blob: 482aec2a96318e854aede3a385cf34eedc1da95f [file] [log] [blame]
Jes Sorensen625efab2007-10-22 11:03:28 +10001/*
2 * Copyright (C) 2006, Rusty Russell <rusty@rustcorp.com.au> IBM Corporation.
3 * Copyright (C) 2007, Jes Sorensen <jes@sgi.com> SGI.
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
13 * NON INFRINGEMENT. See the GNU General Public License for more
14 * details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
19 */
20#include <linux/kernel.h>
21#include <linux/start_kernel.h>
22#include <linux/string.h>
23#include <linux/console.h>
24#include <linux/screen_info.h>
25#include <linux/irq.h>
26#include <linux/interrupt.h>
27#include <linux/clocksource.h>
28#include <linux/clockchips.h>
29#include <linux/cpu.h>
30#include <linux/lguest.h>
31#include <linux/lguest_launcher.h>
Jes Sorensen625efab2007-10-22 11:03:28 +100032#include <asm/paravirt.h>
33#include <asm/param.h>
34#include <asm/page.h>
35#include <asm/pgtable.h>
36#include <asm/desc.h>
37#include <asm/setup.h>
38#include <asm/lguest.h>
39#include <asm/uaccess.h>
40#include <asm/i387.h>
41#include "../lg.h"
42
43static int cpu_had_pge;
44
45static struct {
46 unsigned long offset;
47 unsigned short segment;
48} lguest_entry;
49
50/* Offset from where switcher.S was compiled to where we've copied it */
51static unsigned long switcher_offset(void)
52{
53 return SWITCHER_ADDR - (unsigned long)start_switcher_text;
54}
55
56/* This cpu's struct lguest_pages. */
57static struct lguest_pages *lguest_pages(unsigned int cpu)
58{
59 return &(((struct lguest_pages *)
60 (SWITCHER_ADDR + SHARED_SWITCHER_PAGES*PAGE_SIZE))[cpu]);
61}
62
63static DEFINE_PER_CPU(struct lguest *, last_guest);
64
65/*S:010
Rusty Russelle1e72962007-10-25 15:02:50 +100066 * We approach the Switcher.
Jes Sorensen625efab2007-10-22 11:03:28 +100067 *
68 * Remember that each CPU has two pages which are visible to the Guest when it
69 * runs on that CPU. This has to contain the state for that Guest: we copy the
70 * state in just before we run the Guest.
71 *
72 * Each Guest has "changed" flags which indicate what has changed in the Guest
73 * since it last ran. We saw this set in interrupts_and_traps.c and
74 * segments.c.
75 */
76static void copy_in_guest_info(struct lguest *lg, struct lguest_pages *pages)
77{
78 /* Copying all this data can be quite expensive. We usually run the
79 * same Guest we ran last time (and that Guest hasn't run anywhere else
80 * meanwhile). If that's not the case, we pretend everything in the
81 * Guest has changed. */
82 if (__get_cpu_var(last_guest) != lg || lg->last_pages != pages) {
83 __get_cpu_var(last_guest) = lg;
84 lg->last_pages = pages;
85 lg->changed = CHANGED_ALL;
86 }
87
88 /* These copies are pretty cheap, so we do them unconditionally: */
89 /* Save the current Host top-level page directory. */
90 pages->state.host_cr3 = __pa(current->mm->pgd);
91 /* Set up the Guest's page tables to see this CPU's pages (and no
92 * other CPU's pages). */
93 map_switcher_in_guest(lg, pages);
94 /* Set up the two "TSS" members which tell the CPU what stack to use
95 * for traps which do directly into the Guest (ie. traps at privilege
96 * level 1). */
97 pages->state.guest_tss.esp1 = lg->esp1;
98 pages->state.guest_tss.ss1 = lg->ss1;
99
100 /* Copy direct-to-Guest trap entries. */
101 if (lg->changed & CHANGED_IDT)
102 copy_traps(lg, pages->state.guest_idt, default_idt_entries);
103
104 /* Copy all GDT entries which the Guest can change. */
105 if (lg->changed & CHANGED_GDT)
106 copy_gdt(lg, pages->state.guest_gdt);
107 /* If only the TLS entries have changed, copy them. */
108 else if (lg->changed & CHANGED_GDT_TLS)
109 copy_gdt_tls(lg, pages->state.guest_gdt);
110
111 /* Mark the Guest as unchanged for next time. */
112 lg->changed = 0;
113}
114
115/* Finally: the code to actually call into the Switcher to run the Guest. */
116static void run_guest_once(struct lguest *lg, struct lguest_pages *pages)
117{
118 /* This is a dummy value we need for GCC's sake. */
119 unsigned int clobber;
120
121 /* Copy the guest-specific information into this CPU's "struct
122 * lguest_pages". */
123 copy_in_guest_info(lg, pages);
124
125 /* Set the trap number to 256 (impossible value). If we fault while
126 * switching to the Guest (bad segment registers or bug), this will
127 * cause us to abort the Guest. */
128 lg->regs->trapnum = 256;
129
130 /* Now: we push the "eflags" register on the stack, then do an "lcall".
131 * This is how we change from using the kernel code segment to using
132 * the dedicated lguest code segment, as well as jumping into the
133 * Switcher.
134 *
135 * The lcall also pushes the old code segment (KERNEL_CS) onto the
136 * stack, then the address of this call. This stack layout happens to
Rusty Russelle1e72962007-10-25 15:02:50 +1000137 * exactly match the stack layout created by an interrupt... */
Jes Sorensen625efab2007-10-22 11:03:28 +1000138 asm volatile("pushf; lcall *lguest_entry"
139 /* This is how we tell GCC that %eax ("a") and %ebx ("b")
140 * are changed by this routine. The "=" means output. */
141 : "=a"(clobber), "=b"(clobber)
142 /* %eax contains the pages pointer. ("0" refers to the
143 * 0-th argument above, ie "a"). %ebx contains the
144 * physical address of the Guest's top-level page
145 * directory. */
146 : "0"(pages), "1"(__pa(lg->pgdirs[lg->pgdidx].pgdir))
147 /* We tell gcc that all these registers could change,
148 * which means we don't have to save and restore them in
149 * the Switcher. */
150 : "memory", "%edx", "%ecx", "%edi", "%esi");
151}
152/*:*/
153
Rusty Russelle1e72962007-10-25 15:02:50 +1000154/*M:002 There are hooks in the scheduler which we can register to tell when we
155 * get kicked off the CPU (preempt_notifier_register()). This would allow us
156 * to lazily disable SYSENTER which would regain some performance, and should
157 * also simplify copy_in_guest_info(). Note that we'd still need to restore
158 * things when we exit to Launcher userspace, but that's fairly easy.
159 *
160 * The hooks were designed for KVM, but we can also put them to good use. :*/
161
Jes Sorensen625efab2007-10-22 11:03:28 +1000162/*H:040 This is the i386-specific code to setup and run the Guest. Interrupts
163 * are disabled: we own the CPU. */
164void lguest_arch_run_guest(struct lguest *lg)
165{
Rusty Russelle1e72962007-10-25 15:02:50 +1000166 /* Remember the awfully-named TS bit? If the Guest has asked to set it
167 * we set it now, so we can trap and pass that trap to the Guest if it
168 * uses the FPU. */
Jes Sorensen625efab2007-10-22 11:03:28 +1000169 if (lg->ts)
170 lguest_set_ts();
171
Rusty Russelle1e72962007-10-25 15:02:50 +1000172 /* SYSENTER is an optimized way of doing system calls. We can't allow
173 * it because it always jumps to privilege level 0. A normal Guest
174 * won't try it because we don't advertise it in CPUID, but a malicious
175 * Guest (or malicious Guest userspace program) could, so we tell the
176 * CPU to disable it before running the Guest. */
Jes Sorensen625efab2007-10-22 11:03:28 +1000177 if (boot_cpu_has(X86_FEATURE_SEP))
178 wrmsr(MSR_IA32_SYSENTER_CS, 0, 0);
179
Rusty Russelle1e72962007-10-25 15:02:50 +1000180 /* Now we actually run the Guest. It will return when something
181 * interesting happens, and we can examine its registers to see what it
182 * was doing. */
Jes Sorensen625efab2007-10-22 11:03:28 +1000183 run_guest_once(lg, lguest_pages(raw_smp_processor_id()));
184
Rusty Russelle1e72962007-10-25 15:02:50 +1000185 /* Note that the "regs" pointer contains two extra entries which are
186 * not really registers: a trap number which says what interrupt or
187 * trap made the switcher code come back, and an error code which some
188 * traps set. */
Jes Sorensen625efab2007-10-22 11:03:28 +1000189
Rusty Russelle1e72962007-10-25 15:02:50 +1000190 /* If the Guest page faulted, then the cr2 register will tell us the
191 * bad virtual address. We have to grab this now, because once we
192 * re-enable interrupts an interrupt could fault and thus overwrite
193 * cr2, or we could even move off to a different CPU. */
Jes Sorensen625efab2007-10-22 11:03:28 +1000194 if (lg->regs->trapnum == 14)
195 lg->arch.last_pagefault = read_cr2();
196 /* Similarly, if we took a trap because the Guest used the FPU,
197 * we have to restore the FPU it expects to see. */
198 else if (lg->regs->trapnum == 7)
199 math_state_restore();
200
201 /* Restore SYSENTER if it's supposed to be on. */
202 if (boot_cpu_has(X86_FEATURE_SEP))
203 wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
204}
205
Rusty Russelle1e72962007-10-25 15:02:50 +1000206/*H:130 Now we've examined the hypercall code; our Guest can make requests.
207 * Our Guest is usually so well behaved; it never tries to do things it isn't
208 * allowed to, and uses hypercalls instead. Unfortunately, Linux's paravirtual
209 * infrastructure isn't quite complete, because it doesn't contain replacements
210 * for the Intel I/O instructions. As a result, the Guest sometimes fumbles
211 * across one during the boot process as it probes for various things which are
212 * usually attached to a PC.
Jes Sorensen625efab2007-10-22 11:03:28 +1000213 *
Rusty Russelle1e72962007-10-25 15:02:50 +1000214 * When the Guest uses one of these instructions, we get a trap (General
Jes Sorensen625efab2007-10-22 11:03:28 +1000215 * Protection Fault) and come here. We see if it's one of those troublesome
216 * instructions and skip over it. We return true if we did. */
217static int emulate_insn(struct lguest *lg)
218{
219 u8 insn;
220 unsigned int insnlen = 0, in = 0, shift = 0;
221 /* The eip contains the *virtual* address of the Guest's instruction:
222 * guest_pa just subtracts the Guest's page_offset. */
223 unsigned long physaddr = guest_pa(lg, lg->regs->eip);
224
Rusty Russell47436aa2007-10-22 11:03:36 +1000225 /* This must be the Guest kernel trying to do something, not userspace!
226 * The bottom two bits of the CS segment register are the privilege
227 * level. */
228 if ((lg->regs->cs & 3) != GUEST_PL)
Jes Sorensen625efab2007-10-22 11:03:28 +1000229 return 0;
230
231 /* Decoding x86 instructions is icky. */
Rusty Russell2d37f942007-10-22 11:24:24 +1000232 insn = lgread(lg, physaddr, u8);
Jes Sorensen625efab2007-10-22 11:03:28 +1000233
234 /* 0x66 is an "operand prefix". It means it's using the upper 16 bits
235 of the eax register. */
236 if (insn == 0x66) {
237 shift = 16;
238 /* The instruction is 1 byte so far, read the next byte. */
239 insnlen = 1;
Rusty Russell2d37f942007-10-22 11:24:24 +1000240 insn = lgread(lg, physaddr + insnlen, u8);
Jes Sorensen625efab2007-10-22 11:03:28 +1000241 }
242
243 /* We can ignore the lower bit for the moment and decode the 4 opcodes
244 * we need to emulate. */
245 switch (insn & 0xFE) {
246 case 0xE4: /* in <next byte>,%al */
247 insnlen += 2;
248 in = 1;
249 break;
250 case 0xEC: /* in (%dx),%al */
251 insnlen += 1;
252 in = 1;
253 break;
254 case 0xE6: /* out %al,<next byte> */
255 insnlen += 2;
256 break;
257 case 0xEE: /* out %al,(%dx) */
258 insnlen += 1;
259 break;
260 default:
261 /* OK, we don't know what this is, can't emulate. */
262 return 0;
263 }
264
265 /* If it was an "IN" instruction, they expect the result to be read
266 * into %eax, so we change %eax. We always return all-ones, which
267 * traditionally means "there's nothing there". */
268 if (in) {
269 /* Lower bit tells is whether it's a 16 or 32 bit access */
270 if (insn & 0x1)
271 lg->regs->eax = 0xFFFFFFFF;
272 else
273 lg->regs->eax |= (0xFFFF << shift);
274 }
275 /* Finally, we've "done" the instruction, so move past it. */
276 lg->regs->eip += insnlen;
277 /* Success! */
278 return 1;
279}
280
281/*H:050 Once we've re-enabled interrupts, we look at why the Guest exited. */
282void lguest_arch_handle_trap(struct lguest *lg)
283{
284 switch (lg->regs->trapnum) {
Rusty Russelle1e72962007-10-25 15:02:50 +1000285 case 13: /* We've intercepted a General Protection Fault. */
286 /* Check if this was one of those annoying IN or OUT
287 * instructions which we need to emulate. If so, we just go
288 * back into the Guest after we've done it. */
Jes Sorensen625efab2007-10-22 11:03:28 +1000289 if (lg->regs->errcode == 0) {
290 if (emulate_insn(lg))
291 return;
292 }
293 break;
Rusty Russelle1e72962007-10-25 15:02:50 +1000294 case 14: /* We've intercepted a Page Fault. */
295 /* The Guest accessed a virtual address that wasn't mapped.
296 * This happens a lot: we don't actually set up most of the
297 * page tables for the Guest at all when we start: as it runs
298 * it asks for more and more, and we set them up as
299 * required. In this case, we don't even tell the Guest that
300 * the fault happened.
301 *
302 * The errcode tells whether this was a read or a write, and
303 * whether kernel or userspace code. */
Jes Sorensen625efab2007-10-22 11:03:28 +1000304 if (demand_page(lg, lg->arch.last_pagefault, lg->regs->errcode))
305 return;
306
Rusty Russelle1e72962007-10-25 15:02:50 +1000307 /* OK, it's really not there (or not OK): the Guest needs to
308 * know. We write out the cr2 value so it knows where the
309 * fault occurred.
310 *
311 * Note that if the Guest were really messed up, this could
312 * happen before it's done the LHCALL_LGUEST_INIT hypercall, so
313 * lg->lguest_data could be NULL */
Jes Sorensen625efab2007-10-22 11:03:28 +1000314 if (lg->lguest_data &&
315 put_user(lg->arch.last_pagefault, &lg->lguest_data->cr2))
316 kill_guest(lg, "Writing cr2");
317 break;
318 case 7: /* We've intercepted a Device Not Available fault. */
Rusty Russelle1e72962007-10-25 15:02:50 +1000319 /* If the Guest doesn't want to know, we already restored the
320 * Floating Point Unit, so we just continue without telling
321 * it. */
Jes Sorensen625efab2007-10-22 11:03:28 +1000322 if (!lg->ts)
323 return;
324 break;
325 case 32 ... 255:
Rusty Russellcc6d4fb2007-10-22 11:03:30 +1000326 /* These values mean a real interrupt occurred, in which case
327 * the Host handler has already been run. We just do a
328 * friendly check if another process should now be run, then
329 * return to run the Guest again */
Jes Sorensen625efab2007-10-22 11:03:28 +1000330 cond_resched();
Rusty Russellcc6d4fb2007-10-22 11:03:30 +1000331 return;
332 case LGUEST_TRAP_ENTRY:
Jes Sorensenb410e7b2007-10-22 11:03:31 +1000333 /* Our 'struct hcall_args' maps directly over our regs: we set
334 * up the pointer now to indicate a hypercall is pending. */
335 lg->hcall = (struct hcall_args *)lg->regs;
Jes Sorensen625efab2007-10-22 11:03:28 +1000336 return;
337 }
338
339 /* We didn't handle the trap, so it needs to go to the Guest. */
340 if (!deliver_trap(lg, lg->regs->trapnum))
341 /* If the Guest doesn't have a handler (either it hasn't
342 * registered any yet, or it's one of the faults we don't let
343 * it handle), it dies with a cryptic error message. */
344 kill_guest(lg, "unhandled trap %li at %#lx (%#lx)",
345 lg->regs->trapnum, lg->regs->eip,
346 lg->regs->trapnum == 14 ? lg->arch.last_pagefault
347 : lg->regs->errcode);
348}
349
350/* Now we can look at each of the routines this calls, in increasing order of
351 * complexity: do_hypercalls(), emulate_insn(), maybe_do_interrupt(),
352 * deliver_trap() and demand_page(). After all those, we'll be ready to
353 * examine the Switcher, and our philosophical understanding of the Host/Guest
354 * duality will be complete. :*/
355static void adjust_pge(void *on)
356{
357 if (on)
358 write_cr4(read_cr4() | X86_CR4_PGE);
359 else
360 write_cr4(read_cr4() & ~X86_CR4_PGE);
361}
362
363/*H:020 Now the Switcher is mapped and every thing else is ready, we need to do
364 * some more i386-specific initialization. */
365void __init lguest_arch_host_init(void)
366{
367 int i;
368
369 /* Most of the i386/switcher.S doesn't care that it's been moved; on
370 * Intel, jumps are relative, and it doesn't access any references to
371 * external code or data.
372 *
373 * The only exception is the interrupt handlers in switcher.S: their
374 * addresses are placed in a table (default_idt_entries), so we need to
375 * update the table with the new addresses. switcher_offset() is a
376 * convenience function which returns the distance between the builtin
377 * switcher code and the high-mapped copy we just made. */
378 for (i = 0; i < IDT_ENTRIES; i++)
379 default_idt_entries[i] += switcher_offset();
380
381 /*
382 * Set up the Switcher's per-cpu areas.
383 *
384 * Each CPU gets two pages of its own within the high-mapped region
385 * (aka. "struct lguest_pages"). Much of this can be initialized now,
386 * but some depends on what Guest we are running (which is set up in
387 * copy_in_guest_info()).
388 */
389 for_each_possible_cpu(i) {
390 /* lguest_pages() returns this CPU's two pages. */
391 struct lguest_pages *pages = lguest_pages(i);
392 /* This is a convenience pointer to make the code fit one
393 * statement to a line. */
394 struct lguest_ro_state *state = &pages->state;
395
396 /* The Global Descriptor Table: the Host has a different one
397 * for each CPU. We keep a descriptor for the GDT which says
398 * where it is and how big it is (the size is actually the last
399 * byte, not the size, hence the "-1"). */
400 state->host_gdt_desc.size = GDT_SIZE-1;
401 state->host_gdt_desc.address = (long)get_cpu_gdt_table(i);
402
403 /* All CPUs on the Host use the same Interrupt Descriptor
404 * Table, so we just use store_idt(), which gets this CPU's IDT
405 * descriptor. */
406 store_idt(&state->host_idt_desc);
407
408 /* The descriptors for the Guest's GDT and IDT can be filled
409 * out now, too. We copy the GDT & IDT into ->guest_gdt and
410 * ->guest_idt before actually running the Guest. */
411 state->guest_idt_desc.size = sizeof(state->guest_idt)-1;
412 state->guest_idt_desc.address = (long)&state->guest_idt;
413 state->guest_gdt_desc.size = sizeof(state->guest_gdt)-1;
414 state->guest_gdt_desc.address = (long)&state->guest_gdt;
415
416 /* We know where we want the stack to be when the Guest enters
417 * the switcher: in pages->regs. The stack grows upwards, so
418 * we start it at the end of that structure. */
419 state->guest_tss.esp0 = (long)(&pages->regs + 1);
420 /* And this is the GDT entry to use for the stack: we keep a
421 * couple of special LGUEST entries. */
422 state->guest_tss.ss0 = LGUEST_DS;
423
424 /* x86 can have a finegrained bitmap which indicates what I/O
425 * ports the process can use. We set it to the end of our
426 * structure, meaning "none". */
427 state->guest_tss.io_bitmap_base = sizeof(state->guest_tss);
428
429 /* Some GDT entries are the same across all Guests, so we can
430 * set them up now. */
431 setup_default_gdt_entries(state);
432 /* Most IDT entries are the same for all Guests, too.*/
433 setup_default_idt_entries(state, default_idt_entries);
434
435 /* The Host needs to be able to use the LGUEST segments on this
436 * CPU, too, so put them in the Host GDT. */
437 get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
438 get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
439 }
440
441 /* In the Switcher, we want the %cs segment register to use the
442 * LGUEST_CS GDT entry: we've put that in the Host and Guest GDTs, so
443 * it will be undisturbed when we switch. To change %cs and jump we
444 * need this structure to feed to Intel's "lcall" instruction. */
445 lguest_entry.offset = (long)switch_to_guest + switcher_offset();
446 lguest_entry.segment = LGUEST_CS;
447
448 /* Finally, we need to turn off "Page Global Enable". PGE is an
449 * optimization where page table entries are specially marked to show
450 * they never change. The Host kernel marks all the kernel pages this
451 * way because it's always present, even when userspace is running.
452 *
453 * Lguest breaks this: unbeknownst to the rest of the Host kernel, we
454 * switch to the Guest kernel. If you don't disable this on all CPUs,
455 * you'll get really weird bugs that you'll chase for two days.
456 *
457 * I used to turn PGE off every time we switched to the Guest and back
458 * on when we return, but that slowed the Switcher down noticibly. */
459
460 /* We don't need the complexity of CPUs coming and going while we're
461 * doing this. */
462 lock_cpu_hotplug();
463 if (cpu_has_pge) { /* We have a broader idea of "global". */
464 /* Remember that this was originally set (for cleanup). */
465 cpu_had_pge = 1;
466 /* adjust_pge is a helper function which sets or unsets the PGE
467 * bit on its CPU, depending on the argument (0 == unset). */
468 on_each_cpu(adjust_pge, (void *)0, 0, 1);
469 /* Turn off the feature in the global feature set. */
470 clear_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability);
471 }
472 unlock_cpu_hotplug();
473};
474/*:*/
475
476void __exit lguest_arch_host_fini(void)
477{
478 /* If we had PGE before we started, turn it back on now. */
479 lock_cpu_hotplug();
480 if (cpu_had_pge) {
481 set_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability);
482 /* adjust_pge's argument "1" means set PGE. */
483 on_each_cpu(adjust_pge, (void *)1, 0, 1);
484 }
485 unlock_cpu_hotplug();
486}
Jes Sorensenb410e7b2007-10-22 11:03:31 +1000487
488
489/*H:122 The i386-specific hypercalls simply farm out to the right functions. */
490int lguest_arch_do_hcall(struct lguest *lg, struct hcall_args *args)
491{
492 switch (args->arg0) {
493 case LHCALL_LOAD_GDT:
494 load_guest_gdt(lg, args->arg1, args->arg2);
495 break;
496 case LHCALL_LOAD_IDT_ENTRY:
497 load_guest_idt_entry(lg, args->arg1, args->arg2, args->arg3);
498 break;
499 case LHCALL_LOAD_TLS:
500 guest_load_tls(lg, args->arg1);
501 break;
502 default:
503 /* Bad Guest. Bad! */
504 return -EIO;
505 }
506 return 0;
507}
508
509/*H:126 i386-specific hypercall initialization: */
510int lguest_arch_init_hypercalls(struct lguest *lg)
511{
512 u32 tsc_speed;
513
514 /* The pointer to the Guest's "struct lguest_data" is the only
515 * argument. We check that address now. */
516 if (!lguest_address_ok(lg, lg->hcall->arg1, sizeof(*lg->lguest_data)))
517 return -EFAULT;
518
519 /* Having checked it, we simply set lg->lguest_data to point straight
520 * into the Launcher's memory at the right place and then use
521 * copy_to_user/from_user from now on, instead of lgread/write. I put
522 * this in to show that I'm not immune to writing stupid
523 * optimizations. */
524 lg->lguest_data = lg->mem_base + lg->hcall->arg1;
525
526 /* We insist that the Time Stamp Counter exist and doesn't change with
527 * cpu frequency. Some devious chip manufacturers decided that TSC
528 * changes could be handled in software. I decided that time going
529 * backwards might be good for benchmarks, but it's bad for users.
530 *
531 * We also insist that the TSC be stable: the kernel detects unreliable
532 * TSCs for its own purposes, and we use that here. */
533 if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC) && !check_tsc_unstable())
534 tsc_speed = tsc_khz;
535 else
536 tsc_speed = 0;
537 if (put_user(tsc_speed, &lg->lguest_data->tsc_khz))
538 return -EFAULT;
539
Rusty Russellc18acd72007-10-22 11:03:35 +1000540 /* The interrupt code might not like the system call vector. */
541 if (!check_syscall_vector(lg))
542 kill_guest(lg, "bad syscall vector");
543
Jes Sorensenb410e7b2007-10-22 11:03:31 +1000544 return 0;
545}
Jes Sorensend612cde2007-10-22 11:03:32 +1000546
547/*L:030 lguest_arch_setup_regs()
548 *
549 * Most of the Guest's registers are left alone: we used get_zeroed_page() to
550 * allocate the structure, so they will be 0. */
551void lguest_arch_setup_regs(struct lguest *lg, unsigned long start)
552{
553 struct lguest_regs *regs = lg->regs;
554
555 /* There are four "segment" registers which the Guest needs to boot:
556 * The "code segment" register (cs) refers to the kernel code segment
557 * __KERNEL_CS, and the "data", "extra" and "stack" segment registers
558 * refer to the kernel data segment __KERNEL_DS.
559 *
560 * The privilege level is packed into the lower bits. The Guest runs
561 * at privilege level 1 (GUEST_PL).*/
562 regs->ds = regs->es = regs->ss = __KERNEL_DS|GUEST_PL;
563 regs->cs = __KERNEL_CS|GUEST_PL;
564
565 /* The "eflags" register contains miscellaneous flags. Bit 1 (0x002)
566 * is supposed to always be "1". Bit 9 (0x200) controls whether
567 * interrupts are enabled. We always leave interrupts enabled while
568 * running the Guest. */
Rusty Russell25c47bb2007-10-25 14:09:53 +1000569 regs->eflags = X86_EFLAGS_IF | 0x2;
Jes Sorensend612cde2007-10-22 11:03:32 +1000570
571 /* The "Extended Instruction Pointer" register says where the Guest is
572 * running. */
573 regs->eip = start;
574
575 /* %esi points to our boot information, at physical address 0, so don't
576 * touch it. */
Rusty Russelle1e72962007-10-25 15:02:50 +1000577
Jes Sorensend612cde2007-10-22 11:03:32 +1000578 /* There are a couple of GDT entries the Guest expects when first
579 * booting. */
Jes Sorensend612cde2007-10-22 11:03:32 +1000580 setup_guest_gdt(lg);
581}