struct lg_cpu *cpu) |
d7e28ffe RR |
314 | { |
315 | unsigned int i; | |
316 | ||
bff672e6 RR |
317 | /* Depending on the CONFIG_4KSTACKS option, the Guest can have one or |
318 | * two pages of stack space. */ | |
382ac6b3 | 319 | for (i = 0; i < cpu->lg->stack_pages; i++) |
8057d763 RR |
320 | /* The stack grows *upwards*, so the address we're given is the |
321 | * start of the page after the kernel stack. Subtract one to | |
322 | * get back onto the first stack page, and keep subtracting to | |
323 | * get to the rest of the stack pages. */ | |
1713608f | 324 | pin_page(cpu, cpu->esp1 - 1 - i * PAGE_SIZE); |
d7e28ffe RR |
325 | } |
326 | ||
bff672e6 RR |
327 | /* Direct traps also mean that we need to know whenever the Guest wants to use |
328 | * a different kernel stack, so we can change the IDT entries to use that | |
329 | * stack. The IDT entries expect a virtual address, so unlike most addresses | |
330 | * the Guest gives us, the "esp" (stack pointer) value here is virtual, not | |
331 | * physical. | |
332 | * | |
333 | * In Linux each process has its own kernel stack, so this happens a lot: we | |
334 | * change stacks on each context switch. */ | |
4665ac8e | 335 | void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages) |
d7e28ffe | 336 | { |
e1e72965 | 337 | /* You are not allowed have a stack segment with privilege level 0: bad |
bff672e6 | 338 | * Guest! */ |
d7e28ffe | 339 | if ((seg & 0x3) != GUEST_PL) |
382ac6b3 | 340 | kill_guest(cpu, "bad stack segment %i", seg); |
bff672e6 | 341 | /* We only expect one or two stack pages. */ |
d7e28ffe | 342 | if (pages > 2) |
382ac6b3 | 343 | kill_guest(cpu, "bad stack pages %u", pages); |
bff672e6 | 344 | /* Save where the stack is, and how many pages */ |
4665ac8e GOC |
345 | cpu->ss1 = seg; |
346 | cpu->esp1 = esp; | |
347 | cpu->lg->stack_pages = pages; | |
bff672e6 | 348 | /* Make sure the new stack pages are mapped */ |
4665ac8e | 349 | pin_stack_pages(cpu); |
d7e28ffe RR |
350 | } |
351 | ||
bff672e6 RR |
352 | /* All this reference to mapping stacks leads us neatly into the other complex |
353 | * part of the Host: page table handling. */ | |
354 | ||
355 | /*H:235 This is the routine which actually checks the Guest's IDT entry and | |
e1e72965 | 356 | * transfers it into the entry in "struct lguest": */ |
382ac6b3 | 357 | static void set_trap(struct lg_cpu *cpu, struct desc_struct *trap, |
d7e28ffe RR |
358 | unsigned int num, u32 lo, u32 hi) |
359 | { | |
360 | u8 type = idt_type(lo, hi); | |
361 | ||
bff672e6 | 362 | /* We zero-out a not-present entry */ |
d7e28ffe RR |
363 | if (!idt_present(lo, hi)) { |
364 | trap->a = trap->b = 0; | |
365 | return; | |
366 | } | |
367 | ||
bff672e6 | 368 | /* We only support interrupt and trap gates. */ |
d7e28ffe | 369 | if (type != 0xE && type != 0xF) |
382ac6b3 | 370 | kill_guest(cpu, "bad IDT type %i", type); |
d7e28ffe | 371 | |
bff672e6 RR |
372 | /* We only copy the handler address, present bit, privilege level and |
373 | * type. The privilege level controls where the trap can be triggered | |
374 | * manually with an "int" instruction. This is usually GUEST_PL, | |
375 | * except for system calls which userspace can use. */ | |
d7e28ffe RR |
376 | trap->a = ((__KERNEL_CS|GUEST_PL)<<16) | (lo&0x0000FFFF); |
377 | trap->b = (hi&0xFFFFEF00); | |
378 | } | |
379 | ||
bff672e6 RR |
380 | /*H:230 While we're here, dealing with delivering traps and interrupts to the |
381 | * Guest, we might as well complete the picture: how the Guest tells us where | |
382 | * it wants them to go. This would be simple, except making traps fast | |
383 | * requires some tricks. | |
384 | * | |
385 | * We saw the Guest setting Interrupt Descriptor Table (IDT) entries with the | |
386 | * LHCALL_LOAD_IDT_ENTRY hypercall before: that comes here. */ | |
fc708b3e | 387 | void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int num, u32 lo, u32 hi) |
d7e28ffe | 388 | { |
bff672e6 RR |
389 | /* Guest never handles: NMI, doublefault, spurious interrupt or |
390 | * hypercall. We ignore when it tries to set them. */ | |
d7e28ffe RR |
391 | if (num == 2 || num == 8 || num == 15 || num == LGUEST_TRAP_ENTRY) |
392 | return; | |
393 | ||
bff672e6 RR |
394 | /* Mark the IDT as changed: next time the Guest runs we'll know we have |
395 | * to copy this again. */ | |
ae3749dc | 396 | cpu->changed |= CHANGED_IDT; |
bff672e6 | 397 | |
56adbe9d | 398 | /* Check that the Guest doesn't try to step outside the bounds. */ |
fc708b3e | 399 | if (num >= ARRAY_SIZE(cpu->arch.idt)) |
382ac6b3 | 400 | kill_guest(cpu, "Setting idt entry %u", num); |
56adbe9d | 401 | else |
382ac6b3 | 402 | set_trap(cpu, &cpu->arch.idt[num], num, lo, hi); |
d7e28ffe RR |
403 | } |
404 | ||
bff672e6 RR |
405 | /* The default entry for each interrupt points into the Switcher routines which |
406 | * simply return to the Host. The run_guest() loop will then call | |
407 | * deliver_trap() to bounce it back into the Guest. */ | |
d7e28ffe RR |
408 | static void default_idt_entry(struct desc_struct *idt, |
409 | int trap, | |
410 | const unsigned long handler) | |
411 | { | |
bff672e6 | 412 | /* A present interrupt gate. */ |
d7e28ffe RR |
413 | u32 flags = 0x8e00; |
414 | ||
bff672e6 RR |
415 | /* Set the privilege level on the entry for the hypercall: this allows |
416 | * the Guest to use the "int" instruction to trigger it. */ | |
d7e28ffe RR |
417 | if (trap == LGUEST_TRAP_ENTRY) |
418 | flags |= (GUEST_PL << 13); | |
419 | ||
bff672e6 | 420 | /* Now pack it into the IDT entry in its weird format. */ |
d7e28ffe RR |
421 | idt->a = (LGUEST_CS<<16) | (handler&0x0000FFFF); |
422 | idt->b = (handler&0xFFFF0000) | flags; | |
423 | } | |
424 | ||
bff672e6 | 425 | /* When the Guest first starts, we put default entries into the IDT. */ |
d7e28ffe RR |
426 | void setup_default_idt_entries(struct lguest_ro_state *state, |
427 | const unsigned long *def) | |
428 | { | |
429 | unsigned int i; | |
430 | ||
431 | for (i = 0; i < ARRAY_SIZE(state->guest_idt); i++) | |
432 | default_idt_entry(&state->guest_idt[i], i, def[i]); | |
433 | } | |
434 | ||
bff672e6 RR |
435 | /*H:240 We don't use the IDT entries in the "struct lguest" directly, instead |
436 | * we copy them into the IDT which we've set up for Guests on this CPU, just | |
437 | * before we run the Guest. This routine does that copy. */ | |
fc708b3e | 438 | void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt, |
d7e28ffe RR |
439 | const unsigned long *def) |
440 | { | |
441 | unsigned int i; | |
442 | ||
bff672e6 RR |
443 | /* We can simply copy the direct traps, otherwise we use the default |
444 | * ones in the Switcher: they will return to the Host. */ | |
fc708b3e | 445 | for (i = 0; i < ARRAY_SIZE(cpu->arch.idt); i++) { |
56adbe9d RR |
446 | /* If no Guest can ever override this trap, leave it alone. */ |
447 | if (!direct_trap(i)) | |
448 | continue; | |
449 | ||
450 | /* Only trap gates (type 15) can go direct to the Guest. | |
451 | * Interrupt gates (type 14) disable interrupts as they are | |
452 | * entered, which we never let the Guest do. Not present | |
453 | * entries (type 0x0) also can't go direct, of course. */ | |
fc708b3e GOC |
454 | if (idt_type(cpu->arch.idt[i].a, cpu->arch.idt[i].b) == 0xF) |
455 | idt[i] = cpu->arch.idt[i]; | |
d7e28ffe | 456 | else |
56adbe9d | 457 | /* Reset it to the default. */ |
d7e28ffe RR |
458 | default_idt_entry(&idt[i], i, def[i]); |
459 | } | |
d7e28ffe RR |
460 | } |
461 | ||
e1e72965 RR |
462 | /*H:200 |
463 | * The Guest Clock. | |
464 | * | |
465 | * There are two sources of virtual interrupts. We saw one in lguest_user.c: | |
466 | * the Launcher sending interrupts for virtual devices. The other is the Guest | |
467 | * timer interrupt. | |
468 | * | |
469 | * The Guest uses the LHCALL_SET_CLOCKEVENT hypercall to tell us how long to | |
470 | * the next timer interrupt (in nanoseconds). We use the high-resolution timer | |
471 | * infrastructure to set a callback at that time. | |
472 | * | |
473 | * 0 means "turn off the clock". */ | |
ad8d8f3b | 474 | void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta) |
d7e28ffe RR |
475 | { |
476 | ktime_t expires; | |
477 | ||
478 | if (unlikely(delta == 0)) { | |
479 | /* Clock event device is shutting down. */ | |
ad8d8f3b | 480 | hrtimer_cancel(&cpu->hrt); |
d7e28ffe RR |
481 | return; |
482 | } | |
483 | ||
e1e72965 RR |
484 | /* We use wallclock time here, so the Guest might not be running for |
485 | * all the time between now and the timer interrupt it asked for. This | |
486 | * is almost always the right thing to do. */ | |
d7e28ffe | 487 | expires = ktime_add_ns(ktime_get_real(), delta); |
ad8d8f3b | 488 | hrtimer_start(&cpu->hrt, expires, HRTIMER_MODE_ABS); |
d7e28ffe RR |
489 | } |
490 | ||
e1e72965 | 491 | /* This is the function called when the Guest's timer expires. */ |
d7e28ffe RR |
492 | static enum hrtimer_restart clockdev_fn(struct hrtimer *timer) |
493 | { | |
ad8d8f3b | 494 | struct lg_cpu *cpu = container_of(timer, struct lg_cpu, hrt); |
d7e28ffe | 495 | |
e1e72965 | 496 | /* Remember the first interrupt is the timer interrupt. */ |
177e449d | 497 | set_bit(0, cpu->irqs_pending); |
e1e72965 | 498 | /* If the Guest is actually stopped, we need to wake it up. */ |
66686c2a GOC |
499 | if (cpu->halted) |
500 | wake_up_process(cpu->tsk); | |
d7e28ffe RR |
501 | return HRTIMER_NORESTART; |
502 | } | |
503 | ||
e1e72965 | 504 | /* This sets up the timer for this Guest. */ |
ad8d8f3b | 505 | void init_clockdev(struct lg_cpu *cpu) |
d7e28ffe | 506 | { |
ad8d8f3b GOC |
507 | hrtimer_init(&cpu->hrt, CLOCK_REALTIME, HRTIMER_MODE_ABS); |
508 | cpu->hrt.function = clockdev_fn; | |
d7e28ffe | 509 | } |