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. */ | |
d7e28ffe RR |
44 | static void push_guest_stack(struct lguest *lg, unsigned long *gstack, u32 val) |
45 | { | |
bff672e6 | 46 | /* Stack grows upwards: move stack then write value. */ |
d7e28ffe | 47 | *gstack -= 4; |
2d37f94a | 48 | lgwrite(lg, *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; |
177e449d | 68 | struct lguest *lg = cpu->lg; |
d7e28ffe | 69 | |
bff672e6 RR |
70 | /* There are two cases for interrupts: one where the Guest is already |
71 | * in the kernel, and a more complex one where the Guest is in | |
72 | * userspace. We check the privilege level to find out. */ | |
d7e28ffe | 73 | if ((lg->regs->ss&0x3) != GUEST_PL) { |
bff672e6 RR |
74 | /* The Guest told us their kernel stack with the SET_STACK |
75 | * hypercall: both the virtual address and the segment */ | |
47436aa4 | 76 | virtstack = lg->esp1; |
d7e28ffe | 77 | ss = lg->ss1; |
47436aa4 RR |
78 | |
79 | origstack = gstack = guest_pa(lg, virtstack); | |
bff672e6 RR |
80 | /* We push the old stack segment and pointer onto the new |
81 | * stack: when the Guest does an "iret" back from the interrupt | |
82 | * handler the CPU will notice they're dropping privilege | |
83 | * levels and expect these here. */ | |
d7e28ffe RR |
84 | push_guest_stack(lg, &gstack, lg->regs->ss); |
85 | push_guest_stack(lg, &gstack, lg->regs->esp); | |
86 | } else { | |
bff672e6 | 87 | /* We're staying on the same Guest (kernel) stack. */ |
47436aa4 | 88 | virtstack = lg->regs->esp; |
d7e28ffe | 89 | ss = lg->regs->ss; |
47436aa4 RR |
90 | |
91 | origstack = gstack = guest_pa(lg, virtstack); | |
d7e28ffe RR |
92 | } |
93 | ||
bff672e6 RR |
94 | /* Remember that we never let the Guest actually disable interrupts, so |
95 | * the "Interrupt Flag" bit is always set. We copy that bit from the | |
e1e72965 RR |
96 | * Guest's "irq_enabled" field into the eflags word: we saw the Guest |
97 | * copy it back in "lguest_iret". */ | |
d7e28ffe | 98 | eflags = lg->regs->eflags; |
e5faff45 RR |
99 | if (get_user(irq_enable, &lg->lguest_data->irq_enabled) == 0 |
100 | && !(irq_enable & X86_EFLAGS_IF)) | |
101 | eflags &= ~X86_EFLAGS_IF; | |
d7e28ffe | 102 | |
bff672e6 RR |
103 | /* An interrupt is expected to push three things on the stack: the old |
104 | * "eflags" word, the old code segment, and the old instruction | |
105 | * pointer. */ | |
d7e28ffe RR |
106 | push_guest_stack(lg, &gstack, eflags); |
107 | push_guest_stack(lg, &gstack, lg->regs->cs); | |
108 | push_guest_stack(lg, &gstack, lg->regs->eip); | |
109 | ||
bff672e6 | 110 | /* For the six traps which supply an error code, we push that, too. */ |
d7e28ffe RR |
111 | if (has_err) |
112 | push_guest_stack(lg, &gstack, lg->regs->errcode); | |
113 | ||
bff672e6 RR |
114 | /* Now we've pushed all the old state, we change the stack, the code |
115 | * segment and the address to execute. */ | |
d7e28ffe | 116 | lg->regs->ss = ss; |
47436aa4 | 117 | lg->regs->esp = virtstack + (gstack - origstack); |
d7e28ffe RR |
118 | lg->regs->cs = (__KERNEL_CS|GUEST_PL); |
119 | lg->regs->eip = idt_address(lo, hi); | |
120 | ||
bff672e6 RR |
121 | /* There are two kinds of interrupt handlers: 0xE is an "interrupt |
122 | * gate" which expects interrupts to be disabled on entry. */ | |
d7e28ffe RR |
123 | if (idt_type(lo, hi) == 0xE) |
124 | if (put_user(0, &lg->lguest_data->irq_enabled)) | |
125 | kill_guest(lg, "Disabling interrupts"); | |
126 | } | |
127 | ||
e1e72965 | 128 | /*H:205 |
bff672e6 RR |
129 | * 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. */ | |
177e449d | 133 | void maybe_do_interrupt(struct lg_cpu *cpu) |
d7e28ffe RR |
134 | { |
135 | unsigned int irq; | |
177e449d | 136 | struct lguest *lg = cpu->lg; |
d7e28ffe RR |
137 | DECLARE_BITMAP(blk, LGUEST_IRQS); |
138 | struct desc_struct *idt; | |
139 | ||
bff672e6 | 140 | /* If the Guest hasn't even initialized yet, we can do nothing. */ |
d7e28ffe RR |
141 | if (!lg->lguest_data) |
142 | return; | |
143 | ||
bff672e6 RR |
144 | /* Take our "irqs_pending" array and remove any interrupts the Guest |
145 | * wants blocked: the result ends up in "blk". */ | |
d7e28ffe RR |
146 | if (copy_from_user(&blk, lg->lguest_data->blocked_interrupts, |
147 | sizeof(blk))) | |
148 | return; | |
149 | ||
177e449d | 150 | bitmap_andnot(blk, cpu->irqs_pending, blk, LGUEST_IRQS); |
d7e28ffe | 151 | |
bff672e6 | 152 | /* Find the first interrupt. */ |
d7e28ffe | 153 | irq = find_first_bit(blk, LGUEST_IRQS); |
bff672e6 | 154 | /* None? Nothing to do */ |
d7e28ffe RR |
155 | if (irq >= LGUEST_IRQS) |
156 | return; | |
157 | ||
bff672e6 RR |
158 | /* They may be in the middle of an iret, where they asked us never to |
159 | * deliver interrupts. */ | |
d7e28ffe RR |
160 | if (lg->regs->eip >= lg->noirq_start && lg->regs->eip < lg->noirq_end) |
161 | return; | |
162 | ||
bff672e6 | 163 | /* If they're halted, interrupts restart them. */ |
d7e28ffe RR |
164 | if (lg->halted) { |
165 | /* Re-enable interrupts. */ | |
166 | if (put_user(X86_EFLAGS_IF, &lg->lguest_data->irq_enabled)) | |
167 | kill_guest(lg, "Re-enabling interrupts"); | |
168 | lg->halted = 0; | |
169 | } else { | |
bff672e6 | 170 | /* Otherwise we check if they have interrupts disabled. */ |
d7e28ffe RR |
171 | 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 | ||
bff672e6 RR |
178 | /* 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. */ | |
625efab1 | 181 | idt = &lg->arch.idt[FIRST_EXTERNAL_VECTOR+irq]; |
bff672e6 | 182 | /* If they don't have a handler (yet?), we just ignore it */ |
d7e28ffe | 183 | if (idt_present(idt->a, idt->b)) { |
bff672e6 | 184 | /* OK, mark it no longer pending and deliver it. */ |
177e449d | 185 | clear_bit(irq, cpu->irqs_pending); |
bff672e6 RR |
186 | /* 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. */ | |
177e449d | 189 | set_guest_interrupt(cpu, idt->a, idt->b, 0); |
d7e28ffe | 190 | } |
6c8dca5d RR |
191 | |
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); | |
d7e28ffe | 198 | } |
c18acd73 RR |
199 | /*:*/ |
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. */ | |
207 | static 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. */ | |
214 | bool 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 | ||
224 | int 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 | ||
235 | void free_interrupts(void) | |
236 | { | |
237 | if (syscall_vector != SYSCALL_VECTOR) | |
238 | clear_bit(syscall_vector, used_vectors); | |
239 | } | |
d7e28ffe | 240 | |
bff672e6 RR |
241 | /*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: */ | |
d7e28ffe RR |
244 | static int has_err(unsigned int trap) |
245 | { | |
246 | return (trap == 8 || (trap >= 10 && trap <= 14) || trap == 17); | |
247 | } | |
248 | ||
bff672e6 | 249 | /* deliver_trap() returns true if it could deliver the trap. */ |
177e449d | 250 | int deliver_trap(struct lg_cpu *cpu, unsigned int num) |
d7e28ffe | 251 | { |
0d027c01 RR |
252 | /* Trap numbers are always 8 bit, but we set an impossible trap number |
253 | * for traps inside the Switcher, so check that here. */ | |
177e449d | 254 | if (num >= ARRAY_SIZE(cpu->lg->arch.idt)) |
0d027c01 | 255 | return 0; |
d7e28ffe | 256 | |
bff672e6 RR |
257 | /* 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. */ | |
177e449d | 259 | if (!idt_present(cpu->lg->arch.idt[num].a, cpu->lg->arch.idt[num].b)) |
d7e28ffe | 260 | return 0; |
177e449d GOC |
261 | set_guest_interrupt(cpu, cpu->lg->arch.idt[num].a, |
262 | cpu->lg->arch.idt[num].b, has_err(num)); | |
d7e28ffe RR |
263 | return 1; |
264 | } | |
265 | ||
bff672e6 RR |
266 | /*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. | |
e1e72965 RR |
268 | * Particularly because Guest userspace system calls are traps (usually trap |
269 | * 128). | |
bff672e6 RR |
270 | * |
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 | |
e1e72965 | 275 | * the other hypervisors would beat it up at lunchtime. |
bff672e6 | 276 | * |
56adbe9d RR |
277 | * This routine indicates if a particular trap number could be delivered |
278 | * directly. */ | |
279 | static int direct_trap(unsigned int num) | |
d7e28ffe | 280 | { |
bff672e6 RR |
281 | /* Hardware interrupts don't go to the Guest at all (except system |
282 | * call). */ | |
c18acd73 | 283 | if (num >= FIRST_EXTERNAL_VECTOR && !could_be_syscall(num)) |
d7e28ffe RR |
284 | return 0; |
285 | ||
bff672e6 RR |
286 | /* 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. */ | |
56adbe9d | 290 | return num != 14 && num != 13 && num != 7 && num != LGUEST_TRAP_ENTRY; |
d7e28ffe | 291 | } |
f56a384e RR |
292 | /*:*/ |
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 :*/ | |
d7e28ffe | 306 | |
bff672e6 RR |
307 | /*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. */ | |
d7e28ffe RR |
314 | void pin_stack_pages(struct lguest *lg) |
315 | { | |
316 | unsigned int i; | |
317 | ||
bff672e6 RR |
318 | /* Depending on the CONFIG_4KSTACKS option, the Guest can have one or |
319 | * two pages of stack space. */ | |
d7e28ffe | 320 | for (i = 0; i < lg->stack_pages; i++) |
8057d763 RR |
321 | /* 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); | |
d7e28ffe RR |
326 | } |
327 | ||
bff672e6 RR |
328 | /* 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. */ | |
d7e28ffe RR |
336 | void guest_set_stack(struct lguest *lg, u32 seg, u32 esp, unsigned int pages) |
337 | { | |
e1e72965 | 338 | /* You are not allowed have a stack segment with privilege level 0: bad |
bff672e6 | 339 | * Guest! */ |
d7e28ffe RR |
340 | if ((seg & 0x3) != GUEST_PL) |
341 | kill_guest(lg, "bad stack segment %i", seg); | |
bff672e6 | 342 | /* We only expect one or two stack pages. */ |
d7e28ffe RR |
343 | if (pages > 2) |
344 | kill_guest(lg, "bad stack pages %u", pages); | |
bff672e6 | 345 | /* Save where the stack is, and how many pages */ |
d7e28ffe RR |
346 | lg->ss1 = seg; |
347 | lg->esp1 = esp; | |
348 | lg->stack_pages = pages; | |
bff672e6 | 349 | /* Make sure the new stack pages are mapped */ |
d7e28ffe RR |
350 | pin_stack_pages(lg); |
351 | } | |
352 | ||
bff672e6 RR |
353 | /* 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 | |
e1e72965 | 357 | * transfers it into the entry in "struct lguest": */ |
d7e28ffe RR |
358 | static 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 | ||
bff672e6 | 363 | /* We zero-out a not-present entry */ |
d7e28ffe RR |
364 | if (!idt_present(lo, hi)) { |
365 | trap->a = trap->b = 0; | |
366 | return; | |
367 | } | |
368 | ||
bff672e6 | 369 | /* We only support interrupt and trap gates. */ |
d7e28ffe RR |
370 | if (type != 0xE && type != 0xF) |
371 | kill_guest(lg, "bad IDT type %i", type); | |
372 | ||
bff672e6 RR |
373 | /* 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. */ | |
d7e28ffe RR |
377 | trap->a = ((__KERNEL_CS|GUEST_PL)<<16) | (lo&0x0000FFFF); |
378 | trap->b = (hi&0xFFFFEF00); | |
379 | } | |
380 | ||
bff672e6 RR |
381 | /*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. */ | |
d7e28ffe RR |
388 | void load_guest_idt_entry(struct lguest *lg, unsigned int num, u32 lo, u32 hi) |
389 | { | |
bff672e6 RR |
390 | /* Guest never handles: NMI, doublefault, spurious interrupt or |
391 | * hypercall. We ignore when it tries to set them. */ | |
d7e28ffe RR |
392 | if (num == 2 || num == 8 || num == 15 || num == LGUEST_TRAP_ENTRY) |
393 | return; | |
394 | ||
bff672e6 RR |
395 | /* Mark the IDT as changed: next time the Guest runs we'll know we have |
396 | * to copy this again. */ | |
d7e28ffe | 397 | lg->changed |= CHANGED_IDT; |
bff672e6 | 398 | |
56adbe9d | 399 | /* Check that the Guest doesn't try to step outside the bounds. */ |
625efab1 | 400 | if (num >= ARRAY_SIZE(lg->arch.idt)) |
56adbe9d RR |
401 | kill_guest(lg, "Setting idt entry %u", num); |
402 | else | |
625efab1 | 403 | set_trap(lg, &lg->arch.idt[num], num, lo, hi); |
d7e28ffe RR |
404 | } |
405 | ||
bff672e6 RR |
406 | /* 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. */ | |
d7e28ffe RR |
409 | static void default_idt_entry(struct desc_struct *idt, |
410 | int trap, | |
411 | const unsigned long handler) | |
412 | { | |
bff672e6 | 413 | /* A present interrupt gate. */ |
d7e28ffe RR |
414 | u32 flags = 0x8e00; |
415 | ||
bff672e6 RR |
416 | /* Set the privilege level on the entry for the hypercall: this allows |
417 | * the Guest to use the "int" instruction to trigger it. */ | |
d7e28ffe RR |
418 | if (trap == LGUEST_TRAP_ENTRY) |
419 | flags |= (GUEST_PL << 13); | |
420 | ||
bff672e6 | 421 | /* Now pack it into the IDT entry in its weird format. */ |
d7e28ffe RR |
422 | idt->a = (LGUEST_CS<<16) | (handler&0x0000FFFF); |
423 | idt->b = (handler&0xFFFF0000) | flags; | |
424 | } | |
425 | ||
bff672e6 | 426 | /* When the Guest first starts, we put default entries into the IDT. */ |
d7e28ffe RR |
427 | void 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 | ||
bff672e6 RR |
436 | /*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. */ | |
d7e28ffe RR |
439 | void copy_traps(const struct lguest *lg, struct desc_struct *idt, |
440 | const unsigned long *def) | |
441 | { | |
442 | unsigned int i; | |
443 | ||
bff672e6 RR |
444 | /* We can simply copy the direct traps, otherwise we use the default |
445 | * ones in the Switcher: they will return to the Host. */ | |
625efab1 | 446 | for (i = 0; i < ARRAY_SIZE(lg->arch.idt); i++) { |
56adbe9d RR |
447 | /* 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. */ | |
625efab1 JS |
455 | if (idt_type(lg->arch.idt[i].a, lg->arch.idt[i].b) == 0xF) |
456 | idt[i] = lg->arch.idt[i]; | |
d7e28ffe | 457 | else |
56adbe9d | 458 | /* Reset it to the default. */ |
d7e28ffe RR |
459 | default_idt_entry(&idt[i], i, def[i]); |
460 | } | |
d7e28ffe RR |
461 | } |
462 | ||
e1e72965 RR |
463 | /*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". */ | |
ad8d8f3b | 475 | void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta) |
d7e28ffe RR |
476 | { |
477 | ktime_t expires; | |
478 | ||
479 | if (unlikely(delta == 0)) { | |
480 | /* Clock event device is shutting down. */ | |
ad8d8f3b | 481 | hrtimer_cancel(&cpu->hrt); |
d7e28ffe RR |
482 | return; |
483 | } | |
484 | ||
e1e72965 RR |
485 | /* 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. */ | |
d7e28ffe | 488 | expires = ktime_add_ns(ktime_get_real(), delta); |
ad8d8f3b | 489 | hrtimer_start(&cpu->hrt, expires, HRTIMER_MODE_ABS); |
d7e28ffe RR |
490 | } |
491 | ||
e1e72965 | 492 | /* This is the function called when the Guest's timer expires. */ |
d7e28ffe RR |
493 | static enum hrtimer_restart clockdev_fn(struct hrtimer *timer) |
494 | { | |
ad8d8f3b | 495 | struct lg_cpu *cpu = container_of(timer, struct lg_cpu, hrt); |
d7e28ffe | 496 | |
e1e72965 | 497 | /* Remember the first interrupt is the timer interrupt. */ |
177e449d | 498 | set_bit(0, cpu->irqs_pending); |
e1e72965 | 499 | /* If the Guest is actually stopped, we need to wake it up. */ |
ad8d8f3b GOC |
500 | if (cpu->lg->halted) |
501 | wake_up_process(cpu->lg->tsk); | |
d7e28ffe RR |
502 | return HRTIMER_NORESTART; |
503 | } | |
504 | ||
e1e72965 | 505 | /* This sets up the timer for this Guest. */ |
ad8d8f3b | 506 | void init_clockdev(struct lg_cpu *cpu) |
d7e28ffe | 507 | { |
ad8d8f3b GOC |
508 | hrtimer_init(&cpu->hrt, CLOCK_REALTIME, HRTIMER_MODE_ABS); |
509 | cpu->hrt.function = clockdev_fn; | |
d7e28ffe | 510 | } |