Commit | Line | Data |
---|---|---|
f938d2c8 RR |
1 | /*P:010 |
2 | * A hypervisor allows multiple Operating Systems to run on a single machine. | |
3 | * To quote David Wheeler: "Any problem in computer science can be solved with | |
4 | * another layer of indirection." | |
5 | * | |
6 | * We keep things simple in two ways. First, we start with a normal Linux | |
7 | * kernel and insert a module (lg.ko) which allows us to run other Linux | |
8 | * kernels the same way we'd run processes. We call the first kernel the Host, | |
9 | * and the others the Guests. The program which sets up and configures Guests | |
10 | * (such as the example in Documentation/lguest/lguest.c) is called the | |
11 | * Launcher. | |
12 | * | |
13 | * Secondly, we only run specially modified Guests, not normal kernels. When | |
14 | * you set CONFIG_LGUEST to 'y' or 'm', this automatically sets | |
15 | * CONFIG_LGUEST_GUEST=y, which compiles this file into the kernel so it knows | |
16 | * how to be a Guest. This means that you can use the same kernel you boot | |
17 | * normally (ie. as a Host) as a Guest. | |
07ad157f | 18 | * |
f938d2c8 RR |
19 | * These Guests know that they cannot do privileged operations, such as disable |
20 | * interrupts, and that they have to ask the Host to do such things explicitly. | |
21 | * This file consists of all the replacements for such low-level native | |
22 | * hardware operations: these special Guest versions call the Host. | |
23 | * | |
24 | * So how does the kernel know it's a Guest? The Guest starts at a special | |
25 | * entry point marked with a magic string, which sets up a few things then | |
93b1eab3 | 26 | * calls here. We replace the native functions various "paravirt" structures |
f938d2c8 RR |
27 | * with our Guest versions, then boot like normal. :*/ |
28 | ||
29 | /* | |
07ad157f RR |
30 | * Copyright (C) 2006, Rusty Russell <rusty@rustcorp.com.au> IBM Corporation. |
31 | * | |
32 | * This program is free software; you can redistribute it and/or modify | |
33 | * it under the terms of the GNU General Public License as published by | |
34 | * the Free Software Foundation; either version 2 of the License, or | |
35 | * (at your option) any later version. | |
36 | * | |
37 | * This program is distributed in the hope that it will be useful, but | |
38 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
39 | * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or | |
40 | * NON INFRINGEMENT. See the GNU General Public License for more | |
41 | * details. | |
42 | * | |
43 | * You should have received a copy of the GNU General Public License | |
44 | * along with this program; if not, write to the Free Software | |
45 | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | |
46 | */ | |
47 | #include <linux/kernel.h> | |
48 | #include <linux/start_kernel.h> | |
49 | #include <linux/string.h> | |
50 | #include <linux/console.h> | |
51 | #include <linux/screen_info.h> | |
52 | #include <linux/irq.h> | |
53 | #include <linux/interrupt.h> | |
d7e28ffe RR |
54 | #include <linux/clocksource.h> |
55 | #include <linux/clockchips.h> | |
07ad157f RR |
56 | #include <linux/lguest.h> |
57 | #include <linux/lguest_launcher.h> | |
19f1537b | 58 | #include <linux/virtio_console.h> |
4cfe6c3c | 59 | #include <linux/pm.h> |
07ad157f RR |
60 | #include <asm/paravirt.h> |
61 | #include <asm/param.h> | |
62 | #include <asm/page.h> | |
63 | #include <asm/pgtable.h> | |
64 | #include <asm/desc.h> | |
65 | #include <asm/setup.h> | |
66 | #include <asm/e820.h> | |
67 | #include <asm/mce.h> | |
68 | #include <asm/io.h> | |
625efab1 | 69 | #include <asm/i387.h> |
07ad157f | 70 | |
b2b47c21 RR |
71 | /*G:010 Welcome to the Guest! |
72 | * | |
73 | * The Guest in our tale is a simple creature: identical to the Host but | |
74 | * behaving in simplified but equivalent ways. In particular, the Guest is the | |
75 | * same kernel as the Host (or at least, built from the same source code). :*/ | |
76 | ||
07ad157f RR |
77 | /* Declarations for definitions in lguest_guest.S */ |
78 | extern char lguest_noirq_start[], lguest_noirq_end[]; | |
79 | extern const char lgstart_cli[], lgend_cli[]; | |
80 | extern const char lgstart_sti[], lgend_sti[]; | |
81 | extern const char lgstart_popf[], lgend_popf[]; | |
82 | extern const char lgstart_pushf[], lgend_pushf[]; | |
83 | extern const char lgstart_iret[], lgend_iret[]; | |
84 | extern void lguest_iret(void); | |
85 | ||
86 | struct lguest_data lguest_data = { | |
87 | .hcall_status = { [0 ... LHCALL_RING_SIZE-1] = 0xFF }, | |
88 | .noirq_start = (u32)lguest_noirq_start, | |
89 | .noirq_end = (u32)lguest_noirq_end, | |
47436aa4 | 90 | .kernel_address = PAGE_OFFSET, |
07ad157f | 91 | .blocked_interrupts = { 1 }, /* Block timer interrupts */ |
c18acd73 | 92 | .syscall_vec = SYSCALL_VECTOR, |
07ad157f | 93 | }; |
9d1ca6f1 | 94 | static cycle_t clock_base; |
07ad157f | 95 | |
633872b9 | 96 | /*G:037 async_hcall() is pretty simple: I'm quite proud of it really. We have a |
b2b47c21 RR |
97 | * ring buffer of stored hypercalls which the Host will run though next time we |
98 | * do a normal hypercall. Each entry in the ring has 4 slots for the hypercall | |
99 | * arguments, and a "hcall_status" word which is 0 if the call is ready to go, | |
100 | * and 255 once the Host has finished with it. | |
101 | * | |
102 | * If we come around to a slot which hasn't been finished, then the table is | |
103 | * full and we just make the hypercall directly. This has the nice side | |
104 | * effect of causing the Host to run all the stored calls in the ring buffer | |
105 | * which empties it for next time! */ | |
9b56fdb4 AB |
106 | static void async_hcall(unsigned long call, unsigned long arg1, |
107 | unsigned long arg2, unsigned long arg3) | |
07ad157f RR |
108 | { |
109 | /* Note: This code assumes we're uniprocessor. */ | |
110 | static unsigned int next_call; | |
111 | unsigned long flags; | |
112 | ||
b2b47c21 RR |
113 | /* Disable interrupts if not already disabled: we don't want an |
114 | * interrupt handler making a hypercall while we're already doing | |
115 | * one! */ | |
07ad157f RR |
116 | local_irq_save(flags); |
117 | if (lguest_data.hcall_status[next_call] != 0xFF) { | |
118 | /* Table full, so do normal hcall which will flush table. */ | |
119 | hcall(call, arg1, arg2, arg3); | |
120 | } else { | |
b410e7b1 JS |
121 | lguest_data.hcalls[next_call].arg0 = call; |
122 | lguest_data.hcalls[next_call].arg1 = arg1; | |
123 | lguest_data.hcalls[next_call].arg2 = arg2; | |
124 | lguest_data.hcalls[next_call].arg3 = arg3; | |
b2b47c21 | 125 | /* Arguments must all be written before we mark it to go */ |
07ad157f RR |
126 | wmb(); |
127 | lguest_data.hcall_status[next_call] = 0; | |
128 | if (++next_call == LHCALL_RING_SIZE) | |
129 | next_call = 0; | |
130 | } | |
131 | local_irq_restore(flags); | |
132 | } | |
9b56fdb4 | 133 | |
633872b9 RR |
134 | /*G:035 Notice the lazy_hcall() above, rather than hcall(). This is our first |
135 | * real optimization trick! | |
136 | * | |
137 | * When lazy_mode is set, it means we're allowed to defer all hypercalls and do | |
138 | * them as a batch when lazy_mode is eventually turned off. Because hypercalls | |
139 | * are reasonably expensive, batching them up makes sense. For example, a | |
140 | * large munmap might update dozens of page table entries: that code calls | |
141 | * paravirt_enter_lazy_mmu(), does the dozen updates, then calls | |
142 | * lguest_leave_lazy_mode(). | |
143 | * | |
144 | * So, when we're in lazy mode, we call async_hcall() to store the call for | |
145 | * future processing. */ | |
9b56fdb4 AB |
146 | static void lazy_hcall(unsigned long call, |
147 | unsigned long arg1, | |
148 | unsigned long arg2, | |
149 | unsigned long arg3) | |
150 | { | |
151 | if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_NONE) | |
152 | hcall(call, arg1, arg2, arg3); | |
153 | else | |
154 | async_hcall(call, arg1, arg2, arg3); | |
155 | } | |
633872b9 RR |
156 | |
157 | /* When lazy mode is turned off reset the per-cpu lazy mode variable and then | |
158 | * issue a hypercall to flush any stored calls. */ | |
159 | static void lguest_leave_lazy_mode(void) | |
160 | { | |
161 | paravirt_leave_lazy(paravirt_get_lazy_mode()); | |
162 | hcall(LHCALL_FLUSH_ASYNC, 0, 0, 0); | |
163 | } | |
07ad157f | 164 | |
b2b47c21 | 165 | /*G:033 |
e1e72965 RR |
166 | * After that diversion we return to our first native-instruction |
167 | * replacements: four functions for interrupt control. | |
b2b47c21 RR |
168 | * |
169 | * The simplest way of implementing these would be to have "turn interrupts | |
170 | * off" and "turn interrupts on" hypercalls. Unfortunately, this is too slow: | |
171 | * these are by far the most commonly called functions of those we override. | |
172 | * | |
173 | * So instead we keep an "irq_enabled" field inside our "struct lguest_data", | |
174 | * which the Guest can update with a single instruction. The Host knows to | |
175 | * check there when it wants to deliver an interrupt. | |
176 | */ | |
177 | ||
65ea5b03 PA |
178 | /* save_flags() is expected to return the processor state (ie. "flags"). The |
179 | * flags word contains all kind of stuff, but in practice Linux only cares | |
b2b47c21 | 180 | * about the interrupt flag. Our "save_flags()" just returns that. */ |
07ad157f RR |
181 | static unsigned long save_fl(void) |
182 | { | |
183 | return lguest_data.irq_enabled; | |
184 | } | |
185 | ||
e1e72965 | 186 | /* restore_flags() just sets the flags back to the value given. */ |
07ad157f RR |
187 | static void restore_fl(unsigned long flags) |
188 | { | |
07ad157f RR |
189 | lguest_data.irq_enabled = flags; |
190 | } | |
191 | ||
b2b47c21 | 192 | /* Interrupts go off... */ |
07ad157f RR |
193 | static void irq_disable(void) |
194 | { | |
195 | lguest_data.irq_enabled = 0; | |
196 | } | |
197 | ||
b2b47c21 | 198 | /* Interrupts go on... */ |
07ad157f RR |
199 | static void irq_enable(void) |
200 | { | |
07ad157f RR |
201 | lguest_data.irq_enabled = X86_EFLAGS_IF; |
202 | } | |
f56a384e RR |
203 | /*:*/ |
204 | /*M:003 Note that we don't check for outstanding interrupts when we re-enable | |
205 | * them (or when we unmask an interrupt). This seems to work for the moment, | |
206 | * since interrupts are rare and we'll just get the interrupt on the next timer | |
207 | * tick, but when we turn on CONFIG_NO_HZ, we should revisit this. One way | |
208 | * would be to put the "irq_enabled" field in a page by itself, and have the | |
209 | * Host write-protect it when an interrupt comes in when irqs are disabled. | |
210 | * There will then be a page fault as soon as interrupts are re-enabled. :*/ | |
07ad157f | 211 | |
b2b47c21 RR |
212 | /*G:034 |
213 | * The Interrupt Descriptor Table (IDT). | |
214 | * | |
215 | * The IDT tells the processor what to do when an interrupt comes in. Each | |
216 | * entry in the table is a 64-bit descriptor: this holds the privilege level, | |
217 | * address of the handler, and... well, who cares? The Guest just asks the | |
218 | * Host to make the change anyway, because the Host controls the real IDT. | |
219 | */ | |
07ad157f RR |
220 | static void lguest_write_idt_entry(struct desc_struct *dt, |
221 | int entrynum, u32 low, u32 high) | |
222 | { | |
b2b47c21 | 223 | /* Keep the local copy up to date. */ |
07ad157f | 224 | write_dt_entry(dt, entrynum, low, high); |
b2b47c21 | 225 | /* Tell Host about this new entry. */ |
07ad157f RR |
226 | hcall(LHCALL_LOAD_IDT_ENTRY, entrynum, low, high); |
227 | } | |
228 | ||
b2b47c21 RR |
229 | /* Changing to a different IDT is very rare: we keep the IDT up-to-date every |
230 | * time it is written, so we can simply loop through all entries and tell the | |
231 | * Host about them. */ | |
07ad157f RR |
232 | static void lguest_load_idt(const struct Xgt_desc_struct *desc) |
233 | { | |
234 | unsigned int i; | |
235 | struct desc_struct *idt = (void *)desc->address; | |
236 | ||
237 | for (i = 0; i < (desc->size+1)/8; i++) | |
238 | hcall(LHCALL_LOAD_IDT_ENTRY, i, idt[i].a, idt[i].b); | |
239 | } | |
240 | ||
b2b47c21 RR |
241 | /* |
242 | * The Global Descriptor Table. | |
243 | * | |
244 | * The Intel architecture defines another table, called the Global Descriptor | |
245 | * Table (GDT). You tell the CPU where it is (and its size) using the "lgdt" | |
246 | * instruction, and then several other instructions refer to entries in the | |
247 | * table. There are three entries which the Switcher needs, so the Host simply | |
248 | * controls the entire thing and the Guest asks it to make changes using the | |
249 | * LOAD_GDT hypercall. | |
250 | * | |
251 | * This is the opposite of the IDT code where we have a LOAD_IDT_ENTRY | |
252 | * hypercall and use that repeatedly to load a new IDT. I don't think it | |
253 | * really matters, but wouldn't it be nice if they were the same? | |
254 | */ | |
07ad157f RR |
255 | static void lguest_load_gdt(const struct Xgt_desc_struct *desc) |
256 | { | |
257 | BUG_ON((desc->size+1)/8 != GDT_ENTRIES); | |
258 | hcall(LHCALL_LOAD_GDT, __pa(desc->address), GDT_ENTRIES, 0); | |
259 | } | |
260 | ||
b2b47c21 RR |
261 | /* For a single GDT entry which changes, we do the lazy thing: alter our GDT, |
262 | * then tell the Host to reload the entire thing. This operation is so rare | |
263 | * that this naive implementation is reasonable. */ | |
07ad157f RR |
264 | static void lguest_write_gdt_entry(struct desc_struct *dt, |
265 | int entrynum, u32 low, u32 high) | |
266 | { | |
267 | write_dt_entry(dt, entrynum, low, high); | |
268 | hcall(LHCALL_LOAD_GDT, __pa(dt), GDT_ENTRIES, 0); | |
269 | } | |
270 | ||
b2b47c21 RR |
271 | /* OK, I lied. There are three "thread local storage" GDT entries which change |
272 | * on every context switch (these three entries are how glibc implements | |
273 | * __thread variables). So we have a hypercall specifically for this case. */ | |
07ad157f RR |
274 | static void lguest_load_tls(struct thread_struct *t, unsigned int cpu) |
275 | { | |
0d027c01 RR |
276 | /* There's one problem which normal hardware doesn't have: the Host |
277 | * can't handle us removing entries we're currently using. So we clear | |
278 | * the GS register here: if it's needed it'll be reloaded anyway. */ | |
279 | loadsegment(gs, 0); | |
07ad157f RR |
280 | lazy_hcall(LHCALL_LOAD_TLS, __pa(&t->tls_array), cpu, 0); |
281 | } | |
282 | ||
b2b47c21 | 283 | /*G:038 That's enough excitement for now, back to ploughing through each of |
93b1eab3 | 284 | * the different pv_ops structures (we're about 1/3 of the way through). |
b2b47c21 RR |
285 | * |
286 | * This is the Local Descriptor Table, another weird Intel thingy. Linux only | |
287 | * uses this for some strange applications like Wine. We don't do anything | |
288 | * here, so they'll get an informative and friendly Segmentation Fault. */ | |
07ad157f RR |
289 | static void lguest_set_ldt(const void *addr, unsigned entries) |
290 | { | |
291 | } | |
292 | ||
b2b47c21 RR |
293 | /* This loads a GDT entry into the "Task Register": that entry points to a |
294 | * structure called the Task State Segment. Some comments scattered though the | |
295 | * kernel code indicate that this used for task switching in ages past, along | |
296 | * with blood sacrifice and astrology. | |
297 | * | |
298 | * Now there's nothing interesting in here that we don't get told elsewhere. | |
299 | * But the native version uses the "ltr" instruction, which makes the Host | |
300 | * complain to the Guest about a Segmentation Fault and it'll oops. So we | |
301 | * override the native version with a do-nothing version. */ | |
07ad157f RR |
302 | static void lguest_load_tr_desc(void) |
303 | { | |
304 | } | |
305 | ||
b2b47c21 RR |
306 | /* The "cpuid" instruction is a way of querying both the CPU identity |
307 | * (manufacturer, model, etc) and its features. It was introduced before the | |
308 | * Pentium in 1993 and keeps getting extended by both Intel and AMD. As you | |
309 | * might imagine, after a decade and a half this treatment, it is now a giant | |
310 | * ball of hair. Its entry in the current Intel manual runs to 28 pages. | |
311 | * | |
312 | * This instruction even it has its own Wikipedia entry. The Wikipedia entry | |
313 | * has been translated into 4 languages. I am not making this up! | |
314 | * | |
315 | * We could get funky here and identify ourselves as "GenuineLguest", but | |
316 | * instead we just use the real "cpuid" instruction. Then I pretty much turned | |
317 | * off feature bits until the Guest booted. (Don't say that: you'll damage | |
318 | * lguest sales!) Shut up, inner voice! (Hey, just pointing out that this is | |
319 | * hardly future proof.) Noone's listening! They don't like you anyway, | |
320 | * parenthetic weirdo! | |
321 | * | |
322 | * Replacing the cpuid so we can turn features off is great for the kernel, but | |
323 | * anyone (including userspace) can just use the raw "cpuid" instruction and | |
324 | * the Host won't even notice since it isn't privileged. So we try not to get | |
325 | * too worked up about it. */ | |
65ea5b03 PA |
326 | static void lguest_cpuid(unsigned int *ax, unsigned int *bx, |
327 | unsigned int *cx, unsigned int *dx) | |
07ad157f | 328 | { |
65ea5b03 | 329 | int function = *ax; |
07ad157f | 330 | |
65ea5b03 | 331 | native_cpuid(ax, bx, cx, dx); |
07ad157f RR |
332 | switch (function) { |
333 | case 1: /* Basic feature request. */ | |
334 | /* We only allow kernel to see SSE3, CMPXCHG16B and SSSE3 */ | |
65ea5b03 | 335 | *cx &= 0x00002201; |
d7e28ffe | 336 | /* SSE, SSE2, FXSR, MMX, CMOV, CMPXCHG8B, FPU. */ |
65ea5b03 | 337 | *dx &= 0x07808101; |
b2b47c21 RR |
338 | /* The Host can do a nice optimization if it knows that the |
339 | * kernel mappings (addresses above 0xC0000000 or whatever | |
340 | * PAGE_OFFSET is set to) haven't changed. But Linux calls | |
341 | * flush_tlb_user() for both user and kernel mappings unless | |
342 | * the Page Global Enable (PGE) feature bit is set. */ | |
65ea5b03 | 343 | *dx |= 0x00002000; |
07ad157f RR |
344 | break; |
345 | case 0x80000000: | |
346 | /* Futureproof this a little: if they ask how much extended | |
b2b47c21 | 347 | * processor information there is, limit it to known fields. */ |
65ea5b03 PA |
348 | if (*ax > 0x80000008) |
349 | *ax = 0x80000008; | |
07ad157f RR |
350 | break; |
351 | } | |
352 | } | |
353 | ||
b2b47c21 RR |
354 | /* Intel has four control registers, imaginatively named cr0, cr2, cr3 and cr4. |
355 | * I assume there's a cr1, but it hasn't bothered us yet, so we'll not bother | |
356 | * it. The Host needs to know when the Guest wants to change them, so we have | |
357 | * a whole series of functions like read_cr0() and write_cr0(). | |
358 | * | |
e1e72965 | 359 | * We start with cr0. cr0 allows you to turn on and off all kinds of basic |
b2b47c21 RR |
360 | * features, but Linux only really cares about one: the horrifically-named Task |
361 | * Switched (TS) bit at bit 3 (ie. 8) | |
362 | * | |
363 | * What does the TS bit do? Well, it causes the CPU to trap (interrupt 7) if | |
364 | * the floating point unit is used. Which allows us to restore FPU state | |
365 | * lazily after a task switch, and Linux uses that gratefully, but wouldn't a | |
366 | * name like "FPUTRAP bit" be a little less cryptic? | |
367 | * | |
368 | * We store cr0 (and cr3) locally, because the Host never changes it. The | |
369 | * Guest sometimes wants to read it and we'd prefer not to bother the Host | |
370 | * unnecessarily. */ | |
07ad157f RR |
371 | static unsigned long current_cr0, current_cr3; |
372 | static void lguest_write_cr0(unsigned long val) | |
373 | { | |
25c47bb3 | 374 | lazy_hcall(LHCALL_TS, val & X86_CR0_TS, 0, 0); |
07ad157f RR |
375 | current_cr0 = val; |
376 | } | |
377 | ||
378 | static unsigned long lguest_read_cr0(void) | |
379 | { | |
380 | return current_cr0; | |
381 | } | |
382 | ||
b2b47c21 RR |
383 | /* Intel provided a special instruction to clear the TS bit for people too cool |
384 | * to use write_cr0() to do it. This "clts" instruction is faster, because all | |
385 | * the vowels have been optimized out. */ | |
07ad157f RR |
386 | static void lguest_clts(void) |
387 | { | |
388 | lazy_hcall(LHCALL_TS, 0, 0, 0); | |
25c47bb3 | 389 | current_cr0 &= ~X86_CR0_TS; |
07ad157f RR |
390 | } |
391 | ||
e1e72965 | 392 | /* cr2 is the virtual address of the last page fault, which the Guest only ever |
b2b47c21 RR |
393 | * reads. The Host kindly writes this into our "struct lguest_data", so we |
394 | * just read it out of there. */ | |
07ad157f RR |
395 | static unsigned long lguest_read_cr2(void) |
396 | { | |
397 | return lguest_data.cr2; | |
398 | } | |
399 | ||
e1e72965 | 400 | /* cr3 is the current toplevel pagetable page: the principle is the same as |
b2b47c21 | 401 | * cr0. Keep a local copy, and tell the Host when it changes. */ |
07ad157f RR |
402 | static void lguest_write_cr3(unsigned long cr3) |
403 | { | |
404 | lazy_hcall(LHCALL_NEW_PGTABLE, cr3, 0, 0); | |
405 | current_cr3 = cr3; | |
406 | } | |
407 | ||
408 | static unsigned long lguest_read_cr3(void) | |
409 | { | |
410 | return current_cr3; | |
411 | } | |
412 | ||
e1e72965 | 413 | /* cr4 is used to enable and disable PGE, but we don't care. */ |
07ad157f RR |
414 | static unsigned long lguest_read_cr4(void) |
415 | { | |
416 | return 0; | |
417 | } | |
418 | ||
419 | static void lguest_write_cr4(unsigned long val) | |
420 | { | |
421 | } | |
422 | ||
b2b47c21 RR |
423 | /* |
424 | * Page Table Handling. | |
425 | * | |
426 | * Now would be a good time to take a rest and grab a coffee or similarly | |
427 | * relaxing stimulant. The easy parts are behind us, and the trek gradually | |
428 | * winds uphill from here. | |
429 | * | |
430 | * Quick refresher: memory is divided into "pages" of 4096 bytes each. The CPU | |
431 | * maps virtual addresses to physical addresses using "page tables". We could | |
432 | * use one huge index of 1 million entries: each address is 4 bytes, so that's | |
433 | * 1024 pages just to hold the page tables. But since most virtual addresses | |
e1e72965 | 434 | * are unused, we use a two level index which saves space. The cr3 register |
b2b47c21 RR |
435 | * contains the physical address of the top level "page directory" page, which |
436 | * contains physical addresses of up to 1024 second-level pages. Each of these | |
437 | * second level pages contains up to 1024 physical addresses of actual pages, | |
438 | * or Page Table Entries (PTEs). | |
439 | * | |
440 | * Here's a diagram, where arrows indicate physical addresses: | |
441 | * | |
e1e72965 | 442 | * cr3 ---> +---------+ |
b2b47c21 RR |
443 | * | --------->+---------+ |
444 | * | | | PADDR1 | | |
445 | * Top-level | | PADDR2 | | |
446 | * (PMD) page | | | | |
447 | * | | Lower-level | | |
448 | * | | (PTE) page | | |
449 | * | | | | | |
450 | * .... .... | |
451 | * | |
452 | * So to convert a virtual address to a physical address, we look up the top | |
453 | * level, which points us to the second level, which gives us the physical | |
454 | * address of that page. If the top level entry was not present, or the second | |
455 | * level entry was not present, then the virtual address is invalid (we | |
456 | * say "the page was not mapped"). | |
457 | * | |
458 | * Put another way, a 32-bit virtual address is divided up like so: | |
459 | * | |
460 | * 1 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 | |
461 | * |<---- 10 bits ---->|<---- 10 bits ---->|<------ 12 bits ------>| | |
462 | * Index into top Index into second Offset within page | |
463 | * page directory page pagetable page | |
464 | * | |
465 | * The kernel spends a lot of time changing both the top-level page directory | |
466 | * and lower-level pagetable pages. The Guest doesn't know physical addresses, | |
467 | * so while it maintains these page tables exactly like normal, it also needs | |
468 | * to keep the Host informed whenever it makes a change: the Host will create | |
469 | * the real page tables based on the Guests'. | |
470 | */ | |
471 | ||
472 | /* The Guest calls this to set a second-level entry (pte), ie. to map a page | |
473 | * into a process' address space. We set the entry then tell the Host the | |
474 | * toplevel and address this corresponds to. The Guest uses one pagetable per | |
475 | * process, so we need to tell the Host which one we're changing (mm->pgd). */ | |
07ad157f RR |
476 | static void lguest_set_pte_at(struct mm_struct *mm, unsigned long addr, |
477 | pte_t *ptep, pte_t pteval) | |
478 | { | |
479 | *ptep = pteval; | |
480 | lazy_hcall(LHCALL_SET_PTE, __pa(mm->pgd), addr, pteval.pte_low); | |
481 | } | |
482 | ||
b2b47c21 RR |
483 | /* The Guest calls this to set a top-level entry. Again, we set the entry then |
484 | * tell the Host which top-level page we changed, and the index of the entry we | |
485 | * changed. */ | |
07ad157f RR |
486 | static void lguest_set_pmd(pmd_t *pmdp, pmd_t pmdval) |
487 | { | |
488 | *pmdp = pmdval; | |
489 | lazy_hcall(LHCALL_SET_PMD, __pa(pmdp)&PAGE_MASK, | |
490 | (__pa(pmdp)&(PAGE_SIZE-1))/4, 0); | |
491 | } | |
492 | ||
b2b47c21 RR |
493 | /* There are a couple of legacy places where the kernel sets a PTE, but we |
494 | * don't know the top level any more. This is useless for us, since we don't | |
495 | * know which pagetable is changing or what address, so we just tell the Host | |
496 | * to forget all of them. Fortunately, this is very rare. | |
497 | * | |
498 | * ... except in early boot when the kernel sets up the initial pagetables, | |
499 | * which makes booting astonishingly slow. So we don't even tell the Host | |
e1e72965 | 500 | * anything changed until we've done the first page table switch. */ |
07ad157f RR |
501 | static void lguest_set_pte(pte_t *ptep, pte_t pteval) |
502 | { | |
503 | *ptep = pteval; | |
504 | /* Don't bother with hypercall before initial setup. */ | |
505 | if (current_cr3) | |
506 | lazy_hcall(LHCALL_FLUSH_TLB, 1, 0, 0); | |
507 | } | |
508 | ||
93b1eab3 | 509 | /* Unfortunately for Lguest, the pv_mmu_ops for page tables were based on |
b2b47c21 RR |
510 | * native page table operations. On native hardware you can set a new page |
511 | * table entry whenever you want, but if you want to remove one you have to do | |
512 | * a TLB flush (a TLB is a little cache of page table entries kept by the CPU). | |
513 | * | |
514 | * So the lguest_set_pte_at() and lguest_set_pmd() functions above are only | |
515 | * called when a valid entry is written, not when it's removed (ie. marked not | |
516 | * present). Instead, this is where we come when the Guest wants to remove a | |
517 | * page table entry: we tell the Host to set that entry to 0 (ie. the present | |
518 | * bit is zero). */ | |
07ad157f RR |
519 | static void lguest_flush_tlb_single(unsigned long addr) |
520 | { | |
b2b47c21 | 521 | /* Simply set it to zero: if it was not, it will fault back in. */ |
07ad157f RR |
522 | lazy_hcall(LHCALL_SET_PTE, current_cr3, addr, 0); |
523 | } | |
524 | ||
b2b47c21 RR |
525 | /* This is what happens after the Guest has removed a large number of entries. |
526 | * This tells the Host that any of the page table entries for userspace might | |
527 | * have changed, ie. virtual addresses below PAGE_OFFSET. */ | |
07ad157f RR |
528 | static void lguest_flush_tlb_user(void) |
529 | { | |
530 | lazy_hcall(LHCALL_FLUSH_TLB, 0, 0, 0); | |
531 | } | |
532 | ||
b2b47c21 RR |
533 | /* This is called when the kernel page tables have changed. That's not very |
534 | * common (unless the Guest is using highmem, which makes the Guest extremely | |
535 | * slow), so it's worth separating this from the user flushing above. */ | |
07ad157f RR |
536 | static void lguest_flush_tlb_kernel(void) |
537 | { | |
538 | lazy_hcall(LHCALL_FLUSH_TLB, 1, 0, 0); | |
539 | } | |
540 | ||
b2b47c21 RR |
541 | /* |
542 | * The Unadvanced Programmable Interrupt Controller. | |
543 | * | |
544 | * This is an attempt to implement the simplest possible interrupt controller. | |
545 | * I spent some time looking though routines like set_irq_chip_and_handler, | |
546 | * set_irq_chip_and_handler_name, set_irq_chip_data and set_phasers_to_stun and | |
547 | * I *think* this is as simple as it gets. | |
548 | * | |
549 | * We can tell the Host what interrupts we want blocked ready for using the | |
550 | * lguest_data.interrupts bitmap, so disabling (aka "masking") them is as | |
551 | * simple as setting a bit. We don't actually "ack" interrupts as such, we | |
552 | * just mask and unmask them. I wonder if we should be cleverer? | |
553 | */ | |
07ad157f RR |
554 | static void disable_lguest_irq(unsigned int irq) |
555 | { | |
556 | set_bit(irq, lguest_data.blocked_interrupts); | |
557 | } | |
558 | ||
559 | static void enable_lguest_irq(unsigned int irq) | |
560 | { | |
561 | clear_bit(irq, lguest_data.blocked_interrupts); | |
07ad157f RR |
562 | } |
563 | ||
b2b47c21 | 564 | /* This structure describes the lguest IRQ controller. */ |
07ad157f RR |
565 | static struct irq_chip lguest_irq_controller = { |
566 | .name = "lguest", | |
567 | .mask = disable_lguest_irq, | |
568 | .mask_ack = disable_lguest_irq, | |
569 | .unmask = enable_lguest_irq, | |
570 | }; | |
571 | ||
b2b47c21 RR |
572 | /* This sets up the Interrupt Descriptor Table (IDT) entry for each hardware |
573 | * interrupt (except 128, which is used for system calls), and then tells the | |
574 | * Linux infrastructure that each interrupt is controlled by our level-based | |
575 | * lguest interrupt controller. */ | |
07ad157f RR |
576 | static void __init lguest_init_IRQ(void) |
577 | { | |
578 | unsigned int i; | |
579 | ||
580 | for (i = 0; i < LGUEST_IRQS; i++) { | |
581 | int vector = FIRST_EXTERNAL_VECTOR + i; | |
582 | if (vector != SYSCALL_VECTOR) { | |
583 | set_intr_gate(vector, interrupt[i]); | |
584 | set_irq_chip_and_handler(i, &lguest_irq_controller, | |
585 | handle_level_irq); | |
586 | } | |
587 | } | |
b2b47c21 RR |
588 | /* This call is required to set up for 4k stacks, where we have |
589 | * separate stacks for hard and soft interrupts. */ | |
07ad157f RR |
590 | irq_ctx_init(smp_processor_id()); |
591 | } | |
592 | ||
b2b47c21 RR |
593 | /* |
594 | * Time. | |
595 | * | |
596 | * It would be far better for everyone if the Guest had its own clock, but | |
6c8dca5d | 597 | * until then the Host gives us the time on every interrupt. |
b2b47c21 | 598 | */ |
07ad157f RR |
599 | static unsigned long lguest_get_wallclock(void) |
600 | { | |
6c8dca5d | 601 | return lguest_data.time.tv_sec; |
07ad157f RR |
602 | } |
603 | ||
d7e28ffe RR |
604 | static cycle_t lguest_clock_read(void) |
605 | { | |
6c8dca5d RR |
606 | unsigned long sec, nsec; |
607 | ||
608 | /* If the Host tells the TSC speed, we can trust that. */ | |
d7e28ffe RR |
609 | if (lguest_data.tsc_khz) |
610 | return native_read_tsc(); | |
6c8dca5d RR |
611 | |
612 | /* If we can't use the TSC, we read the time value written by the Host. | |
613 | * Since it's in two parts (seconds and nanoseconds), we risk reading | |
614 | * it just as it's changing from 99 & 0.999999999 to 100 and 0, and | |
615 | * getting 99 and 0. As Linux tends to come apart under the stress of | |
616 | * time travel, we must be careful: */ | |
617 | do { | |
618 | /* First we read the seconds part. */ | |
619 | sec = lguest_data.time.tv_sec; | |
620 | /* This read memory barrier tells the compiler and the CPU that | |
621 | * this can't be reordered: we have to complete the above | |
622 | * before going on. */ | |
623 | rmb(); | |
624 | /* Now we read the nanoseconds part. */ | |
625 | nsec = lguest_data.time.tv_nsec; | |
626 | /* Make sure we've done that. */ | |
627 | rmb(); | |
628 | /* Now if the seconds part has changed, try again. */ | |
629 | } while (unlikely(lguest_data.time.tv_sec != sec)); | |
630 | ||
631 | /* Our non-TSC clock is in real nanoseconds. */ | |
632 | return sec*1000000000ULL + nsec; | |
d7e28ffe RR |
633 | } |
634 | ||
635 | /* This is what we tell the kernel is our clocksource. */ | |
636 | static struct clocksource lguest_clock = { | |
637 | .name = "lguest", | |
638 | .rating = 400, | |
639 | .read = lguest_clock_read, | |
6c8dca5d | 640 | .mask = CLOCKSOURCE_MASK(64), |
37250097 RR |
641 | .mult = 1 << 22, |
642 | .shift = 22, | |
05aa026a | 643 | .flags = CLOCK_SOURCE_IS_CONTINUOUS, |
d7e28ffe RR |
644 | }; |
645 | ||
6c8dca5d | 646 | /* The "scheduler clock" is just our real clock, adjusted to start at zero */ |
9d1ca6f1 RR |
647 | static unsigned long long lguest_sched_clock(void) |
648 | { | |
649 | return cyc2ns(&lguest_clock, lguest_clock_read() - clock_base); | |
650 | } | |
651 | ||
d7e28ffe RR |
652 | /* We also need a "struct clock_event_device": Linux asks us to set it to go |
653 | * off some time in the future. Actually, James Morris figured all this out, I | |
654 | * just applied the patch. */ | |
655 | static int lguest_clockevent_set_next_event(unsigned long delta, | |
656 | struct clock_event_device *evt) | |
657 | { | |
658 | if (delta < LG_CLOCK_MIN_DELTA) { | |
659 | if (printk_ratelimit()) | |
660 | printk(KERN_DEBUG "%s: small delta %lu ns\n", | |
661 | __FUNCTION__, delta); | |
662 | return -ETIME; | |
663 | } | |
664 | hcall(LHCALL_SET_CLOCKEVENT, delta, 0, 0); | |
665 | return 0; | |
666 | } | |
667 | ||
668 | static void lguest_clockevent_set_mode(enum clock_event_mode mode, | |
669 | struct clock_event_device *evt) | |
670 | { | |
671 | switch (mode) { | |
672 | case CLOCK_EVT_MODE_UNUSED: | |
673 | case CLOCK_EVT_MODE_SHUTDOWN: | |
674 | /* A 0 argument shuts the clock down. */ | |
675 | hcall(LHCALL_SET_CLOCKEVENT, 0, 0, 0); | |
676 | break; | |
677 | case CLOCK_EVT_MODE_ONESHOT: | |
678 | /* This is what we expect. */ | |
679 | break; | |
680 | case CLOCK_EVT_MODE_PERIODIC: | |
681 | BUG(); | |
18de5bc4 TG |
682 | case CLOCK_EVT_MODE_RESUME: |
683 | break; | |
d7e28ffe RR |
684 | } |
685 | } | |
686 | ||
687 | /* This describes our primitive timer chip. */ | |
688 | static struct clock_event_device lguest_clockevent = { | |
689 | .name = "lguest", | |
690 | .features = CLOCK_EVT_FEAT_ONESHOT, | |
691 | .set_next_event = lguest_clockevent_set_next_event, | |
692 | .set_mode = lguest_clockevent_set_mode, | |
693 | .rating = INT_MAX, | |
694 | .mult = 1, | |
695 | .shift = 0, | |
696 | .min_delta_ns = LG_CLOCK_MIN_DELTA, | |
697 | .max_delta_ns = LG_CLOCK_MAX_DELTA, | |
698 | }; | |
699 | ||
700 | /* This is the Guest timer interrupt handler (hardware interrupt 0). We just | |
701 | * call the clockevent infrastructure and it does whatever needs doing. */ | |
07ad157f RR |
702 | static void lguest_time_irq(unsigned int irq, struct irq_desc *desc) |
703 | { | |
d7e28ffe RR |
704 | unsigned long flags; |
705 | ||
706 | /* Don't interrupt us while this is running. */ | |
707 | local_irq_save(flags); | |
708 | lguest_clockevent.event_handler(&lguest_clockevent); | |
709 | local_irq_restore(flags); | |
07ad157f RR |
710 | } |
711 | ||
b2b47c21 RR |
712 | /* At some point in the boot process, we get asked to set up our timing |
713 | * infrastructure. The kernel doesn't expect timer interrupts before this, but | |
714 | * we cleverly initialized the "blocked_interrupts" field of "struct | |
715 | * lguest_data" so that timer interrupts were blocked until now. */ | |
07ad157f RR |
716 | static void lguest_time_init(void) |
717 | { | |
b2b47c21 | 718 | /* Set up the timer interrupt (0) to go to our simple timer routine */ |
07ad157f | 719 | set_irq_handler(0, lguest_time_irq); |
07ad157f | 720 | |
e1e72965 RR |
721 | /* Our clock structure looks like arch/x86/kernel/tsc_32.c if we can |
722 | * use the TSC, otherwise it's a dumb nanosecond-resolution clock. | |
723 | * Either way, the "rating" is set so high that it's always chosen over | |
724 | * any other clocksource. */ | |
05aa026a | 725 | if (lguest_data.tsc_khz) |
d7e28ffe RR |
726 | lguest_clock.mult = clocksource_khz2mult(lguest_data.tsc_khz, |
727 | lguest_clock.shift); | |
9d1ca6f1 | 728 | clock_base = lguest_clock_read(); |
d7e28ffe RR |
729 | clocksource_register(&lguest_clock); |
730 | ||
6c8dca5d | 731 | /* Now we've set up our clock, we can use it as the scheduler clock */ |
93b1eab3 | 732 | pv_time_ops.sched_clock = lguest_sched_clock; |
6c8dca5d | 733 | |
b2b47c21 RR |
734 | /* We can't set cpumask in the initializer: damn C limitations! Set it |
735 | * here and register our timer device. */ | |
d7e28ffe RR |
736 | lguest_clockevent.cpumask = cpumask_of_cpu(0); |
737 | clockevents_register_device(&lguest_clockevent); | |
738 | ||
b2b47c21 | 739 | /* Finally, we unblock the timer interrupt. */ |
d7e28ffe | 740 | enable_lguest_irq(0); |
07ad157f RR |
741 | } |
742 | ||
b2b47c21 RR |
743 | /* |
744 | * Miscellaneous bits and pieces. | |
745 | * | |
746 | * Here is an oddball collection of functions which the Guest needs for things | |
747 | * to work. They're pretty simple. | |
748 | */ | |
749 | ||
e1e72965 | 750 | /* The Guest needs to tell the Host what stack it expects traps to use. For |
b2b47c21 RR |
751 | * native hardware, this is part of the Task State Segment mentioned above in |
752 | * lguest_load_tr_desc(), but to help hypervisors there's this special call. | |
753 | * | |
754 | * We tell the Host the segment we want to use (__KERNEL_DS is the kernel data | |
755 | * segment), the privilege level (we're privilege level 1, the Host is 0 and | |
756 | * will not tolerate us trying to use that), the stack pointer, and the number | |
757 | * of pages in the stack. */ | |
faca6227 | 758 | static void lguest_load_sp0(struct tss_struct *tss, |
07ad157f RR |
759 | struct thread_struct *thread) |
760 | { | |
faca6227 | 761 | lazy_hcall(LHCALL_SET_STACK, __KERNEL_DS|0x1, thread->sp0, |
07ad157f RR |
762 | THREAD_SIZE/PAGE_SIZE); |
763 | } | |
764 | ||
b2b47c21 | 765 | /* Let's just say, I wouldn't do debugging under a Guest. */ |
07ad157f RR |
766 | static void lguest_set_debugreg(int regno, unsigned long value) |
767 | { | |
768 | /* FIXME: Implement */ | |
769 | } | |
770 | ||
b2b47c21 RR |
771 | /* There are times when the kernel wants to make sure that no memory writes are |
772 | * caught in the cache (that they've all reached real hardware devices). This | |
773 | * doesn't matter for the Guest which has virtual hardware. | |
774 | * | |
775 | * On the Pentium 4 and above, cpuid() indicates that the Cache Line Flush | |
776 | * (clflush) instruction is available and the kernel uses that. Otherwise, it | |
777 | * uses the older "Write Back and Invalidate Cache" (wbinvd) instruction. | |
778 | * Unlike clflush, wbinvd can only be run at privilege level 0. So we can | |
779 | * ignore clflush, but replace wbinvd. | |
780 | */ | |
07ad157f RR |
781 | static void lguest_wbinvd(void) |
782 | { | |
783 | } | |
784 | ||
b2b47c21 RR |
785 | /* If the Guest expects to have an Advanced Programmable Interrupt Controller, |
786 | * we play dumb by ignoring writes and returning 0 for reads. So it's no | |
787 | * longer Programmable nor Controlling anything, and I don't think 8 lines of | |
788 | * code qualifies for Advanced. It will also never interrupt anything. It | |
789 | * does, however, allow us to get through the Linux boot code. */ | |
07ad157f | 790 | #ifdef CONFIG_X86_LOCAL_APIC |
42e0a9aa | 791 | static void lguest_apic_write(unsigned long reg, u32 v) |
07ad157f RR |
792 | { |
793 | } | |
794 | ||
42e0a9aa | 795 | static u32 lguest_apic_read(unsigned long reg) |
07ad157f RR |
796 | { |
797 | return 0; | |
798 | } | |
799 | #endif | |
800 | ||
b2b47c21 | 801 | /* STOP! Until an interrupt comes in. */ |
07ad157f RR |
802 | static void lguest_safe_halt(void) |
803 | { | |
804 | hcall(LHCALL_HALT, 0, 0, 0); | |
805 | } | |
806 | ||
b2b47c21 RR |
807 | /* Perhaps CRASH isn't the best name for this hypercall, but we use it to get a |
808 | * message out when we're crashing as well as elegant termination like powering | |
809 | * off. | |
810 | * | |
811 | * Note that the Host always prefers that the Guest speak in physical addresses | |
812 | * rather than virtual addresses, so we use __pa() here. */ | |
07ad157f RR |
813 | static void lguest_power_off(void) |
814 | { | |
815 | hcall(LHCALL_CRASH, __pa("Power down"), 0, 0); | |
816 | } | |
817 | ||
b2b47c21 RR |
818 | /* |
819 | * Panicing. | |
820 | * | |
821 | * Don't. But if you did, this is what happens. | |
822 | */ | |
07ad157f RR |
823 | static int lguest_panic(struct notifier_block *nb, unsigned long l, void *p) |
824 | { | |
825 | hcall(LHCALL_CRASH, __pa(p), 0, 0); | |
b2b47c21 | 826 | /* The hcall won't return, but to keep gcc happy, we're "done". */ |
07ad157f RR |
827 | return NOTIFY_DONE; |
828 | } | |
829 | ||
830 | static struct notifier_block paniced = { | |
831 | .notifier_call = lguest_panic | |
832 | }; | |
833 | ||
b2b47c21 | 834 | /* Setting up memory is fairly easy. */ |
07ad157f RR |
835 | static __init char *lguest_memory_setup(void) |
836 | { | |
b2b47c21 RR |
837 | /* We do this here and not earlier because lockcheck barfs if we do it |
838 | * before start_kernel() */ | |
07ad157f RR |
839 | atomic_notifier_chain_register(&panic_notifier_list, &paniced); |
840 | ||
b2b47c21 RR |
841 | /* The Linux bootloader header contains an "e820" memory map: the |
842 | * Launcher populated the first entry with our memory limit. */ | |
30c82645 PA |
843 | add_memory_region(boot_params.e820_map[0].addr, |
844 | boot_params.e820_map[0].size, | |
845 | boot_params.e820_map[0].type); | |
b2b47c21 RR |
846 | |
847 | /* This string is for the boot messages. */ | |
07ad157f RR |
848 | return "LGUEST"; |
849 | } | |
850 | ||
e1e72965 RR |
851 | /* We will eventually use the virtio console device to produce console output, |
852 | * but before that is set up we use LHCALL_NOTIFY on normal memory to produce | |
853 | * console output. */ | |
19f1537b RR |
854 | static __init int early_put_chars(u32 vtermno, const char *buf, int count) |
855 | { | |
856 | char scratch[17]; | |
857 | unsigned int len = count; | |
858 | ||
e1e72965 RR |
859 | /* We use a nul-terminated string, so we have to make a copy. Icky, |
860 | * huh? */ | |
19f1537b RR |
861 | if (len > sizeof(scratch) - 1) |
862 | len = sizeof(scratch) - 1; | |
863 | scratch[len] = '\0'; | |
864 | memcpy(scratch, buf, len); | |
865 | hcall(LHCALL_NOTIFY, __pa(scratch), 0, 0); | |
866 | ||
867 | /* This routine returns the number of bytes actually written. */ | |
868 | return len; | |
869 | } | |
870 | ||
b2b47c21 RR |
871 | /*G:050 |
872 | * Patching (Powerfully Placating Performance Pedants) | |
873 | * | |
93b1eab3 | 874 | * We have already seen that pv_ops structures let us replace simple |
b2b47c21 RR |
875 | * native instructions with calls to the appropriate back end all throughout |
876 | * the kernel. This allows the same kernel to run as a Guest and as a native | |
877 | * kernel, but it's slow because of all the indirect branches. | |
878 | * | |
879 | * Remember that David Wheeler quote about "Any problem in computer science can | |
880 | * be solved with another layer of indirection"? The rest of that quote is | |
881 | * "... But that usually will create another problem." This is the first of | |
882 | * those problems. | |
883 | * | |
884 | * Our current solution is to allow the paravirt back end to optionally patch | |
885 | * over the indirect calls to replace them with something more efficient. We | |
886 | * patch the four most commonly called functions: disable interrupts, enable | |
e1e72965 | 887 | * interrupts, restore interrupts and save interrupts. We usually have 6 or 10 |
b2b47c21 RR |
888 | * bytes to patch into: the Guest versions of these operations are small enough |
889 | * that we can fit comfortably. | |
890 | * | |
891 | * First we need assembly templates of each of the patchable Guest operations, | |
892 | * and these are in lguest_asm.S. */ | |
893 | ||
894 | /*G:060 We construct a table from the assembler templates: */ | |
07ad157f RR |
895 | static const struct lguest_insns |
896 | { | |
897 | const char *start, *end; | |
898 | } lguest_insns[] = { | |
93b1eab3 JF |
899 | [PARAVIRT_PATCH(pv_irq_ops.irq_disable)] = { lgstart_cli, lgend_cli }, |
900 | [PARAVIRT_PATCH(pv_irq_ops.irq_enable)] = { lgstart_sti, lgend_sti }, | |
901 | [PARAVIRT_PATCH(pv_irq_ops.restore_fl)] = { lgstart_popf, lgend_popf }, | |
902 | [PARAVIRT_PATCH(pv_irq_ops.save_fl)] = { lgstart_pushf, lgend_pushf }, | |
07ad157f | 903 | }; |
b2b47c21 RR |
904 | |
905 | /* Now our patch routine is fairly simple (based on the native one in | |
906 | * paravirt.c). If we have a replacement, we copy it in and return how much of | |
907 | * the available space we used. */ | |
ab144f5e AK |
908 | static unsigned lguest_patch(u8 type, u16 clobber, void *ibuf, |
909 | unsigned long addr, unsigned len) | |
07ad157f RR |
910 | { |
911 | unsigned int insn_len; | |
912 | ||
b2b47c21 | 913 | /* Don't do anything special if we don't have a replacement */ |
07ad157f | 914 | if (type >= ARRAY_SIZE(lguest_insns) || !lguest_insns[type].start) |
ab144f5e | 915 | return paravirt_patch_default(type, clobber, ibuf, addr, len); |
07ad157f RR |
916 | |
917 | insn_len = lguest_insns[type].end - lguest_insns[type].start; | |
918 | ||
b2b47c21 RR |
919 | /* Similarly if we can't fit replacement (shouldn't happen, but let's |
920 | * be thorough). */ | |
07ad157f | 921 | if (len < insn_len) |
ab144f5e | 922 | return paravirt_patch_default(type, clobber, ibuf, addr, len); |
07ad157f | 923 | |
b2b47c21 | 924 | /* Copy in our instructions. */ |
ab144f5e | 925 | memcpy(ibuf, lguest_insns[type].start, insn_len); |
07ad157f RR |
926 | return insn_len; |
927 | } | |
928 | ||
93b1eab3 JF |
929 | /*G:030 Once we get to lguest_init(), we know we're a Guest. The pv_ops |
930 | * structures in the kernel provide points for (almost) every routine we have | |
931 | * to override to avoid privileged instructions. */ | |
814a0e5c | 932 | __init void lguest_init(void) |
07ad157f | 933 | { |
b2b47c21 RR |
934 | /* We're under lguest, paravirt is enabled, and we're running at |
935 | * privilege level 1, not 0 as normal. */ | |
93b1eab3 JF |
936 | pv_info.name = "lguest"; |
937 | pv_info.paravirt_enabled = 1; | |
938 | pv_info.kernel_rpl = 1; | |
07ad157f | 939 | |
b2b47c21 RR |
940 | /* We set up all the lguest overrides for sensitive operations. These |
941 | * are detailed with the operations themselves. */ | |
93b1eab3 JF |
942 | |
943 | /* interrupt-related operations */ | |
944 | pv_irq_ops.init_IRQ = lguest_init_IRQ; | |
945 | pv_irq_ops.save_fl = save_fl; | |
946 | pv_irq_ops.restore_fl = restore_fl; | |
947 | pv_irq_ops.irq_disable = irq_disable; | |
948 | pv_irq_ops.irq_enable = irq_enable; | |
949 | pv_irq_ops.safe_halt = lguest_safe_halt; | |
950 | ||
951 | /* init-time operations */ | |
952 | pv_init_ops.memory_setup = lguest_memory_setup; | |
953 | pv_init_ops.patch = lguest_patch; | |
954 | ||
955 | /* Intercepts of various cpu instructions */ | |
956 | pv_cpu_ops.load_gdt = lguest_load_gdt; | |
957 | pv_cpu_ops.cpuid = lguest_cpuid; | |
958 | pv_cpu_ops.load_idt = lguest_load_idt; | |
959 | pv_cpu_ops.iret = lguest_iret; | |
faca6227 | 960 | pv_cpu_ops.load_sp0 = lguest_load_sp0; |
93b1eab3 JF |
961 | pv_cpu_ops.load_tr_desc = lguest_load_tr_desc; |
962 | pv_cpu_ops.set_ldt = lguest_set_ldt; | |
963 | pv_cpu_ops.load_tls = lguest_load_tls; | |
964 | pv_cpu_ops.set_debugreg = lguest_set_debugreg; | |
965 | pv_cpu_ops.clts = lguest_clts; | |
966 | pv_cpu_ops.read_cr0 = lguest_read_cr0; | |
967 | pv_cpu_ops.write_cr0 = lguest_write_cr0; | |
968 | pv_cpu_ops.read_cr4 = lguest_read_cr4; | |
969 | pv_cpu_ops.write_cr4 = lguest_write_cr4; | |
970 | pv_cpu_ops.write_gdt_entry = lguest_write_gdt_entry; | |
971 | pv_cpu_ops.write_idt_entry = lguest_write_idt_entry; | |
972 | pv_cpu_ops.wbinvd = lguest_wbinvd; | |
8965c1c0 JF |
973 | pv_cpu_ops.lazy_mode.enter = paravirt_enter_lazy_cpu; |
974 | pv_cpu_ops.lazy_mode.leave = lguest_leave_lazy_mode; | |
93b1eab3 JF |
975 | |
976 | /* pagetable management */ | |
977 | pv_mmu_ops.write_cr3 = lguest_write_cr3; | |
978 | pv_mmu_ops.flush_tlb_user = lguest_flush_tlb_user; | |
979 | pv_mmu_ops.flush_tlb_single = lguest_flush_tlb_single; | |
980 | pv_mmu_ops.flush_tlb_kernel = lguest_flush_tlb_kernel; | |
981 | pv_mmu_ops.set_pte = lguest_set_pte; | |
982 | pv_mmu_ops.set_pte_at = lguest_set_pte_at; | |
983 | pv_mmu_ops.set_pmd = lguest_set_pmd; | |
984 | pv_mmu_ops.read_cr2 = lguest_read_cr2; | |
985 | pv_mmu_ops.read_cr3 = lguest_read_cr3; | |
8965c1c0 JF |
986 | pv_mmu_ops.lazy_mode.enter = paravirt_enter_lazy_mmu; |
987 | pv_mmu_ops.lazy_mode.leave = lguest_leave_lazy_mode; | |
93b1eab3 | 988 | |
07ad157f | 989 | #ifdef CONFIG_X86_LOCAL_APIC |
93b1eab3 JF |
990 | /* apic read/write intercepts */ |
991 | pv_apic_ops.apic_write = lguest_apic_write; | |
992 | pv_apic_ops.apic_write_atomic = lguest_apic_write; | |
993 | pv_apic_ops.apic_read = lguest_apic_read; | |
07ad157f | 994 | #endif |
93b1eab3 JF |
995 | |
996 | /* time operations */ | |
997 | pv_time_ops.get_wallclock = lguest_get_wallclock; | |
998 | pv_time_ops.time_init = lguest_time_init; | |
999 | ||
b2b47c21 RR |
1000 | /* Now is a good time to look at the implementations of these functions |
1001 | * before returning to the rest of lguest_init(). */ | |
1002 | ||
1003 | /*G:070 Now we've seen all the paravirt_ops, we return to | |
1004 | * lguest_init() where the rest of the fairly chaotic boot setup | |
47436aa4 | 1005 | * occurs. */ |
07ad157f | 1006 | |
b2b47c21 RR |
1007 | /* The native boot code sets up initial page tables immediately after |
1008 | * the kernel itself, and sets init_pg_tables_end so they're not | |
1009 | * clobbered. The Launcher places our initial pagetables somewhere at | |
1010 | * the top of our physical memory, so we don't need extra space: set | |
1011 | * init_pg_tables_end to the end of the kernel. */ | |
07ad157f RR |
1012 | init_pg_tables_end = __pa(pg0); |
1013 | ||
b2b47c21 RR |
1014 | /* Load the %fs segment register (the per-cpu segment register) with |
1015 | * the normal data segment to get through booting. */ | |
07ad157f RR |
1016 | asm volatile ("mov %0, %%fs" : : "r" (__KERNEL_DS) : "memory"); |
1017 | ||
b2b47c21 | 1018 | /* The Host uses the top of the Guest's virtual address space for the |
e1e72965 | 1019 | * Host<->Guest Switcher, and it tells us how big that is in |
b2b47c21 | 1020 | * lguest_data.reserve_mem, set up on the LGUEST_INIT hypercall. */ |
07ad157f RR |
1021 | reserve_top_address(lguest_data.reserve_mem); |
1022 | ||
b2b47c21 RR |
1023 | /* If we don't initialize the lock dependency checker now, it crashes |
1024 | * paravirt_disable_iospace. */ | |
07ad157f RR |
1025 | lockdep_init(); |
1026 | ||
b2b47c21 RR |
1027 | /* The IDE code spends about 3 seconds probing for disks: if we reserve |
1028 | * all the I/O ports up front it can't get them and so doesn't probe. | |
1029 | * Other device drivers are similar (but less severe). This cuts the | |
1030 | * kernel boot time on my machine from 4.1 seconds to 0.45 seconds. */ | |
07ad157f RR |
1031 | paravirt_disable_iospace(); |
1032 | ||
b2b47c21 RR |
1033 | /* This is messy CPU setup stuff which the native boot code does before |
1034 | * start_kernel, so we have to do, too: */ | |
07ad157f RR |
1035 | cpu_detect(&new_cpu_data); |
1036 | /* head.S usually sets up the first capability word, so do it here. */ | |
1037 | new_cpu_data.x86_capability[0] = cpuid_edx(1); | |
1038 | ||
1039 | /* Math is always hard! */ | |
1040 | new_cpu_data.hard_math = 1; | |
1041 | ||
1042 | #ifdef CONFIG_X86_MCE | |
1043 | mce_disabled = 1; | |
1044 | #endif | |
07ad157f RR |
1045 | #ifdef CONFIG_ACPI |
1046 | acpi_disabled = 1; | |
1047 | acpi_ht = 0; | |
1048 | #endif | |
1049 | ||
b2b47c21 RR |
1050 | /* We set the perferred console to "hvc". This is the "hypervisor |
1051 | * virtual console" driver written by the PowerPC people, which we also | |
1052 | * adapted for lguest's use. */ | |
07ad157f RR |
1053 | add_preferred_console("hvc", 0, NULL); |
1054 | ||
19f1537b RR |
1055 | /* Register our very early console. */ |
1056 | virtio_cons_early_init(early_put_chars); | |
1057 | ||
b2b47c21 RR |
1058 | /* Last of all, we set the power management poweroff hook to point to |
1059 | * the Guest routine to power off. */ | |
07ad157f | 1060 | pm_power_off = lguest_power_off; |
b2b47c21 RR |
1061 | |
1062 | /* Now we're set up, call start_kernel() in init/main.c and we proceed | |
1063 | * to boot as normal. It never returns. */ | |
07ad157f RR |
1064 | start_kernel(); |
1065 | } | |
b2b47c21 RR |
1066 | /* |
1067 | * This marks the end of stage II of our journey, The Guest. | |
1068 | * | |
e1e72965 RR |
1069 | * It is now time for us to explore the layer of virtual drivers and complete |
1070 | * our understanding of the Guest in "make Drivers". | |
b2b47c21 | 1071 | */ |