]>
Commit | Line | Data |
---|---|---|
2e04ef76 RR |
1 | /*P:500 |
2 | * Just as userspace programs request kernel operations through a system | |
f938d2c8 RR |
3 | * call, the Guest requests Host operations through a "hypercall". You might |
4 | * notice this nomenclature doesn't really follow any logic, but the name has | |
5 | * been around for long enough that we're stuck with it. As you'd expect, this | |
2e04ef76 RR |
6 | * code is basically a one big switch statement. |
7 | :*/ | |
f938d2c8 RR |
8 | |
9 | /* Copyright (C) 2006 Rusty Russell IBM Corporation | |
d7e28ffe RR |
10 | |
11 | This program is free software; you can redistribute it and/or modify | |
12 | it under the terms of the GNU General Public License as published by | |
13 | the Free Software Foundation; either version 2 of the License, or | |
14 | (at your option) any later version. | |
15 | ||
16 | This program is distributed in the hope that it will be useful, | |
17 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
18 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
19 | GNU General Public License for more details. | |
20 | ||
21 | You should have received a copy of the GNU General Public License | |
22 | along with this program; if not, write to the Free Software | |
23 | Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
24 | */ | |
25 | #include <linux/uaccess.h> | |
26 | #include <linux/syscalls.h> | |
27 | #include <linux/mm.h> | |
ca94f2bd | 28 | #include <linux/ktime.h> |
d7e28ffe RR |
29 | #include <asm/page.h> |
30 | #include <asm/pgtable.h> | |
d7e28ffe RR |
31 | #include "lg.h" |
32 | ||
2e04ef76 RR |
33 | /*H:120 |
34 | * This is the core hypercall routine: where the Guest gets what it wants. | |
35 | * Or gets killed. Or, in the case of LHCALL_SHUTDOWN, both. | |
36 | */ | |
73044f05 | 37 | static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args) |
d7e28ffe | 38 | { |
b410e7b1 | 39 | switch (args->arg0) { |
d7e28ffe | 40 | case LHCALL_FLUSH_ASYNC: |
2e04ef76 RR |
41 | /* |
42 | * This call does nothing, except by breaking out of the Guest | |
43 | * it makes us process all the asynchronous hypercalls. | |
44 | */ | |
d7e28ffe | 45 | break; |
a32a8813 | 46 | case LHCALL_SEND_INTERRUPTS: |
2e04ef76 RR |
47 | /* |
48 | * This call does nothing too, but by breaking out of the Guest | |
49 | * it makes us process any pending interrupts. | |
50 | */ | |
a32a8813 | 51 | break; |
d7e28ffe | 52 | case LHCALL_LGUEST_INIT: |
2e04ef76 RR |
53 | /* |
54 | * You can't get here unless you're already initialized. Don't | |
55 | * do that. | |
56 | */ | |
382ac6b3 | 57 | kill_guest(cpu, "already have lguest_data"); |
d7e28ffe | 58 | break; |
ec04b13f | 59 | case LHCALL_SHUTDOWN: { |
d7e28ffe | 60 | char msg[128]; |
2e04ef76 | 61 | /* |
a91d74a3 | 62 | * Shutdown is such a trivial hypercall that we do it in five |
2e04ef76 RR |
63 | * lines right here. |
64 | * | |
65 | * If the lgread fails, it will call kill_guest() itself; the | |
66 | * kill_guest() with the message will be ignored. | |
67 | */ | |
382ac6b3 | 68 | __lgread(cpu, msg, args->arg1, sizeof(msg)); |
d7e28ffe | 69 | msg[sizeof(msg)-1] = '\0'; |
382ac6b3 | 70 | kill_guest(cpu, "CRASH: %s", msg); |
ec04b13f | 71 | if (args->arg2 == LGUEST_SHUTDOWN_RESTART) |
382ac6b3 | 72 | cpu->lg->dead = ERR_PTR(-ERESTART); |
d7e28ffe RR |
73 | break; |
74 | } | |
75 | case LHCALL_FLUSH_TLB: | |
2e04ef76 | 76 | /* FLUSH_TLB comes in two flavors, depending on the argument: */ |
b410e7b1 | 77 | if (args->arg1) |
4665ac8e | 78 | guest_pagetable_clear_all(cpu); |
d7e28ffe | 79 | else |
1713608f | 80 | guest_pagetable_flush_user(cpu); |
d7e28ffe | 81 | break; |
bff672e6 | 82 | |
2e04ef76 RR |
83 | /* |
84 | * All these calls simply pass the arguments through to the right | |
85 | * routines. | |
86 | */ | |
d7e28ffe | 87 | case LHCALL_NEW_PGTABLE: |
4665ac8e | 88 | guest_new_pagetable(cpu, args->arg1); |
d7e28ffe RR |
89 | break; |
90 | case LHCALL_SET_STACK: | |
4665ac8e | 91 | guest_set_stack(cpu, args->arg1, args->arg2, args->arg3); |
d7e28ffe RR |
92 | break; |
93 | case LHCALL_SET_PTE: | |
acdd0b62 MZ |
94 | #ifdef CONFIG_X86_PAE |
95 | guest_set_pte(cpu, args->arg1, args->arg2, | |
96 | __pte(args->arg3 | (u64)args->arg4 << 32)); | |
97 | #else | |
382ac6b3 | 98 | guest_set_pte(cpu, args->arg1, args->arg2, __pte(args->arg3)); |
acdd0b62 | 99 | #endif |
d7e28ffe | 100 | break; |
ebe0ba84 MZ |
101 | case LHCALL_SET_PGD: |
102 | guest_set_pgd(cpu->lg, args->arg1, args->arg2); | |
d7e28ffe | 103 | break; |
acdd0b62 MZ |
104 | #ifdef CONFIG_X86_PAE |
105 | case LHCALL_SET_PMD: | |
106 | guest_set_pmd(cpu->lg, args->arg1, args->arg2); | |
107 | break; | |
108 | #endif | |
d7e28ffe | 109 | case LHCALL_SET_CLOCKEVENT: |
ad8d8f3b | 110 | guest_set_clockevent(cpu, args->arg1); |
d7e28ffe RR |
111 | break; |
112 | case LHCALL_TS: | |
bff672e6 | 113 | /* This sets the TS flag, as we saw used in run_guest(). */ |
4665ac8e | 114 | cpu->ts = args->arg1; |
d7e28ffe RR |
115 | break; |
116 | case LHCALL_HALT: | |
bff672e6 | 117 | /* Similarly, this sets the halted flag for run_guest(). */ |
66686c2a | 118 | cpu->halted = 1; |
d7e28ffe RR |
119 | break; |
120 | default: | |
e1e72965 | 121 | /* It should be an architecture-specific hypercall. */ |
73044f05 | 122 | if (lguest_arch_do_hcall(cpu, args)) |
382ac6b3 | 123 | kill_guest(cpu, "Bad hypercall %li\n", args->arg0); |
d7e28ffe RR |
124 | } |
125 | } | |
126 | ||
2e04ef76 RR |
127 | /*H:124 |
128 | * Asynchronous hypercalls are easy: we just look in the array in the | |
b410e7b1 | 129 | * Guest's "struct lguest_data" to see if any new ones are marked "ready". |
bff672e6 RR |
130 | * |
131 | * We are careful to do these in order: obviously we respect the order the | |
132 | * Guest put them in the ring, but we also promise the Guest that they will | |
133 | * happen before any normal hypercall (which is why we check this before | |
2e04ef76 RR |
134 | * checking for a normal hcall). |
135 | */ | |
73044f05 | 136 | static void do_async_hcalls(struct lg_cpu *cpu) |
d7e28ffe RR |
137 | { |
138 | unsigned int i; | |
139 | u8 st[LHCALL_RING_SIZE]; | |
140 | ||
bff672e6 | 141 | /* For simplicity, we copy the entire call status array in at once. */ |
382ac6b3 | 142 | if (copy_from_user(&st, &cpu->lg->lguest_data->hcall_status, sizeof(st))) |
d7e28ffe RR |
143 | return; |
144 | ||
bff672e6 | 145 | /* We process "struct lguest_data"s hcalls[] ring once. */ |
d7e28ffe | 146 | for (i = 0; i < ARRAY_SIZE(st); i++) { |
b410e7b1 | 147 | struct hcall_args args; |
2e04ef76 RR |
148 | /* |
149 | * We remember where we were up to from last time. This makes | |
bff672e6 | 150 | * sure that the hypercalls are done in the order the Guest |
2e04ef76 RR |
151 | * places them in the ring. |
152 | */ | |
73044f05 | 153 | unsigned int n = cpu->next_hcall; |
d7e28ffe | 154 | |
bff672e6 | 155 | /* 0xFF means there's no call here (yet). */ |
d7e28ffe RR |
156 | if (st[n] == 0xFF) |
157 | break; | |
158 | ||
2e04ef76 RR |
159 | /* |
160 | * OK, we have hypercall. Increment the "next_hcall" cursor, | |
161 | * and wrap back to 0 if we reach the end. | |
162 | */ | |
73044f05 GOC |
163 | if (++cpu->next_hcall == LHCALL_RING_SIZE) |
164 | cpu->next_hcall = 0; | |
d7e28ffe | 165 | |
2e04ef76 RR |
166 | /* |
167 | * Copy the hypercall arguments into a local copy of the | |
168 | * hcall_args struct. | |
169 | */ | |
382ac6b3 | 170 | if (copy_from_user(&args, &cpu->lg->lguest_data->hcalls[n], |
b410e7b1 | 171 | sizeof(struct hcall_args))) { |
382ac6b3 | 172 | kill_guest(cpu, "Fetching async hypercalls"); |
d7e28ffe RR |
173 | break; |
174 | } | |
175 | ||
bff672e6 | 176 | /* Do the hypercall, same as a normal one. */ |
73044f05 | 177 | do_hcall(cpu, &args); |
bff672e6 RR |
178 | |
179 | /* Mark the hypercall done. */ | |
382ac6b3 GOC |
180 | if (put_user(0xFF, &cpu->lg->lguest_data->hcall_status[n])) { |
181 | kill_guest(cpu, "Writing result for async hypercall"); | |
d7e28ffe RR |
182 | break; |
183 | } | |
184 | ||
2e04ef76 RR |
185 | /* |
186 | * Stop doing hypercalls if they want to notify the Launcher: | |
187 | * it needs to service this first. | |
188 | */ | |
69a09dc1 | 189 | if (cpu->pending.trap) |
d7e28ffe RR |
190 | break; |
191 | } | |
192 | } | |
193 | ||
2e04ef76 RR |
194 | /* |
195 | * Last of all, we look at what happens first of all. The very first time the | |
196 | * Guest makes a hypercall, we end up here to set things up: | |
197 | */ | |
73044f05 | 198 | static void initialize(struct lg_cpu *cpu) |
d7e28ffe | 199 | { |
2e04ef76 RR |
200 | /* |
201 | * You can't do anything until you're initialized. The Guest knows the | |
202 | * rules, so we're unforgiving here. | |
203 | */ | |
73044f05 | 204 | if (cpu->hcall->arg0 != LHCALL_LGUEST_INIT) { |
382ac6b3 | 205 | kill_guest(cpu, "hypercall %li before INIT", cpu->hcall->arg0); |
d7e28ffe RR |
206 | return; |
207 | } | |
208 | ||
73044f05 | 209 | if (lguest_arch_init_hypercalls(cpu)) |
382ac6b3 | 210 | kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data); |
3c6b5bfa | 211 | |
2e04ef76 RR |
212 | /* |
213 | * The Guest tells us where we're not to deliver interrupts by putting | |
2f921b5b | 214 | * the instruction address into "struct lguest_data". |
2e04ef76 | 215 | */ |
2f921b5b | 216 | if (get_user(cpu->lg->noirq_iret, &cpu->lg->lguest_data->noirq_iret)) |
382ac6b3 | 217 | kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data); |
d7e28ffe | 218 | |
2e04ef76 RR |
219 | /* |
220 | * We write the current time into the Guest's data page once so it can | |
221 | * set its clock. | |
222 | */ | |
382ac6b3 | 223 | write_timestamp(cpu); |
6c8dca5d | 224 | |
47436aa4 | 225 | /* page_tables.c will also do some setup. */ |
382ac6b3 | 226 | page_table_guest_data_init(cpu); |
47436aa4 | 227 | |
2e04ef76 RR |
228 | /* |
229 | * This is the one case where the above accesses might have been the | |
bff672e6 | 230 | * first write to a Guest page. This may have caused a copy-on-write |
e1e72965 | 231 | * fault, but the old page might be (read-only) in the Guest |
2e04ef76 RR |
232 | * pagetable. |
233 | */ | |
4665ac8e | 234 | guest_pagetable_clear_all(cpu); |
d7e28ffe | 235 | } |
a6bd8e13 RR |
236 | /*:*/ |
237 | ||
2e04ef76 RR |
238 | /*M:013 |
239 | * If a Guest reads from a page (so creates a mapping) that it has never | |
a6bd8e13 RR |
240 | * written to, and then the Launcher writes to it (ie. the output of a virtual |
241 | * device), the Guest will still see the old page. In practice, this never | |
242 | * happens: why would the Guest read a page which it has never written to? But | |
2e04ef76 | 243 | * a similar scenario might one day bite us, so it's worth mentioning. |
a91d74a3 RR |
244 | * |
245 | * Note that if we used a shared anonymous mapping in the Launcher instead of | |
246 | * mapping /dev/zero private, we wouldn't worry about cop-on-write. And we | |
247 | * need that to switch the Launcher to processes (away from threads) anyway. | |
2e04ef76 | 248 | :*/ |
d7e28ffe | 249 | |
bff672e6 RR |
250 | /*H:100 |
251 | * Hypercalls | |
252 | * | |
253 | * Remember from the Guest, hypercalls come in two flavors: normal and | |
254 | * asynchronous. This file handles both of types. | |
255 | */ | |
73044f05 | 256 | void do_hypercalls(struct lg_cpu *cpu) |
d7e28ffe | 257 | { |
cc6d4fbc | 258 | /* Not initialized yet? This hypercall must do it. */ |
73044f05 | 259 | if (unlikely(!cpu->lg->lguest_data)) { |
cc6d4fbc | 260 | /* Set up the "struct lguest_data" */ |
73044f05 | 261 | initialize(cpu); |
cc6d4fbc | 262 | /* Hcall is done. */ |
73044f05 | 263 | cpu->hcall = NULL; |
d7e28ffe RR |
264 | return; |
265 | } | |
266 | ||
2e04ef76 RR |
267 | /* |
268 | * The Guest has initialized. | |
bff672e6 | 269 | * |
2e04ef76 RR |
270 | * Look in the hypercall ring for the async hypercalls: |
271 | */ | |
73044f05 | 272 | do_async_hcalls(cpu); |
bff672e6 | 273 | |
2e04ef76 RR |
274 | /* |
275 | * If we stopped reading the hypercall ring because the Guest did a | |
15045275 | 276 | * NOTIFY to the Launcher, we want to return now. Otherwise we do |
2e04ef76 RR |
277 | * the hypercall. |
278 | */ | |
69a09dc1 | 279 | if (!cpu->pending.trap) { |
73044f05 | 280 | do_hcall(cpu, cpu->hcall); |
2e04ef76 RR |
281 | /* |
282 | * Tricky point: we reset the hcall pointer to mark the | |
cc6d4fbc RR |
283 | * hypercall as "done". We use the hcall pointer rather than |
284 | * the trap number to indicate a hypercall is pending. | |
285 | * Normally it doesn't matter: the Guest will run again and | |
286 | * update the trap number before we come back here. | |
287 | * | |
e1e72965 | 288 | * However, if we are signalled or the Guest sends I/O to the |
cc6d4fbc RR |
289 | * Launcher, the run_guest() loop will exit without running the |
290 | * Guest. When it comes back it would try to re-run the | |
2e04ef76 RR |
291 | * hypercall. Finding that bug sucked. |
292 | */ | |
73044f05 | 293 | cpu->hcall = NULL; |
d7e28ffe RR |
294 | } |
295 | } | |
6c8dca5d | 296 | |
2e04ef76 RR |
297 | /* |
298 | * This routine supplies the Guest with time: it's used for wallclock time at | |
299 | * initial boot and as a rough time source if the TSC isn't available. | |
300 | */ | |
382ac6b3 | 301 | void write_timestamp(struct lg_cpu *cpu) |
6c8dca5d RR |
302 | { |
303 | struct timespec now; | |
304 | ktime_get_real_ts(&now); | |
382ac6b3 GOC |
305 | if (copy_to_user(&cpu->lg->lguest_data->time, |
306 | &now, sizeof(struct timespec))) | |
307 | kill_guest(cpu, "Writing timestamp"); | |
6c8dca5d | 308 | } |