Commit | Line | Data |
---|---|---|
f938d2c8 RR |
1 | /*P:500 Just as userspace programs request kernel operations through a system |
2 | * call, the Guest requests Host operations through a "hypercall". You might | |
3 | * notice this nomenclature doesn't really follow any logic, but the name has | |
4 | * been around for long enough that we're stuck with it. As you'd expect, this | |
5 | * code is basically a one big switch statement. :*/ | |
6 | ||
7 | /* Copyright (C) 2006 Rusty Russell IBM Corporation | |
d7e28ffe RR |
8 | |
9 | This program is free software; you can redistribute it and/or modify | |
10 | it under the terms of the GNU General Public License as published by | |
11 | the Free Software Foundation; either version 2 of the License, or | |
12 | (at your option) any later version. | |
13 | ||
14 | This program is distributed in the hope that it will be useful, | |
15 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | GNU General Public License for more details. | |
18 | ||
19 | You should have received a copy of the GNU General Public License | |
20 | along with this program; if not, write to the Free Software | |
21 | Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
22 | */ | |
23 | #include <linux/uaccess.h> | |
24 | #include <linux/syscalls.h> | |
25 | #include <linux/mm.h> | |
26 | #include <asm/page.h> | |
27 | #include <asm/pgtable.h> | |
28 | #include <irq_vectors.h> | |
29 | #include "lg.h" | |
30 | ||
bff672e6 RR |
31 | /*H:120 This is the core hypercall routine: where the Guest gets what it |
32 | * wants. Or gets killed. Or, in the case of LHCALL_CRASH, both. | |
33 | * | |
34 | * Remember from the Guest: %eax == which call to make, and the arguments are | |
35 | * packed into %edx, %ebx and %ecx if needed. */ | |
d7e28ffe RR |
36 | static void do_hcall(struct lguest *lg, struct lguest_regs *regs) |
37 | { | |
38 | switch (regs->eax) { | |
39 | case LHCALL_FLUSH_ASYNC: | |
bff672e6 RR |
40 | /* This call does nothing, except by breaking out of the Guest |
41 | * it makes us process all the asynchronous hypercalls. */ | |
d7e28ffe RR |
42 | break; |
43 | case LHCALL_LGUEST_INIT: | |
bff672e6 RR |
44 | /* You can't get here unless you're already initialized. Don't |
45 | * do that. */ | |
d7e28ffe RR |
46 | kill_guest(lg, "already have lguest_data"); |
47 | break; | |
48 | case LHCALL_CRASH: { | |
bff672e6 RR |
49 | /* Crash is such a trivial hypercall that we do it in four |
50 | * lines right here. */ | |
d7e28ffe | 51 | char msg[128]; |
bff672e6 RR |
52 | /* If the lgread fails, it will call kill_guest() itself; the |
53 | * kill_guest() with the message will be ignored. */ | |
d7e28ffe RR |
54 | lgread(lg, msg, regs->edx, sizeof(msg)); |
55 | msg[sizeof(msg)-1] = '\0'; | |
56 | kill_guest(lg, "CRASH: %s", msg); | |
57 | break; | |
58 | } | |
59 | case LHCALL_FLUSH_TLB: | |
bff672e6 RR |
60 | /* FLUSH_TLB comes in two flavors, depending on the |
61 | * argument: */ | |
d7e28ffe RR |
62 | if (regs->edx) |
63 | guest_pagetable_clear_all(lg); | |
64 | else | |
65 | guest_pagetable_flush_user(lg); | |
66 | break; | |
d7e28ffe | 67 | case LHCALL_BIND_DMA: |
bff672e6 RR |
68 | /* BIND_DMA really wants four arguments, but it's the only call |
69 | * which does. So the Guest packs the number of buffers and | |
70 | * the interrupt number into the final argument, and we decode | |
71 | * it here. This can legitimately fail, since we currently | |
72 | * place a limit on the number of DMA pools a Guest can have. | |
73 | * So we return true or false from this call. */ | |
d7e28ffe RR |
74 | regs->eax = bind_dma(lg, regs->edx, regs->ebx, |
75 | regs->ecx >> 8, regs->ecx & 0xFF); | |
76 | break; | |
bff672e6 RR |
77 | |
78 | /* All these calls simply pass the arguments through to the right | |
79 | * routines. */ | |
d7e28ffe RR |
80 | case LHCALL_SEND_DMA: |
81 | send_dma(lg, regs->edx, regs->ebx); | |
82 | break; | |
83 | case LHCALL_LOAD_GDT: | |
84 | load_guest_gdt(lg, regs->edx, regs->ebx); | |
85 | break; | |
86 | case LHCALL_LOAD_IDT_ENTRY: | |
87 | load_guest_idt_entry(lg, regs->edx, regs->ebx, regs->ecx); | |
88 | break; | |
89 | case LHCALL_NEW_PGTABLE: | |
90 | guest_new_pagetable(lg, regs->edx); | |
91 | break; | |
92 | case LHCALL_SET_STACK: | |
93 | guest_set_stack(lg, regs->edx, regs->ebx, regs->ecx); | |
94 | break; | |
95 | case LHCALL_SET_PTE: | |
96 | guest_set_pte(lg, regs->edx, regs->ebx, mkgpte(regs->ecx)); | |
97 | break; | |
98 | case LHCALL_SET_PMD: | |
99 | guest_set_pmd(lg, regs->edx, regs->ebx); | |
100 | break; | |
101 | case LHCALL_LOAD_TLS: | |
102 | guest_load_tls(lg, regs->edx); | |
103 | break; | |
104 | case LHCALL_SET_CLOCKEVENT: | |
105 | guest_set_clockevent(lg, regs->edx); | |
106 | break; | |
bff672e6 | 107 | |
d7e28ffe | 108 | case LHCALL_TS: |
bff672e6 | 109 | /* This sets the TS flag, as we saw used in run_guest(). */ |
d7e28ffe RR |
110 | lg->ts = regs->edx; |
111 | break; | |
112 | case LHCALL_HALT: | |
bff672e6 | 113 | /* Similarly, this sets the halted flag for run_guest(). */ |
d7e28ffe RR |
114 | lg->halted = 1; |
115 | break; | |
116 | default: | |
117 | kill_guest(lg, "Bad hypercall %li\n", regs->eax); | |
118 | } | |
119 | } | |
120 | ||
bff672e6 RR |
121 | /* Asynchronous hypercalls are easy: we just look in the array in the Guest's |
122 | * "struct lguest_data" and see if there are any new ones marked "ready". | |
123 | * | |
124 | * We are careful to do these in order: obviously we respect the order the | |
125 | * Guest put them in the ring, but we also promise the Guest that they will | |
126 | * happen before any normal hypercall (which is why we check this before | |
127 | * checking for a normal hcall). */ | |
d7e28ffe RR |
128 | static void do_async_hcalls(struct lguest *lg) |
129 | { | |
130 | unsigned int i; | |
131 | u8 st[LHCALL_RING_SIZE]; | |
132 | ||
bff672e6 | 133 | /* For simplicity, we copy the entire call status array in at once. */ |
d7e28ffe RR |
134 | if (copy_from_user(&st, &lg->lguest_data->hcall_status, sizeof(st))) |
135 | return; | |
136 | ||
bff672e6 RR |
137 | |
138 | /* We process "struct lguest_data"s hcalls[] ring once. */ | |
d7e28ffe RR |
139 | for (i = 0; i < ARRAY_SIZE(st); i++) { |
140 | struct lguest_regs regs; | |
bff672e6 RR |
141 | /* We remember where we were up to from last time. This makes |
142 | * sure that the hypercalls are done in the order the Guest | |
143 | * places them in the ring. */ | |
d7e28ffe RR |
144 | unsigned int n = lg->next_hcall; |
145 | ||
bff672e6 | 146 | /* 0xFF means there's no call here (yet). */ |
d7e28ffe RR |
147 | if (st[n] == 0xFF) |
148 | break; | |
149 | ||
bff672e6 RR |
150 | /* OK, we have hypercall. Increment the "next_hcall" cursor, |
151 | * and wrap back to 0 if we reach the end. */ | |
d7e28ffe RR |
152 | if (++lg->next_hcall == LHCALL_RING_SIZE) |
153 | lg->next_hcall = 0; | |
154 | ||
bff672e6 RR |
155 | /* We copy the hypercall arguments into a fake register |
156 | * structure. This makes life simple for do_hcall(). */ | |
d7e28ffe RR |
157 | if (get_user(regs.eax, &lg->lguest_data->hcalls[n].eax) |
158 | || get_user(regs.edx, &lg->lguest_data->hcalls[n].edx) | |
159 | || get_user(regs.ecx, &lg->lguest_data->hcalls[n].ecx) | |
160 | || get_user(regs.ebx, &lg->lguest_data->hcalls[n].ebx)) { | |
161 | kill_guest(lg, "Fetching async hypercalls"); | |
162 | break; | |
163 | } | |
164 | ||
bff672e6 | 165 | /* Do the hypercall, same as a normal one. */ |
d7e28ffe | 166 | do_hcall(lg, ®s); |
bff672e6 RR |
167 | |
168 | /* Mark the hypercall done. */ | |
d7e28ffe RR |
169 | if (put_user(0xFF, &lg->lguest_data->hcall_status[n])) { |
170 | kill_guest(lg, "Writing result for async hypercall"); | |
171 | break; | |
172 | } | |
173 | ||
bff672e6 RR |
174 | /* Stop doing hypercalls if we've just done a DMA to the |
175 | * Launcher: it needs to service this first. */ | |
d7e28ffe RR |
176 | if (lg->dma_is_pending) |
177 | break; | |
178 | } | |
179 | } | |
180 | ||
bff672e6 RR |
181 | /* Last of all, we look at what happens first of all. The very first time the |
182 | * Guest makes a hypercall, we end up here to set things up: */ | |
d7e28ffe RR |
183 | static void initialize(struct lguest *lg) |
184 | { | |
185 | u32 tsc_speed; | |
186 | ||
bff672e6 RR |
187 | /* You can't do anything until you're initialized. The Guest knows the |
188 | * rules, so we're unforgiving here. */ | |
d7e28ffe RR |
189 | if (lg->regs->eax != LHCALL_LGUEST_INIT) { |
190 | kill_guest(lg, "hypercall %li before LGUEST_INIT", | |
191 | lg->regs->eax); | |
192 | return; | |
193 | } | |
194 | ||
bff672e6 RR |
195 | /* We insist that the Time Stamp Counter exist and doesn't change with |
196 | * cpu frequency. Some devious chip manufacturers decided that TSC | |
197 | * changes could be handled in software. I decided that time going | |
198 | * backwards might be good for benchmarks, but it's bad for users. | |
199 | * | |
200 | * We also insist that the TSC be stable: the kernel detects unreliable | |
201 | * TSCs for its own purposes, and we use that here. */ | |
d7e28ffe RR |
202 | if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC) && !check_tsc_unstable()) |
203 | tsc_speed = tsc_khz; | |
204 | else | |
205 | tsc_speed = 0; | |
206 | ||
bff672e6 RR |
207 | /* The pointer to the Guest's "struct lguest_data" is the only |
208 | * argument. */ | |
d7e28ffe | 209 | lg->lguest_data = (struct lguest_data __user *)lg->regs->edx; |
bff672e6 RR |
210 | /* If we check the address they gave is OK now, we can simply |
211 | * copy_to_user/from_user from now on rather than using lgread/lgwrite. | |
212 | * I put this in to show that I'm not immune to writing stupid | |
213 | * optimizations. */ | |
d7e28ffe RR |
214 | if (!lguest_address_ok(lg, lg->regs->edx, sizeof(*lg->lguest_data))) { |
215 | kill_guest(lg, "bad guest page %p", lg->lguest_data); | |
216 | return; | |
217 | } | |
bff672e6 RR |
218 | /* The Guest tells us where we're not to deliver interrupts by putting |
219 | * the range of addresses into "struct lguest_data". */ | |
d7e28ffe RR |
220 | if (get_user(lg->noirq_start, &lg->lguest_data->noirq_start) |
221 | || get_user(lg->noirq_end, &lg->lguest_data->noirq_end) | |
bff672e6 RR |
222 | /* We tell the Guest that it can't use the top 4MB of virtual |
223 | * addresses used by the Switcher. */ | |
d7e28ffe RR |
224 | || put_user(4U*1024*1024, &lg->lguest_data->reserve_mem) |
225 | || put_user(tsc_speed, &lg->lguest_data->tsc_khz) | |
bff672e6 | 226 | /* We also give the Guest a unique id, as used in lguest_net.c. */ |
d7e28ffe RR |
227 | || put_user(lg->guestid, &lg->lguest_data->guestid)) |
228 | kill_guest(lg, "bad guest page %p", lg->lguest_data); | |
229 | ||
6c8dca5d RR |
230 | /* We write the current time into the Guest's data page once now. */ |
231 | write_timestamp(lg); | |
232 | ||
bff672e6 RR |
233 | /* This is the one case where the above accesses might have been the |
234 | * first write to a Guest page. This may have caused a copy-on-write | |
235 | * fault, but the Guest might be referring to the old (read-only) | |
236 | * page. */ | |
d7e28ffe RR |
237 | guest_pagetable_clear_all(lg); |
238 | } | |
bff672e6 RR |
239 | /* Now we've examined the hypercall code; our Guest can make requests. There |
240 | * is one other way we can do things for the Guest, as we see in | |
241 | * emulate_insn(). */ | |
d7e28ffe | 242 | |
bff672e6 RR |
243 | /*H:110 Tricky point: we mark the hypercall as "done" once we've done it. |
244 | * Normally we don't need to do this: the Guest will run again and update the | |
245 | * trap number before we come back around the run_guest() loop to | |
246 | * do_hypercalls(). | |
247 | * | |
248 | * However, if we are signalled or the Guest sends DMA to the Launcher, that | |
249 | * loop will exit without running the Guest. When it comes back it would try | |
250 | * to re-run the hypercall. */ | |
d7e28ffe RR |
251 | static void clear_hcall(struct lguest *lg) |
252 | { | |
253 | lg->regs->trapnum = 255; | |
254 | } | |
255 | ||
bff672e6 RR |
256 | /*H:100 |
257 | * Hypercalls | |
258 | * | |
259 | * Remember from the Guest, hypercalls come in two flavors: normal and | |
260 | * asynchronous. This file handles both of types. | |
261 | */ | |
d7e28ffe RR |
262 | void do_hypercalls(struct lguest *lg) |
263 | { | |
bff672e6 | 264 | /* Not initialized yet? */ |
d7e28ffe | 265 | if (unlikely(!lg->lguest_data)) { |
bff672e6 RR |
266 | /* Did the Guest make a hypercall? We might have come back for |
267 | * some other reason (an interrupt, a different trap). */ | |
d7e28ffe | 268 | if (lg->regs->trapnum == LGUEST_TRAP_ENTRY) { |
bff672e6 | 269 | /* Set up the "struct lguest_data" */ |
d7e28ffe | 270 | initialize(lg); |
bff672e6 | 271 | /* The hypercall is done. */ |
d7e28ffe RR |
272 | clear_hcall(lg); |
273 | } | |
274 | return; | |
275 | } | |
276 | ||
bff672e6 RR |
277 | /* The Guest has initialized. |
278 | * | |
279 | * Look in the hypercall ring for the async hypercalls: */ | |
d7e28ffe | 280 | do_async_hcalls(lg); |
bff672e6 RR |
281 | |
282 | /* If we stopped reading the hypercall ring because the Guest did a | |
283 | * SEND_DMA to the Launcher, we want to return now. Otherwise if the | |
284 | * Guest asked us to do a hypercall, we do it. */ | |
d7e28ffe RR |
285 | if (!lg->dma_is_pending && lg->regs->trapnum == LGUEST_TRAP_ENTRY) { |
286 | do_hcall(lg, lg->regs); | |
bff672e6 | 287 | /* The hypercall is done. */ |
d7e28ffe RR |
288 | clear_hcall(lg); |
289 | } | |
290 | } | |
6c8dca5d RR |
291 | |
292 | /* This routine supplies the Guest with time: it's used for wallclock time at | |
293 | * initial boot and as a rough time source if the TSC isn't available. */ | |
294 | void write_timestamp(struct lguest *lg) | |
295 | { | |
296 | struct timespec now; | |
297 | ktime_get_real_ts(&now); | |
298 | if (put_user(now, &lg->lguest_data->time)) | |
299 | kill_guest(lg, "Writing timestamp"); | |
300 | } |