Linux 2.6.39-rc2
[pohmelfs.git] / arch / powerpc / kernel / kvm.c
blobb06bdae04064f59d2c4c4c98e76304dee6ca98c8
1 /*
2 * Copyright (C) 2010 SUSE Linux Products GmbH. All rights reserved.
4 * Authors:
5 * Alexander Graf <agraf@suse.de>
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License, version 2, as
9 * published by the Free Software Foundation.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
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, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
21 #include <linux/kvm_host.h>
22 #include <linux/init.h>
23 #include <linux/kvm_para.h>
24 #include <linux/slab.h>
25 #include <linux/of.h>
27 #include <asm/reg.h>
28 #include <asm/sections.h>
29 #include <asm/cacheflush.h>
30 #include <asm/disassemble.h>
32 #define KVM_MAGIC_PAGE (-4096L)
33 #define magic_var(x) KVM_MAGIC_PAGE + offsetof(struct kvm_vcpu_arch_shared, x)
35 #define KVM_INST_LWZ 0x80000000
36 #define KVM_INST_STW 0x90000000
37 #define KVM_INST_LD 0xe8000000
38 #define KVM_INST_STD 0xf8000000
39 #define KVM_INST_NOP 0x60000000
40 #define KVM_INST_B 0x48000000
41 #define KVM_INST_B_MASK 0x03ffffff
42 #define KVM_INST_B_MAX 0x01ffffff
44 #define KVM_MASK_RT 0x03e00000
45 #define KVM_RT_30 0x03c00000
46 #define KVM_MASK_RB 0x0000f800
47 #define KVM_INST_MFMSR 0x7c0000a6
48 #define KVM_INST_MFSPR_SPRG0 0x7c1042a6
49 #define KVM_INST_MFSPR_SPRG1 0x7c1142a6
50 #define KVM_INST_MFSPR_SPRG2 0x7c1242a6
51 #define KVM_INST_MFSPR_SPRG3 0x7c1342a6
52 #define KVM_INST_MFSPR_SRR0 0x7c1a02a6
53 #define KVM_INST_MFSPR_SRR1 0x7c1b02a6
54 #define KVM_INST_MFSPR_DAR 0x7c1302a6
55 #define KVM_INST_MFSPR_DSISR 0x7c1202a6
57 #define KVM_INST_MTSPR_SPRG0 0x7c1043a6
58 #define KVM_INST_MTSPR_SPRG1 0x7c1143a6
59 #define KVM_INST_MTSPR_SPRG2 0x7c1243a6
60 #define KVM_INST_MTSPR_SPRG3 0x7c1343a6
61 #define KVM_INST_MTSPR_SRR0 0x7c1a03a6
62 #define KVM_INST_MTSPR_SRR1 0x7c1b03a6
63 #define KVM_INST_MTSPR_DAR 0x7c1303a6
64 #define KVM_INST_MTSPR_DSISR 0x7c1203a6
66 #define KVM_INST_TLBSYNC 0x7c00046c
67 #define KVM_INST_MTMSRD_L0 0x7c000164
68 #define KVM_INST_MTMSRD_L1 0x7c010164
69 #define KVM_INST_MTMSR 0x7c000124
71 #define KVM_INST_WRTEEI_0 0x7c000146
72 #define KVM_INST_WRTEEI_1 0x7c008146
74 #define KVM_INST_MTSRIN 0x7c0001e4
76 static bool kvm_patching_worked = true;
77 static char kvm_tmp[1024 * 1024];
78 static int kvm_tmp_index;
80 static inline void kvm_patch_ins(u32 *inst, u32 new_inst)
82 *inst = new_inst;
83 flush_icache_range((ulong)inst, (ulong)inst + 4);
86 static void kvm_patch_ins_ll(u32 *inst, long addr, u32 rt)
88 #ifdef CONFIG_64BIT
89 kvm_patch_ins(inst, KVM_INST_LD | rt | (addr & 0x0000fffc));
90 #else
91 kvm_patch_ins(inst, KVM_INST_LWZ | rt | (addr & 0x0000fffc));
92 #endif
95 static void kvm_patch_ins_ld(u32 *inst, long addr, u32 rt)
97 #ifdef CONFIG_64BIT
98 kvm_patch_ins(inst, KVM_INST_LD | rt | (addr & 0x0000fffc));
99 #else
100 kvm_patch_ins(inst, KVM_INST_LWZ | rt | ((addr + 4) & 0x0000fffc));
101 #endif
104 static void kvm_patch_ins_lwz(u32 *inst, long addr, u32 rt)
106 kvm_patch_ins(inst, KVM_INST_LWZ | rt | (addr & 0x0000ffff));
109 static void kvm_patch_ins_std(u32 *inst, long addr, u32 rt)
111 #ifdef CONFIG_64BIT
112 kvm_patch_ins(inst, KVM_INST_STD | rt | (addr & 0x0000fffc));
113 #else
114 kvm_patch_ins(inst, KVM_INST_STW | rt | ((addr + 4) & 0x0000fffc));
115 #endif
118 static void kvm_patch_ins_stw(u32 *inst, long addr, u32 rt)
120 kvm_patch_ins(inst, KVM_INST_STW | rt | (addr & 0x0000fffc));
123 static void kvm_patch_ins_nop(u32 *inst)
125 kvm_patch_ins(inst, KVM_INST_NOP);
128 static void kvm_patch_ins_b(u32 *inst, int addr)
130 #if defined(CONFIG_RELOCATABLE) && defined(CONFIG_PPC_BOOK3S)
131 /* On relocatable kernels interrupts handlers and our code
132 can be in different regions, so we don't patch them */
134 extern u32 __end_interrupts;
135 if ((ulong)inst < (ulong)&__end_interrupts)
136 return;
137 #endif
139 kvm_patch_ins(inst, KVM_INST_B | (addr & KVM_INST_B_MASK));
142 static u32 *kvm_alloc(int len)
144 u32 *p;
146 if ((kvm_tmp_index + len) > ARRAY_SIZE(kvm_tmp)) {
147 printk(KERN_ERR "KVM: No more space (%d + %d)\n",
148 kvm_tmp_index, len);
149 kvm_patching_worked = false;
150 return NULL;
153 p = (void*)&kvm_tmp[kvm_tmp_index];
154 kvm_tmp_index += len;
156 return p;
159 extern u32 kvm_emulate_mtmsrd_branch_offs;
160 extern u32 kvm_emulate_mtmsrd_reg_offs;
161 extern u32 kvm_emulate_mtmsrd_orig_ins_offs;
162 extern u32 kvm_emulate_mtmsrd_len;
163 extern u32 kvm_emulate_mtmsrd[];
165 static void kvm_patch_ins_mtmsrd(u32 *inst, u32 rt)
167 u32 *p;
168 int distance_start;
169 int distance_end;
170 ulong next_inst;
172 p = kvm_alloc(kvm_emulate_mtmsrd_len * 4);
173 if (!p)
174 return;
176 /* Find out where we are and put everything there */
177 distance_start = (ulong)p - (ulong)inst;
178 next_inst = ((ulong)inst + 4);
179 distance_end = next_inst - (ulong)&p[kvm_emulate_mtmsrd_branch_offs];
181 /* Make sure we only write valid b instructions */
182 if (distance_start > KVM_INST_B_MAX) {
183 kvm_patching_worked = false;
184 return;
187 /* Modify the chunk to fit the invocation */
188 memcpy(p, kvm_emulate_mtmsrd, kvm_emulate_mtmsrd_len * 4);
189 p[kvm_emulate_mtmsrd_branch_offs] |= distance_end & KVM_INST_B_MASK;
190 switch (get_rt(rt)) {
191 case 30:
192 kvm_patch_ins_ll(&p[kvm_emulate_mtmsrd_reg_offs],
193 magic_var(scratch2), KVM_RT_30);
194 break;
195 case 31:
196 kvm_patch_ins_ll(&p[kvm_emulate_mtmsrd_reg_offs],
197 magic_var(scratch1), KVM_RT_30);
198 break;
199 default:
200 p[kvm_emulate_mtmsrd_reg_offs] |= rt;
201 break;
204 p[kvm_emulate_mtmsrd_orig_ins_offs] = *inst;
205 flush_icache_range((ulong)p, (ulong)p + kvm_emulate_mtmsrd_len * 4);
207 /* Patch the invocation */
208 kvm_patch_ins_b(inst, distance_start);
211 extern u32 kvm_emulate_mtmsr_branch_offs;
212 extern u32 kvm_emulate_mtmsr_reg1_offs;
213 extern u32 kvm_emulate_mtmsr_reg2_offs;
214 extern u32 kvm_emulate_mtmsr_orig_ins_offs;
215 extern u32 kvm_emulate_mtmsr_len;
216 extern u32 kvm_emulate_mtmsr[];
218 static void kvm_patch_ins_mtmsr(u32 *inst, u32 rt)
220 u32 *p;
221 int distance_start;
222 int distance_end;
223 ulong next_inst;
225 p = kvm_alloc(kvm_emulate_mtmsr_len * 4);
226 if (!p)
227 return;
229 /* Find out where we are and put everything there */
230 distance_start = (ulong)p - (ulong)inst;
231 next_inst = ((ulong)inst + 4);
232 distance_end = next_inst - (ulong)&p[kvm_emulate_mtmsr_branch_offs];
234 /* Make sure we only write valid b instructions */
235 if (distance_start > KVM_INST_B_MAX) {
236 kvm_patching_worked = false;
237 return;
240 /* Modify the chunk to fit the invocation */
241 memcpy(p, kvm_emulate_mtmsr, kvm_emulate_mtmsr_len * 4);
242 p[kvm_emulate_mtmsr_branch_offs] |= distance_end & KVM_INST_B_MASK;
244 /* Make clobbered registers work too */
245 switch (get_rt(rt)) {
246 case 30:
247 kvm_patch_ins_ll(&p[kvm_emulate_mtmsr_reg1_offs],
248 magic_var(scratch2), KVM_RT_30);
249 kvm_patch_ins_ll(&p[kvm_emulate_mtmsr_reg2_offs],
250 magic_var(scratch2), KVM_RT_30);
251 break;
252 case 31:
253 kvm_patch_ins_ll(&p[kvm_emulate_mtmsr_reg1_offs],
254 magic_var(scratch1), KVM_RT_30);
255 kvm_patch_ins_ll(&p[kvm_emulate_mtmsr_reg2_offs],
256 magic_var(scratch1), KVM_RT_30);
257 break;
258 default:
259 p[kvm_emulate_mtmsr_reg1_offs] |= rt;
260 p[kvm_emulate_mtmsr_reg2_offs] |= rt;
261 break;
264 p[kvm_emulate_mtmsr_orig_ins_offs] = *inst;
265 flush_icache_range((ulong)p, (ulong)p + kvm_emulate_mtmsr_len * 4);
267 /* Patch the invocation */
268 kvm_patch_ins_b(inst, distance_start);
271 #ifdef CONFIG_BOOKE
273 extern u32 kvm_emulate_wrteei_branch_offs;
274 extern u32 kvm_emulate_wrteei_ee_offs;
275 extern u32 kvm_emulate_wrteei_len;
276 extern u32 kvm_emulate_wrteei[];
278 static void kvm_patch_ins_wrteei(u32 *inst)
280 u32 *p;
281 int distance_start;
282 int distance_end;
283 ulong next_inst;
285 p = kvm_alloc(kvm_emulate_wrteei_len * 4);
286 if (!p)
287 return;
289 /* Find out where we are and put everything there */
290 distance_start = (ulong)p - (ulong)inst;
291 next_inst = ((ulong)inst + 4);
292 distance_end = next_inst - (ulong)&p[kvm_emulate_wrteei_branch_offs];
294 /* Make sure we only write valid b instructions */
295 if (distance_start > KVM_INST_B_MAX) {
296 kvm_patching_worked = false;
297 return;
300 /* Modify the chunk to fit the invocation */
301 memcpy(p, kvm_emulate_wrteei, kvm_emulate_wrteei_len * 4);
302 p[kvm_emulate_wrteei_branch_offs] |= distance_end & KVM_INST_B_MASK;
303 p[kvm_emulate_wrteei_ee_offs] |= (*inst & MSR_EE);
304 flush_icache_range((ulong)p, (ulong)p + kvm_emulate_wrteei_len * 4);
306 /* Patch the invocation */
307 kvm_patch_ins_b(inst, distance_start);
310 #endif
312 #ifdef CONFIG_PPC_BOOK3S_32
314 extern u32 kvm_emulate_mtsrin_branch_offs;
315 extern u32 kvm_emulate_mtsrin_reg1_offs;
316 extern u32 kvm_emulate_mtsrin_reg2_offs;
317 extern u32 kvm_emulate_mtsrin_orig_ins_offs;
318 extern u32 kvm_emulate_mtsrin_len;
319 extern u32 kvm_emulate_mtsrin[];
321 static void kvm_patch_ins_mtsrin(u32 *inst, u32 rt, u32 rb)
323 u32 *p;
324 int distance_start;
325 int distance_end;
326 ulong next_inst;
328 p = kvm_alloc(kvm_emulate_mtsrin_len * 4);
329 if (!p)
330 return;
332 /* Find out where we are and put everything there */
333 distance_start = (ulong)p - (ulong)inst;
334 next_inst = ((ulong)inst + 4);
335 distance_end = next_inst - (ulong)&p[kvm_emulate_mtsrin_branch_offs];
337 /* Make sure we only write valid b instructions */
338 if (distance_start > KVM_INST_B_MAX) {
339 kvm_patching_worked = false;
340 return;
343 /* Modify the chunk to fit the invocation */
344 memcpy(p, kvm_emulate_mtsrin, kvm_emulate_mtsrin_len * 4);
345 p[kvm_emulate_mtsrin_branch_offs] |= distance_end & KVM_INST_B_MASK;
346 p[kvm_emulate_mtsrin_reg1_offs] |= (rb << 10);
347 p[kvm_emulate_mtsrin_reg2_offs] |= rt;
348 p[kvm_emulate_mtsrin_orig_ins_offs] = *inst;
349 flush_icache_range((ulong)p, (ulong)p + kvm_emulate_mtsrin_len * 4);
351 /* Patch the invocation */
352 kvm_patch_ins_b(inst, distance_start);
355 #endif
357 static void kvm_map_magic_page(void *data)
359 u32 *features = data;
361 ulong in[8];
362 ulong out[8];
364 in[0] = KVM_MAGIC_PAGE;
365 in[1] = KVM_MAGIC_PAGE;
367 kvm_hypercall(in, out, HC_VENDOR_KVM | KVM_HC_PPC_MAP_MAGIC_PAGE);
369 *features = out[0];
372 static void kvm_check_ins(u32 *inst, u32 features)
374 u32 _inst = *inst;
375 u32 inst_no_rt = _inst & ~KVM_MASK_RT;
376 u32 inst_rt = _inst & KVM_MASK_RT;
378 switch (inst_no_rt) {
379 /* Loads */
380 case KVM_INST_MFMSR:
381 kvm_patch_ins_ld(inst, magic_var(msr), inst_rt);
382 break;
383 case KVM_INST_MFSPR_SPRG0:
384 kvm_patch_ins_ld(inst, magic_var(sprg0), inst_rt);
385 break;
386 case KVM_INST_MFSPR_SPRG1:
387 kvm_patch_ins_ld(inst, magic_var(sprg1), inst_rt);
388 break;
389 case KVM_INST_MFSPR_SPRG2:
390 kvm_patch_ins_ld(inst, magic_var(sprg2), inst_rt);
391 break;
392 case KVM_INST_MFSPR_SPRG3:
393 kvm_patch_ins_ld(inst, magic_var(sprg3), inst_rt);
394 break;
395 case KVM_INST_MFSPR_SRR0:
396 kvm_patch_ins_ld(inst, magic_var(srr0), inst_rt);
397 break;
398 case KVM_INST_MFSPR_SRR1:
399 kvm_patch_ins_ld(inst, magic_var(srr1), inst_rt);
400 break;
401 case KVM_INST_MFSPR_DAR:
402 kvm_patch_ins_ld(inst, magic_var(dar), inst_rt);
403 break;
404 case KVM_INST_MFSPR_DSISR:
405 kvm_patch_ins_lwz(inst, magic_var(dsisr), inst_rt);
406 break;
408 /* Stores */
409 case KVM_INST_MTSPR_SPRG0:
410 kvm_patch_ins_std(inst, magic_var(sprg0), inst_rt);
411 break;
412 case KVM_INST_MTSPR_SPRG1:
413 kvm_patch_ins_std(inst, magic_var(sprg1), inst_rt);
414 break;
415 case KVM_INST_MTSPR_SPRG2:
416 kvm_patch_ins_std(inst, magic_var(sprg2), inst_rt);
417 break;
418 case KVM_INST_MTSPR_SPRG3:
419 kvm_patch_ins_std(inst, magic_var(sprg3), inst_rt);
420 break;
421 case KVM_INST_MTSPR_SRR0:
422 kvm_patch_ins_std(inst, magic_var(srr0), inst_rt);
423 break;
424 case KVM_INST_MTSPR_SRR1:
425 kvm_patch_ins_std(inst, magic_var(srr1), inst_rt);
426 break;
427 case KVM_INST_MTSPR_DAR:
428 kvm_patch_ins_std(inst, magic_var(dar), inst_rt);
429 break;
430 case KVM_INST_MTSPR_DSISR:
431 kvm_patch_ins_stw(inst, magic_var(dsisr), inst_rt);
432 break;
434 /* Nops */
435 case KVM_INST_TLBSYNC:
436 kvm_patch_ins_nop(inst);
437 break;
439 /* Rewrites */
440 case KVM_INST_MTMSRD_L1:
441 kvm_patch_ins_mtmsrd(inst, inst_rt);
442 break;
443 case KVM_INST_MTMSR:
444 case KVM_INST_MTMSRD_L0:
445 kvm_patch_ins_mtmsr(inst, inst_rt);
446 break;
449 switch (inst_no_rt & ~KVM_MASK_RB) {
450 #ifdef CONFIG_PPC_BOOK3S_32
451 case KVM_INST_MTSRIN:
452 if (features & KVM_MAGIC_FEAT_SR) {
453 u32 inst_rb = _inst & KVM_MASK_RB;
454 kvm_patch_ins_mtsrin(inst, inst_rt, inst_rb);
456 break;
457 break;
458 #endif
461 switch (_inst) {
462 #ifdef CONFIG_BOOKE
463 case KVM_INST_WRTEEI_0:
464 case KVM_INST_WRTEEI_1:
465 kvm_patch_ins_wrteei(inst);
466 break;
467 #endif
471 static void kvm_use_magic_page(void)
473 u32 *p;
474 u32 *start, *end;
475 u32 tmp;
476 u32 features;
478 /* Tell the host to map the magic page to -4096 on all CPUs */
479 on_each_cpu(kvm_map_magic_page, &features, 1);
481 /* Quick self-test to see if the mapping works */
482 if (__get_user(tmp, (u32*)KVM_MAGIC_PAGE)) {
483 kvm_patching_worked = false;
484 return;
487 /* Now loop through all code and find instructions */
488 start = (void*)_stext;
489 end = (void*)_etext;
491 for (p = start; p < end; p++)
492 kvm_check_ins(p, features);
494 printk(KERN_INFO "KVM: Live patching for a fast VM %s\n",
495 kvm_patching_worked ? "worked" : "failed");
498 unsigned long kvm_hypercall(unsigned long *in,
499 unsigned long *out,
500 unsigned long nr)
502 unsigned long register r0 asm("r0");
503 unsigned long register r3 asm("r3") = in[0];
504 unsigned long register r4 asm("r4") = in[1];
505 unsigned long register r5 asm("r5") = in[2];
506 unsigned long register r6 asm("r6") = in[3];
507 unsigned long register r7 asm("r7") = in[4];
508 unsigned long register r8 asm("r8") = in[5];
509 unsigned long register r9 asm("r9") = in[6];
510 unsigned long register r10 asm("r10") = in[7];
511 unsigned long register r11 asm("r11") = nr;
512 unsigned long register r12 asm("r12");
514 asm volatile("bl kvm_hypercall_start"
515 : "=r"(r0), "=r"(r3), "=r"(r4), "=r"(r5), "=r"(r6),
516 "=r"(r7), "=r"(r8), "=r"(r9), "=r"(r10), "=r"(r11),
517 "=r"(r12)
518 : "r"(r3), "r"(r4), "r"(r5), "r"(r6), "r"(r7), "r"(r8),
519 "r"(r9), "r"(r10), "r"(r11)
520 : "memory", "cc", "xer", "ctr", "lr");
522 out[0] = r4;
523 out[1] = r5;
524 out[2] = r6;
525 out[3] = r7;
526 out[4] = r8;
527 out[5] = r9;
528 out[6] = r10;
529 out[7] = r11;
531 return r3;
533 EXPORT_SYMBOL_GPL(kvm_hypercall);
535 static int kvm_para_setup(void)
537 extern u32 kvm_hypercall_start;
538 struct device_node *hyper_node;
539 u32 *insts;
540 int len, i;
542 hyper_node = of_find_node_by_path("/hypervisor");
543 if (!hyper_node)
544 return -1;
546 insts = (u32*)of_get_property(hyper_node, "hcall-instructions", &len);
547 if (len % 4)
548 return -1;
549 if (len > (4 * 4))
550 return -1;
552 for (i = 0; i < (len / 4); i++)
553 kvm_patch_ins(&(&kvm_hypercall_start)[i], insts[i]);
555 return 0;
558 static __init void kvm_free_tmp(void)
560 unsigned long start, end;
562 start = (ulong)&kvm_tmp[kvm_tmp_index + (PAGE_SIZE - 1)] & PAGE_MASK;
563 end = (ulong)&kvm_tmp[ARRAY_SIZE(kvm_tmp)] & PAGE_MASK;
565 /* Free the tmp space we don't need */
566 for (; start < end; start += PAGE_SIZE) {
567 ClearPageReserved(virt_to_page(start));
568 init_page_count(virt_to_page(start));
569 free_page(start);
570 totalram_pages++;
574 static int __init kvm_guest_init(void)
576 if (!kvm_para_available())
577 goto free_tmp;
579 if (kvm_para_setup())
580 goto free_tmp;
582 if (kvm_para_has_feature(KVM_FEATURE_MAGIC_PAGE))
583 kvm_use_magic_page();
585 #ifdef CONFIG_PPC_BOOK3S_64
586 /* Enable napping */
587 powersave_nap = 1;
588 #endif
590 free_tmp:
591 kvm_free_tmp();
593 return 0;
596 postcore_initcall(kvm_guest_init);