Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
e1dab14c VK |
2 | /* |
3 | * Core of Xen paravirt_ops implementation. | |
4 | * | |
5 | * This file contains the xen_paravirt_ops structure itself, and the | |
6 | * implementations for: | |
7 | * - privileged instructions | |
8 | * - interrupt flags | |
9 | * - segment operations | |
10 | * - booting and setup | |
11 | * | |
12 | * Jeremy Fitzhardinge <jeremy@xensource.com>, XenSource Inc, 2007 | |
13 | */ | |
14 | ||
15 | #include <linux/cpu.h> | |
16 | #include <linux/kernel.h> | |
17 | #include <linux/init.h> | |
18 | #include <linux/smp.h> | |
19 | #include <linux/preempt.h> | |
20 | #include <linux/hardirq.h> | |
21 | #include <linux/percpu.h> | |
22 | #include <linux/delay.h> | |
23 | #include <linux/start_kernel.h> | |
24 | #include <linux/sched.h> | |
25 | #include <linux/kprobes.h> | |
26 | #include <linux/bootmem.h> | |
27 | #include <linux/export.h> | |
28 | #include <linux/mm.h> | |
29 | #include <linux/page-flags.h> | |
30 | #include <linux/highmem.h> | |
31 | #include <linux/console.h> | |
32 | #include <linux/pci.h> | |
33 | #include <linux/gfp.h> | |
34 | #include <linux/memblock.h> | |
35 | #include <linux/edd.h> | |
36 | #include <linux/frame.h> | |
37 | ||
38 | #include <xen/xen.h> | |
39 | #include <xen/events.h> | |
40 | #include <xen/interface/xen.h> | |
41 | #include <xen/interface/version.h> | |
42 | #include <xen/interface/physdev.h> | |
43 | #include <xen/interface/vcpu.h> | |
44 | #include <xen/interface/memory.h> | |
45 | #include <xen/interface/nmi.h> | |
46 | #include <xen/interface/xen-mca.h> | |
47 | #include <xen/features.h> | |
48 | #include <xen/page.h> | |
49 | #include <xen/hvc-console.h> | |
50 | #include <xen/acpi.h> | |
51 | ||
52 | #include <asm/paravirt.h> | |
53 | #include <asm/apic.h> | |
54 | #include <asm/page.h> | |
55 | #include <asm/xen/pci.h> | |
56 | #include <asm/xen/hypercall.h> | |
57 | #include <asm/xen/hypervisor.h> | |
58 | #include <asm/xen/cpuid.h> | |
59 | #include <asm/fixmap.h> | |
60 | #include <asm/processor.h> | |
61 | #include <asm/proto.h> | |
62 | #include <asm/msr-index.h> | |
63 | #include <asm/traps.h> | |
64 | #include <asm/setup.h> | |
65 | #include <asm/desc.h> | |
66 | #include <asm/pgalloc.h> | |
67 | #include <asm/pgtable.h> | |
68 | #include <asm/tlbflush.h> | |
69 | #include <asm/reboot.h> | |
70 | #include <asm/stackprotector.h> | |
71 | #include <asm/hypervisor.h> | |
72 | #include <asm/mach_traps.h> | |
73 | #include <asm/mwait.h> | |
74 | #include <asm/pci_x86.h> | |
75 | #include <asm/cpu.h> | |
76 | ||
77 | #ifdef CONFIG_ACPI | |
78 | #include <linux/acpi.h> | |
79 | #include <asm/acpi.h> | |
80 | #include <acpi/pdc_intel.h> | |
81 | #include <acpi/processor.h> | |
82 | #include <xen/interface/platform.h> | |
83 | #endif | |
84 | ||
85 | #include "xen-ops.h" | |
86 | #include "mmu.h" | |
87 | #include "smp.h" | |
88 | #include "multicalls.h" | |
89 | #include "pmu.h" | |
90 | ||
91 | void *xen_initial_gdt; | |
92 | ||
e1dab14c VK |
93 | static int xen_cpu_up_prepare_pv(unsigned int cpu); |
94 | static int xen_cpu_dead_pv(unsigned int cpu); | |
95 | ||
96 | struct tls_descs { | |
97 | struct desc_struct desc[3]; | |
98 | }; | |
99 | ||
100 | /* | |
101 | * Updating the 3 TLS descriptors in the GDT on every task switch is | |
102 | * surprisingly expensive so we avoid updating them if they haven't | |
103 | * changed. Since Xen writes different descriptors than the one | |
104 | * passed in the update_descriptor hypercall we keep shadow copies to | |
105 | * compare against. | |
106 | */ | |
107 | static DEFINE_PER_CPU(struct tls_descs, shadow_tls_desc); | |
108 | ||
e1dab14c VK |
109 | static void __init xen_banner(void) |
110 | { | |
111 | unsigned version = HYPERVISOR_xen_version(XENVER_version, NULL); | |
112 | struct xen_extraversion extra; | |
113 | HYPERVISOR_xen_version(XENVER_extraversion, &extra); | |
114 | ||
989513a7 | 115 | pr_info("Booting paravirtualized kernel on %s\n", pv_info.name); |
e1dab14c VK |
116 | printk(KERN_INFO "Xen version: %d.%d%s%s\n", |
117 | version >> 16, version & 0xffff, extra.extraversion, | |
118 | xen_feature(XENFEAT_mmu_pt_update_preserve_ad) ? " (preserve-AD)" : ""); | |
119 | } | |
120 | /* Check if running on Xen version (major, minor) or later */ | |
121 | bool | |
122 | xen_running_on_version_or_later(unsigned int major, unsigned int minor) | |
123 | { | |
124 | unsigned int version; | |
125 | ||
126 | if (!xen_domain()) | |
127 | return false; | |
128 | ||
129 | version = HYPERVISOR_xen_version(XENVER_version, NULL); | |
130 | if ((((version >> 16) == major) && ((version & 0xffff) >= minor)) || | |
131 | ((version >> 16) > major)) | |
132 | return true; | |
133 | return false; | |
134 | } | |
135 | ||
e1dab14c VK |
136 | static __read_mostly unsigned int cpuid_leaf5_ecx_val; |
137 | static __read_mostly unsigned int cpuid_leaf5_edx_val; | |
138 | ||
139 | static void xen_cpuid(unsigned int *ax, unsigned int *bx, | |
140 | unsigned int *cx, unsigned int *dx) | |
141 | { | |
142 | unsigned maskebx = ~0; | |
6807cf65 | 143 | |
e1dab14c VK |
144 | /* |
145 | * Mask out inconvenient features, to try and disable as many | |
146 | * unsupported kernel subsystems as possible. | |
147 | */ | |
148 | switch (*ax) { | |
e1dab14c VK |
149 | case CPUID_MWAIT_LEAF: |
150 | /* Synthesize the values.. */ | |
151 | *ax = 0; | |
152 | *bx = 0; | |
153 | *cx = cpuid_leaf5_ecx_val; | |
154 | *dx = cpuid_leaf5_edx_val; | |
155 | return; | |
156 | ||
e1dab14c VK |
157 | case 0xb: |
158 | /* Suppress extended topology stuff */ | |
159 | maskebx = 0; | |
160 | break; | |
161 | } | |
162 | ||
163 | asm(XEN_EMULATE_PREFIX "cpuid" | |
164 | : "=a" (*ax), | |
165 | "=b" (*bx), | |
166 | "=c" (*cx), | |
167 | "=d" (*dx) | |
168 | : "0" (*ax), "2" (*cx)); | |
169 | ||
170 | *bx &= maskebx; | |
e1dab14c VK |
171 | } |
172 | STACK_FRAME_NON_STANDARD(xen_cpuid); /* XEN_EMULATE_PREFIX */ | |
173 | ||
174 | static bool __init xen_check_mwait(void) | |
175 | { | |
176 | #ifdef CONFIG_ACPI | |
177 | struct xen_platform_op op = { | |
178 | .cmd = XENPF_set_processor_pminfo, | |
179 | .u.set_pminfo.id = -1, | |
180 | .u.set_pminfo.type = XEN_PM_PDC, | |
181 | }; | |
182 | uint32_t buf[3]; | |
183 | unsigned int ax, bx, cx, dx; | |
184 | unsigned int mwait_mask; | |
185 | ||
186 | /* We need to determine whether it is OK to expose the MWAIT | |
187 | * capability to the kernel to harvest deeper than C3 states from ACPI | |
188 | * _CST using the processor_harvest_xen.c module. For this to work, we | |
189 | * need to gather the MWAIT_LEAF values (which the cstate.c code | |
190 | * checks against). The hypervisor won't expose the MWAIT flag because | |
191 | * it would break backwards compatibility; so we will find out directly | |
192 | * from the hardware and hypercall. | |
193 | */ | |
194 | if (!xen_initial_domain()) | |
195 | return false; | |
196 | ||
197 | /* | |
198 | * When running under platform earlier than Xen4.2, do not expose | |
199 | * mwait, to avoid the risk of loading native acpi pad driver | |
200 | */ | |
201 | if (!xen_running_on_version_or_later(4, 2)) | |
202 | return false; | |
203 | ||
204 | ax = 1; | |
205 | cx = 0; | |
206 | ||
207 | native_cpuid(&ax, &bx, &cx, &dx); | |
208 | ||
209 | mwait_mask = (1 << (X86_FEATURE_EST % 32)) | | |
210 | (1 << (X86_FEATURE_MWAIT % 32)); | |
211 | ||
212 | if ((cx & mwait_mask) != mwait_mask) | |
213 | return false; | |
214 | ||
215 | /* We need to emulate the MWAIT_LEAF and for that we need both | |
216 | * ecx and edx. The hypercall provides only partial information. | |
217 | */ | |
218 | ||
219 | ax = CPUID_MWAIT_LEAF; | |
220 | bx = 0; | |
221 | cx = 0; | |
222 | dx = 0; | |
223 | ||
224 | native_cpuid(&ax, &bx, &cx, &dx); | |
225 | ||
226 | /* Ask the Hypervisor whether to clear ACPI_PDC_C_C2C3_FFH. If so, | |
227 | * don't expose MWAIT_LEAF and let ACPI pick the IOPORT version of C3. | |
228 | */ | |
229 | buf[0] = ACPI_PDC_REVISION_ID; | |
230 | buf[1] = 1; | |
231 | buf[2] = (ACPI_PDC_C_CAPABILITY_SMP | ACPI_PDC_EST_CAPABILITY_SWSMP); | |
232 | ||
233 | set_xen_guest_handle(op.u.set_pminfo.pdc, buf); | |
234 | ||
235 | if ((HYPERVISOR_platform_op(&op) == 0) && | |
236 | (buf[2] & (ACPI_PDC_C_C1_FFH | ACPI_PDC_C_C2C3_FFH))) { | |
237 | cpuid_leaf5_ecx_val = cx; | |
238 | cpuid_leaf5_edx_val = dx; | |
239 | } | |
240 | return true; | |
241 | #else | |
242 | return false; | |
243 | #endif | |
244 | } | |
e1dab14c | 245 | |
6807cf65 JG |
246 | static bool __init xen_check_xsave(void) |
247 | { | |
40f4ac0b | 248 | unsigned int cx, xsave_mask; |
e1dab14c | 249 | |
40f4ac0b JG |
250 | cx = cpuid_ecx(1); |
251 | ||
252 | xsave_mask = (1 << (X86_FEATURE_XSAVE % 32)) | | |
253 | (1 << (X86_FEATURE_OSXSAVE % 32)); | |
254 | ||
255 | /* Xen will set CR4.OSXSAVE if supported and not disabled by force */ | |
256 | return (cx & xsave_mask) == xsave_mask; | |
e1dab14c VK |
257 | } |
258 | ||
0808e80c JG |
259 | static void __init xen_init_capabilities(void) |
260 | { | |
0808e80c | 261 | setup_force_cpu_cap(X86_FEATURE_XENPV); |
3ee99df3 | 262 | setup_clear_cpu_cap(X86_FEATURE_DCA); |
fd9145fd | 263 | setup_clear_cpu_cap(X86_FEATURE_APERFMPERF); |
88f3256f | 264 | setup_clear_cpu_cap(X86_FEATURE_MTRR); |
aa107156 | 265 | setup_clear_cpu_cap(X86_FEATURE_ACC); |
e657fccb | 266 | setup_clear_cpu_cap(X86_FEATURE_X2APIC); |
f2f931c6 | 267 | setup_clear_cpu_cap(X86_FEATURE_SME); |
b778d6bf | 268 | |
660da7c9 AL |
269 | /* |
270 | * Xen PV would need some work to support PCID: CR3 handling as well | |
271 | * as xen_flush_tlb_others() would need updating. | |
272 | */ | |
273 | setup_clear_cpu_cap(X86_FEATURE_PCID); | |
b778d6bf JG |
274 | |
275 | if (!xen_initial_domain()) | |
276 | setup_clear_cpu_cap(X86_FEATURE_ACPI); | |
ea01598b JG |
277 | |
278 | if (xen_check_mwait()) | |
279 | setup_force_cpu_cap(X86_FEATURE_MWAIT); | |
280 | else | |
281 | setup_clear_cpu_cap(X86_FEATURE_MWAIT); | |
6807cf65 | 282 | |
40f4ac0b | 283 | if (!xen_check_xsave()) { |
6807cf65 JG |
284 | setup_clear_cpu_cap(X86_FEATURE_XSAVE); |
285 | setup_clear_cpu_cap(X86_FEATURE_OSXSAVE); | |
286 | } | |
0808e80c JG |
287 | } |
288 | ||
e1dab14c VK |
289 | static void xen_set_debugreg(int reg, unsigned long val) |
290 | { | |
291 | HYPERVISOR_set_debugreg(reg, val); | |
292 | } | |
293 | ||
294 | static unsigned long xen_get_debugreg(int reg) | |
295 | { | |
296 | return HYPERVISOR_get_debugreg(reg); | |
297 | } | |
298 | ||
299 | static void xen_end_context_switch(struct task_struct *next) | |
300 | { | |
301 | xen_mc_flush(); | |
302 | paravirt_end_context_switch(next); | |
303 | } | |
304 | ||
305 | static unsigned long xen_store_tr(void) | |
306 | { | |
307 | return 0; | |
308 | } | |
309 | ||
310 | /* | |
311 | * Set the page permissions for a particular virtual address. If the | |
312 | * address is a vmalloc mapping (or other non-linear mapping), then | |
313 | * find the linear mapping of the page and also set its protections to | |
314 | * match. | |
315 | */ | |
316 | static void set_aliased_prot(void *v, pgprot_t prot) | |
317 | { | |
318 | int level; | |
319 | pte_t *ptep; | |
320 | pte_t pte; | |
321 | unsigned long pfn; | |
322 | struct page *page; | |
323 | unsigned char dummy; | |
324 | ||
325 | ptep = lookup_address((unsigned long)v, &level); | |
326 | BUG_ON(ptep == NULL); | |
327 | ||
328 | pfn = pte_pfn(*ptep); | |
329 | page = pfn_to_page(pfn); | |
330 | ||
331 | pte = pfn_pte(pfn, prot); | |
332 | ||
333 | /* | |
334 | * Careful: update_va_mapping() will fail if the virtual address | |
335 | * we're poking isn't populated in the page tables. We don't | |
336 | * need to worry about the direct map (that's always in the page | |
337 | * tables), but we need to be careful about vmap space. In | |
338 | * particular, the top level page table can lazily propagate | |
339 | * entries between processes, so if we've switched mms since we | |
340 | * vmapped the target in the first place, we might not have the | |
341 | * top-level page table entry populated. | |
342 | * | |
343 | * We disable preemption because we want the same mm active when | |
344 | * we probe the target and when we issue the hypercall. We'll | |
345 | * have the same nominal mm, but if we're a kernel thread, lazy | |
346 | * mm dropping could change our pgd. | |
347 | * | |
348 | * Out of an abundance of caution, this uses __get_user() to fault | |
349 | * in the target address just in case there's some obscure case | |
350 | * in which the target address isn't readable. | |
351 | */ | |
352 | ||
353 | preempt_disable(); | |
354 | ||
355 | probe_kernel_read(&dummy, v, 1); | |
356 | ||
357 | if (HYPERVISOR_update_va_mapping((unsigned long)v, pte, 0)) | |
358 | BUG(); | |
359 | ||
360 | if (!PageHighMem(page)) { | |
361 | void *av = __va(PFN_PHYS(pfn)); | |
362 | ||
363 | if (av != v) | |
364 | if (HYPERVISOR_update_va_mapping((unsigned long)av, pte, 0)) | |
365 | BUG(); | |
366 | } else | |
367 | kmap_flush_unused(); | |
368 | ||
369 | preempt_enable(); | |
370 | } | |
371 | ||
372 | static void xen_alloc_ldt(struct desc_struct *ldt, unsigned entries) | |
373 | { | |
374 | const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; | |
375 | int i; | |
376 | ||
377 | /* | |
378 | * We need to mark the all aliases of the LDT pages RO. We | |
379 | * don't need to call vm_flush_aliases(), though, since that's | |
380 | * only responsible for flushing aliases out the TLBs, not the | |
381 | * page tables, and Xen will flush the TLB for us if needed. | |
382 | * | |
383 | * To avoid confusing future readers: none of this is necessary | |
384 | * to load the LDT. The hypervisor only checks this when the | |
385 | * LDT is faulted in due to subsequent descriptor access. | |
386 | */ | |
387 | ||
388 | for (i = 0; i < entries; i += entries_per_page) | |
389 | set_aliased_prot(ldt + i, PAGE_KERNEL_RO); | |
390 | } | |
391 | ||
392 | static void xen_free_ldt(struct desc_struct *ldt, unsigned entries) | |
393 | { | |
394 | const unsigned entries_per_page = PAGE_SIZE / LDT_ENTRY_SIZE; | |
395 | int i; | |
396 | ||
397 | for (i = 0; i < entries; i += entries_per_page) | |
398 | set_aliased_prot(ldt + i, PAGE_KERNEL); | |
399 | } | |
400 | ||
401 | static void xen_set_ldt(const void *addr, unsigned entries) | |
402 | { | |
403 | struct mmuext_op *op; | |
404 | struct multicall_space mcs = xen_mc_entry(sizeof(*op)); | |
405 | ||
406 | trace_xen_cpu_set_ldt(addr, entries); | |
407 | ||
408 | op = mcs.args; | |
409 | op->cmd = MMUEXT_SET_LDT; | |
410 | op->arg1.linear_addr = (unsigned long)addr; | |
411 | op->arg2.nr_ents = entries; | |
412 | ||
413 | MULTI_mmuext_op(mcs.mc, op, 1, NULL, DOMID_SELF); | |
414 | ||
415 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
416 | } | |
417 | ||
418 | static void xen_load_gdt(const struct desc_ptr *dtr) | |
419 | { | |
420 | unsigned long va = dtr->address; | |
421 | unsigned int size = dtr->size + 1; | |
422 | unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE); | |
423 | unsigned long frames[pages]; | |
424 | int f; | |
425 | ||
426 | /* | |
427 | * A GDT can be up to 64k in size, which corresponds to 8192 | |
428 | * 8-byte entries, or 16 4k pages.. | |
429 | */ | |
430 | ||
431 | BUG_ON(size > 65536); | |
432 | BUG_ON(va & ~PAGE_MASK); | |
433 | ||
434 | for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) { | |
435 | int level; | |
436 | pte_t *ptep; | |
437 | unsigned long pfn, mfn; | |
438 | void *virt; | |
439 | ||
440 | /* | |
441 | * The GDT is per-cpu and is in the percpu data area. | |
442 | * That can be virtually mapped, so we need to do a | |
443 | * page-walk to get the underlying MFN for the | |
444 | * hypercall. The page can also be in the kernel's | |
445 | * linear range, so we need to RO that mapping too. | |
446 | */ | |
447 | ptep = lookup_address(va, &level); | |
448 | BUG_ON(ptep == NULL); | |
449 | ||
450 | pfn = pte_pfn(*ptep); | |
451 | mfn = pfn_to_mfn(pfn); | |
452 | virt = __va(PFN_PHYS(pfn)); | |
453 | ||
454 | frames[f] = mfn; | |
455 | ||
456 | make_lowmem_page_readonly((void *)va); | |
457 | make_lowmem_page_readonly(virt); | |
458 | } | |
459 | ||
460 | if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct))) | |
461 | BUG(); | |
462 | } | |
463 | ||
464 | /* | |
465 | * load_gdt for early boot, when the gdt is only mapped once | |
466 | */ | |
467 | static void __init xen_load_gdt_boot(const struct desc_ptr *dtr) | |
468 | { | |
469 | unsigned long va = dtr->address; | |
470 | unsigned int size = dtr->size + 1; | |
471 | unsigned pages = DIV_ROUND_UP(size, PAGE_SIZE); | |
472 | unsigned long frames[pages]; | |
473 | int f; | |
474 | ||
475 | /* | |
476 | * A GDT can be up to 64k in size, which corresponds to 8192 | |
477 | * 8-byte entries, or 16 4k pages.. | |
478 | */ | |
479 | ||
480 | BUG_ON(size > 65536); | |
481 | BUG_ON(va & ~PAGE_MASK); | |
482 | ||
483 | for (f = 0; va < dtr->address + size; va += PAGE_SIZE, f++) { | |
484 | pte_t pte; | |
485 | unsigned long pfn, mfn; | |
486 | ||
487 | pfn = virt_to_pfn(va); | |
488 | mfn = pfn_to_mfn(pfn); | |
489 | ||
490 | pte = pfn_pte(pfn, PAGE_KERNEL_RO); | |
491 | ||
492 | if (HYPERVISOR_update_va_mapping((unsigned long)va, pte, 0)) | |
493 | BUG(); | |
494 | ||
495 | frames[f] = mfn; | |
496 | } | |
497 | ||
498 | if (HYPERVISOR_set_gdt(frames, size / sizeof(struct desc_struct))) | |
499 | BUG(); | |
500 | } | |
501 | ||
502 | static inline bool desc_equal(const struct desc_struct *d1, | |
503 | const struct desc_struct *d2) | |
504 | { | |
9a98e778 | 505 | return !memcmp(d1, d2, sizeof(*d1)); |
e1dab14c VK |
506 | } |
507 | ||
508 | static void load_TLS_descriptor(struct thread_struct *t, | |
509 | unsigned int cpu, unsigned int i) | |
510 | { | |
511 | struct desc_struct *shadow = &per_cpu(shadow_tls_desc, cpu).desc[i]; | |
512 | struct desc_struct *gdt; | |
513 | xmaddr_t maddr; | |
514 | struct multicall_space mc; | |
515 | ||
516 | if (desc_equal(shadow, &t->tls_array[i])) | |
517 | return; | |
518 | ||
519 | *shadow = t->tls_array[i]; | |
520 | ||
521 | gdt = get_cpu_gdt_rw(cpu); | |
522 | maddr = arbitrary_virt_to_machine(&gdt[GDT_ENTRY_TLS_MIN+i]); | |
523 | mc = __xen_mc_entry(0); | |
524 | ||
525 | MULTI_update_descriptor(mc.mc, maddr.maddr, t->tls_array[i]); | |
526 | } | |
527 | ||
528 | static void xen_load_tls(struct thread_struct *t, unsigned int cpu) | |
529 | { | |
530 | /* | |
531 | * XXX sleazy hack: If we're being called in a lazy-cpu zone | |
532 | * and lazy gs handling is enabled, it means we're in a | |
533 | * context switch, and %gs has just been saved. This means we | |
534 | * can zero it out to prevent faults on exit from the | |
535 | * hypervisor if the next process has no %gs. Either way, it | |
536 | * has been saved, and the new value will get loaded properly. | |
537 | * This will go away as soon as Xen has been modified to not | |
538 | * save/restore %gs for normal hypercalls. | |
539 | * | |
540 | * On x86_64, this hack is not used for %gs, because gs points | |
541 | * to KERNEL_GS_BASE (and uses it for PDA references), so we | |
542 | * must not zero %gs on x86_64 | |
543 | * | |
544 | * For x86_64, we need to zero %fs, otherwise we may get an | |
545 | * exception between the new %fs descriptor being loaded and | |
546 | * %fs being effectively cleared at __switch_to(). | |
547 | */ | |
548 | if (paravirt_get_lazy_mode() == PARAVIRT_LAZY_CPU) { | |
549 | #ifdef CONFIG_X86_32 | |
550 | lazy_load_gs(0); | |
551 | #else | |
552 | loadsegment(fs, 0); | |
553 | #endif | |
554 | } | |
555 | ||
556 | xen_mc_batch(); | |
557 | ||
558 | load_TLS_descriptor(t, cpu, 0); | |
559 | load_TLS_descriptor(t, cpu, 1); | |
560 | load_TLS_descriptor(t, cpu, 2); | |
561 | ||
562 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
563 | } | |
564 | ||
565 | #ifdef CONFIG_X86_64 | |
566 | static void xen_load_gs_index(unsigned int idx) | |
567 | { | |
568 | if (HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, idx)) | |
569 | BUG(); | |
570 | } | |
571 | #endif | |
572 | ||
573 | static void xen_write_ldt_entry(struct desc_struct *dt, int entrynum, | |
574 | const void *ptr) | |
575 | { | |
576 | xmaddr_t mach_lp = arbitrary_virt_to_machine(&dt[entrynum]); | |
577 | u64 entry = *(u64 *)ptr; | |
578 | ||
579 | trace_xen_cpu_write_ldt_entry(dt, entrynum, entry); | |
580 | ||
581 | preempt_disable(); | |
582 | ||
583 | xen_mc_flush(); | |
584 | if (HYPERVISOR_update_descriptor(mach_lp.maddr, entry)) | |
585 | BUG(); | |
586 | ||
587 | preempt_enable(); | |
588 | } | |
589 | ||
5878d5d6 JG |
590 | #ifdef CONFIG_X86_64 |
591 | struct trap_array_entry { | |
592 | void (*orig)(void); | |
593 | void (*xen)(void); | |
594 | bool ist_okay; | |
595 | }; | |
596 | ||
597 | static struct trap_array_entry trap_array[] = { | |
598 | { debug, xen_xendebug, true }, | |
599 | { int3, xen_xenint3, true }, | |
600 | { double_fault, xen_double_fault, true }, | |
601 | #ifdef CONFIG_X86_MCE | |
602 | { machine_check, xen_machine_check, true }, | |
603 | #endif | |
604 | { nmi, xen_nmi, true }, | |
605 | { overflow, xen_overflow, false }, | |
606 | #ifdef CONFIG_IA32_EMULATION | |
607 | { entry_INT80_compat, xen_entry_INT80_compat, false }, | |
608 | #endif | |
609 | { page_fault, xen_page_fault, false }, | |
610 | { divide_error, xen_divide_error, false }, | |
611 | { bounds, xen_bounds, false }, | |
612 | { invalid_op, xen_invalid_op, false }, | |
613 | { device_not_available, xen_device_not_available, false }, | |
614 | { coprocessor_segment_overrun, xen_coprocessor_segment_overrun, false }, | |
615 | { invalid_TSS, xen_invalid_TSS, false }, | |
616 | { segment_not_present, xen_segment_not_present, false }, | |
617 | { stack_segment, xen_stack_segment, false }, | |
618 | { general_protection, xen_general_protection, false }, | |
619 | { spurious_interrupt_bug, xen_spurious_interrupt_bug, false }, | |
620 | { coprocessor_error, xen_coprocessor_error, false }, | |
621 | { alignment_check, xen_alignment_check, false }, | |
622 | { simd_coprocessor_error, xen_simd_coprocessor_error, false }, | |
623 | }; | |
624 | ||
625 | static bool get_trap_addr(void **addr, unsigned int ist) | |
626 | { | |
627 | unsigned int nr; | |
628 | bool ist_okay = false; | |
629 | ||
630 | /* | |
631 | * Replace trap handler addresses by Xen specific ones. | |
632 | * Check for known traps using IST and whitelist them. | |
633 | * The debugger ones are the only ones we care about. | |
634 | * Xen will handle faults like double_fault, * so we should never see | |
635 | * them. Warn if there's an unexpected IST-using fault handler. | |
636 | */ | |
637 | for (nr = 0; nr < ARRAY_SIZE(trap_array); nr++) { | |
638 | struct trap_array_entry *entry = trap_array + nr; | |
639 | ||
640 | if (*addr == entry->orig) { | |
641 | *addr = entry->xen; | |
642 | ist_okay = entry->ist_okay; | |
643 | break; | |
644 | } | |
645 | } | |
646 | ||
647 | if (WARN_ON(ist != 0 && !ist_okay)) | |
648 | return false; | |
649 | ||
650 | return true; | |
651 | } | |
652 | #endif | |
653 | ||
e1dab14c VK |
654 | static int cvt_gate_to_trap(int vector, const gate_desc *val, |
655 | struct trap_info *info) | |
656 | { | |
657 | unsigned long addr; | |
658 | ||
64b163fa | 659 | if (val->bits.type != GATE_TRAP && val->bits.type != GATE_INTERRUPT) |
e1dab14c VK |
660 | return 0; |
661 | ||
662 | info->vector = vector; | |
663 | ||
64b163fa | 664 | addr = gate_offset(val); |
e1dab14c | 665 | #ifdef CONFIG_X86_64 |
5878d5d6 | 666 | if (!get_trap_addr((void **)&addr, val->bits.ist)) |
e1dab14c | 667 | return 0; |
e1dab14c VK |
668 | #endif /* CONFIG_X86_64 */ |
669 | info->address = addr; | |
670 | ||
64b163fa TG |
671 | info->cs = gate_segment(val); |
672 | info->flags = val->bits.dpl; | |
e1dab14c | 673 | /* interrupt gates clear IF */ |
64b163fa | 674 | if (val->bits.type == GATE_INTERRUPT) |
e1dab14c VK |
675 | info->flags |= 1 << 2; |
676 | ||
677 | return 1; | |
678 | } | |
679 | ||
680 | /* Locations of each CPU's IDT */ | |
681 | static DEFINE_PER_CPU(struct desc_ptr, idt_desc); | |
682 | ||
683 | /* Set an IDT entry. If the entry is part of the current IDT, then | |
684 | also update Xen. */ | |
685 | static void xen_write_idt_entry(gate_desc *dt, int entrynum, const gate_desc *g) | |
686 | { | |
687 | unsigned long p = (unsigned long)&dt[entrynum]; | |
688 | unsigned long start, end; | |
689 | ||
690 | trace_xen_cpu_write_idt_entry(dt, entrynum, g); | |
691 | ||
692 | preempt_disable(); | |
693 | ||
694 | start = __this_cpu_read(idt_desc.address); | |
695 | end = start + __this_cpu_read(idt_desc.size) + 1; | |
696 | ||
697 | xen_mc_flush(); | |
698 | ||
699 | native_write_idt_entry(dt, entrynum, g); | |
700 | ||
701 | if (p >= start && (p + 8) <= end) { | |
702 | struct trap_info info[2]; | |
703 | ||
704 | info[1].address = 0; | |
705 | ||
706 | if (cvt_gate_to_trap(entrynum, g, &info[0])) | |
707 | if (HYPERVISOR_set_trap_table(info)) | |
708 | BUG(); | |
709 | } | |
710 | ||
711 | preempt_enable(); | |
712 | } | |
713 | ||
714 | static void xen_convert_trap_info(const struct desc_ptr *desc, | |
715 | struct trap_info *traps) | |
716 | { | |
717 | unsigned in, out, count; | |
718 | ||
719 | count = (desc->size+1) / sizeof(gate_desc); | |
720 | BUG_ON(count > 256); | |
721 | ||
722 | for (in = out = 0; in < count; in++) { | |
723 | gate_desc *entry = (gate_desc *)(desc->address) + in; | |
724 | ||
725 | if (cvt_gate_to_trap(in, entry, &traps[out])) | |
726 | out++; | |
727 | } | |
728 | traps[out].address = 0; | |
729 | } | |
730 | ||
731 | void xen_copy_trap_info(struct trap_info *traps) | |
732 | { | |
733 | const struct desc_ptr *desc = this_cpu_ptr(&idt_desc); | |
734 | ||
735 | xen_convert_trap_info(desc, traps); | |
736 | } | |
737 | ||
738 | /* Load a new IDT into Xen. In principle this can be per-CPU, so we | |
739 | hold a spinlock to protect the static traps[] array (static because | |
740 | it avoids allocation, and saves stack space). */ | |
741 | static void xen_load_idt(const struct desc_ptr *desc) | |
742 | { | |
743 | static DEFINE_SPINLOCK(lock); | |
744 | static struct trap_info traps[257]; | |
745 | ||
746 | trace_xen_cpu_load_idt(desc); | |
747 | ||
748 | spin_lock(&lock); | |
749 | ||
750 | memcpy(this_cpu_ptr(&idt_desc), desc, sizeof(idt_desc)); | |
751 | ||
752 | xen_convert_trap_info(desc, traps); | |
753 | ||
754 | xen_mc_flush(); | |
755 | if (HYPERVISOR_set_trap_table(traps)) | |
756 | BUG(); | |
757 | ||
758 | spin_unlock(&lock); | |
759 | } | |
760 | ||
761 | /* Write a GDT descriptor entry. Ignore LDT descriptors, since | |
762 | they're handled differently. */ | |
763 | static void xen_write_gdt_entry(struct desc_struct *dt, int entry, | |
764 | const void *desc, int type) | |
765 | { | |
766 | trace_xen_cpu_write_gdt_entry(dt, entry, desc, type); | |
767 | ||
768 | preempt_disable(); | |
769 | ||
770 | switch (type) { | |
771 | case DESC_LDT: | |
772 | case DESC_TSS: | |
773 | /* ignore */ | |
774 | break; | |
775 | ||
776 | default: { | |
777 | xmaddr_t maddr = arbitrary_virt_to_machine(&dt[entry]); | |
778 | ||
779 | xen_mc_flush(); | |
780 | if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) | |
781 | BUG(); | |
782 | } | |
783 | ||
784 | } | |
785 | ||
786 | preempt_enable(); | |
787 | } | |
788 | ||
789 | /* | |
790 | * Version of write_gdt_entry for use at early boot-time needed to | |
791 | * update an entry as simply as possible. | |
792 | */ | |
793 | static void __init xen_write_gdt_entry_boot(struct desc_struct *dt, int entry, | |
794 | const void *desc, int type) | |
795 | { | |
796 | trace_xen_cpu_write_gdt_entry(dt, entry, desc, type); | |
797 | ||
798 | switch (type) { | |
799 | case DESC_LDT: | |
800 | case DESC_TSS: | |
801 | /* ignore */ | |
802 | break; | |
803 | ||
804 | default: { | |
805 | xmaddr_t maddr = virt_to_machine(&dt[entry]); | |
806 | ||
807 | if (HYPERVISOR_update_descriptor(maddr.maddr, *(u64 *)desc)) | |
808 | dt[entry] = *(struct desc_struct *)desc; | |
809 | } | |
810 | ||
811 | } | |
812 | } | |
813 | ||
814 | static void xen_load_sp0(struct tss_struct *tss, | |
815 | struct thread_struct *thread) | |
816 | { | |
817 | struct multicall_space mcs; | |
818 | ||
819 | mcs = xen_mc_entry(0); | |
820 | MULTI_stack_switch(mcs.mc, __KERNEL_DS, thread->sp0); | |
821 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
822 | tss->x86_tss.sp0 = thread->sp0; | |
823 | } | |
824 | ||
825 | void xen_set_iopl_mask(unsigned mask) | |
826 | { | |
827 | struct physdev_set_iopl set_iopl; | |
828 | ||
829 | /* Force the change at ring 0. */ | |
830 | set_iopl.iopl = (mask == 0) ? 1 : (mask >> 12) & 3; | |
831 | HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl); | |
832 | } | |
833 | ||
834 | static void xen_io_delay(void) | |
835 | { | |
836 | } | |
837 | ||
838 | static DEFINE_PER_CPU(unsigned long, xen_cr0_value); | |
839 | ||
840 | static unsigned long xen_read_cr0(void) | |
841 | { | |
842 | unsigned long cr0 = this_cpu_read(xen_cr0_value); | |
843 | ||
844 | if (unlikely(cr0 == 0)) { | |
845 | cr0 = native_read_cr0(); | |
846 | this_cpu_write(xen_cr0_value, cr0); | |
847 | } | |
848 | ||
849 | return cr0; | |
850 | } | |
851 | ||
852 | static void xen_write_cr0(unsigned long cr0) | |
853 | { | |
854 | struct multicall_space mcs; | |
855 | ||
856 | this_cpu_write(xen_cr0_value, cr0); | |
857 | ||
858 | /* Only pay attention to cr0.TS; everything else is | |
859 | ignored. */ | |
860 | mcs = xen_mc_entry(0); | |
861 | ||
862 | MULTI_fpu_taskswitch(mcs.mc, (cr0 & X86_CR0_TS) != 0); | |
863 | ||
864 | xen_mc_issue(PARAVIRT_LAZY_CPU); | |
865 | } | |
866 | ||
867 | static void xen_write_cr4(unsigned long cr4) | |
868 | { | |
869 | cr4 &= ~(X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PCE); | |
870 | ||
871 | native_write_cr4(cr4); | |
872 | } | |
873 | #ifdef CONFIG_X86_64 | |
874 | static inline unsigned long xen_read_cr8(void) | |
875 | { | |
876 | return 0; | |
877 | } | |
878 | static inline void xen_write_cr8(unsigned long val) | |
879 | { | |
880 | BUG_ON(val); | |
881 | } | |
882 | #endif | |
883 | ||
884 | static u64 xen_read_msr_safe(unsigned int msr, int *err) | |
885 | { | |
886 | u64 val; | |
887 | ||
888 | if (pmu_msr_read(msr, &val, err)) | |
889 | return val; | |
890 | ||
891 | val = native_read_msr_safe(msr, err); | |
892 | switch (msr) { | |
893 | case MSR_IA32_APICBASE: | |
894 | #ifdef CONFIG_X86_X2APIC | |
895 | if (!(cpuid_ecx(1) & (1 << (X86_FEATURE_X2APIC & 31)))) | |
896 | #endif | |
897 | val &= ~X2APIC_ENABLE; | |
898 | break; | |
899 | } | |
900 | return val; | |
901 | } | |
902 | ||
903 | static int xen_write_msr_safe(unsigned int msr, unsigned low, unsigned high) | |
904 | { | |
905 | int ret; | |
906 | ||
907 | ret = 0; | |
908 | ||
909 | switch (msr) { | |
910 | #ifdef CONFIG_X86_64 | |
911 | unsigned which; | |
912 | u64 base; | |
913 | ||
914 | case MSR_FS_BASE: which = SEGBASE_FS; goto set; | |
915 | case MSR_KERNEL_GS_BASE: which = SEGBASE_GS_USER; goto set; | |
916 | case MSR_GS_BASE: which = SEGBASE_GS_KERNEL; goto set; | |
917 | ||
918 | set: | |
919 | base = ((u64)high << 32) | low; | |
920 | if (HYPERVISOR_set_segment_base(which, base) != 0) | |
921 | ret = -EIO; | |
922 | break; | |
923 | #endif | |
924 | ||
925 | case MSR_STAR: | |
926 | case MSR_CSTAR: | |
927 | case MSR_LSTAR: | |
928 | case MSR_SYSCALL_MASK: | |
929 | case MSR_IA32_SYSENTER_CS: | |
930 | case MSR_IA32_SYSENTER_ESP: | |
931 | case MSR_IA32_SYSENTER_EIP: | |
932 | /* Fast syscall setup is all done in hypercalls, so | |
933 | these are all ignored. Stub them out here to stop | |
934 | Xen console noise. */ | |
935 | break; | |
936 | ||
937 | default: | |
938 | if (!pmu_msr_write(msr, low, high, &ret)) | |
939 | ret = native_write_msr_safe(msr, low, high); | |
940 | } | |
941 | ||
942 | return ret; | |
943 | } | |
944 | ||
945 | static u64 xen_read_msr(unsigned int msr) | |
946 | { | |
947 | /* | |
948 | * This will silently swallow a #GP from RDMSR. It may be worth | |
949 | * changing that. | |
950 | */ | |
951 | int err; | |
952 | ||
953 | return xen_read_msr_safe(msr, &err); | |
954 | } | |
955 | ||
956 | static void xen_write_msr(unsigned int msr, unsigned low, unsigned high) | |
957 | { | |
958 | /* | |
959 | * This will silently swallow a #GP from WRMSR. It may be worth | |
960 | * changing that. | |
961 | */ | |
962 | xen_write_msr_safe(msr, low, high); | |
963 | } | |
964 | ||
965 | void xen_setup_shared_info(void) | |
966 | { | |
989513a7 | 967 | set_fixmap(FIX_PARAVIRT_BOOTMAP, xen_start_info->shared_info); |
e1dab14c | 968 | |
989513a7 JG |
969 | HYPERVISOR_shared_info = |
970 | (struct shared_info *)fix_to_virt(FIX_PARAVIRT_BOOTMAP); | |
e1dab14c | 971 | |
e1dab14c | 972 | xen_setup_mfn_list_list(); |
d162809f | 973 | |
0e4d5837 AA |
974 | if (system_state == SYSTEM_BOOTING) { |
975 | #ifndef CONFIG_SMP | |
976 | /* | |
977 | * In UP this is as good a place as any to set up shared info. | |
978 | * Limit this to boot only, at restore vcpu setup is done via | |
979 | * xen_vcpu_restore(). | |
980 | */ | |
981 | xen_setup_vcpu_info_placement(); | |
982 | #endif | |
983 | /* | |
984 | * Now that shared info is set up we can start using routines | |
985 | * that point to pvclock area. | |
986 | */ | |
d162809f | 987 | xen_init_time_ops(); |
0e4d5837 | 988 | } |
e1dab14c VK |
989 | } |
990 | ||
991 | /* This is called once we have the cpu_possible_mask */ | |
0e4d5837 | 992 | void __ref xen_setup_vcpu_info_placement(void) |
e1dab14c VK |
993 | { |
994 | int cpu; | |
995 | ||
996 | for_each_possible_cpu(cpu) { | |
997 | /* Set up direct vCPU id mapping for PV guests. */ | |
998 | per_cpu(xen_vcpu_id, cpu) = cpu; | |
c9b5d98b AA |
999 | |
1000 | /* | |
1001 | * xen_vcpu_setup(cpu) can fail -- in which case it | |
1002 | * falls back to the shared_info version for cpus | |
1003 | * where xen_vcpu_nr(cpu) < MAX_VIRT_CPUS. | |
1004 | * | |
1005 | * xen_cpu_up_prepare_pv() handles the rest by failing | |
1006 | * them in hotplug. | |
1007 | */ | |
1008 | (void) xen_vcpu_setup(cpu); | |
e1dab14c VK |
1009 | } |
1010 | ||
1011 | /* | |
1012 | * xen_vcpu_setup managed to place the vcpu_info within the | |
1013 | * percpu area for all cpus, so make use of it. | |
1014 | */ | |
1015 | if (xen_have_vcpu_info_placement) { | |
1016 | pv_irq_ops.save_fl = __PV_IS_CALLEE_SAVE(xen_save_fl_direct); | |
1017 | pv_irq_ops.restore_fl = __PV_IS_CALLEE_SAVE(xen_restore_fl_direct); | |
1018 | pv_irq_ops.irq_disable = __PV_IS_CALLEE_SAVE(xen_irq_disable_direct); | |
1019 | pv_irq_ops.irq_enable = __PV_IS_CALLEE_SAVE(xen_irq_enable_direct); | |
1020 | pv_mmu_ops.read_cr2 = xen_read_cr2_direct; | |
1021 | } | |
1022 | } | |
1023 | ||
e1dab14c VK |
1024 | static const struct pv_info xen_info __initconst = { |
1025 | .shared_kernel_pmd = 0, | |
1026 | ||
1027 | #ifdef CONFIG_X86_64 | |
1028 | .extra_user_64bit_cs = FLAT_USER_CS64, | |
1029 | #endif | |
1030 | .name = "Xen", | |
1031 | }; | |
1032 | ||
e1dab14c VK |
1033 | static const struct pv_cpu_ops xen_cpu_ops __initconst = { |
1034 | .cpuid = xen_cpuid, | |
1035 | ||
1036 | .set_debugreg = xen_set_debugreg, | |
1037 | .get_debugreg = xen_get_debugreg, | |
1038 | ||
1039 | .read_cr0 = xen_read_cr0, | |
1040 | .write_cr0 = xen_write_cr0, | |
1041 | ||
e1dab14c VK |
1042 | .write_cr4 = xen_write_cr4, |
1043 | ||
1044 | #ifdef CONFIG_X86_64 | |
1045 | .read_cr8 = xen_read_cr8, | |
1046 | .write_cr8 = xen_write_cr8, | |
1047 | #endif | |
1048 | ||
1049 | .wbinvd = native_wbinvd, | |
1050 | ||
1051 | .read_msr = xen_read_msr, | |
1052 | .write_msr = xen_write_msr, | |
1053 | ||
1054 | .read_msr_safe = xen_read_msr_safe, | |
1055 | .write_msr_safe = xen_write_msr_safe, | |
1056 | ||
1057 | .read_pmc = xen_read_pmc, | |
1058 | ||
1059 | .iret = xen_iret, | |
1060 | #ifdef CONFIG_X86_64 | |
1061 | .usergs_sysret64 = xen_sysret64, | |
1062 | #endif | |
1063 | ||
1064 | .load_tr_desc = paravirt_nop, | |
1065 | .set_ldt = xen_set_ldt, | |
1066 | .load_gdt = xen_load_gdt, | |
1067 | .load_idt = xen_load_idt, | |
1068 | .load_tls = xen_load_tls, | |
1069 | #ifdef CONFIG_X86_64 | |
1070 | .load_gs_index = xen_load_gs_index, | |
1071 | #endif | |
1072 | ||
1073 | .alloc_ldt = xen_alloc_ldt, | |
1074 | .free_ldt = xen_free_ldt, | |
1075 | ||
e1dab14c VK |
1076 | .store_tr = xen_store_tr, |
1077 | ||
1078 | .write_ldt_entry = xen_write_ldt_entry, | |
1079 | .write_gdt_entry = xen_write_gdt_entry, | |
1080 | .write_idt_entry = xen_write_idt_entry, | |
1081 | .load_sp0 = xen_load_sp0, | |
1082 | ||
1083 | .set_iopl_mask = xen_set_iopl_mask, | |
1084 | .io_delay = xen_io_delay, | |
1085 | ||
1086 | /* Xen takes care of %gs when switching to usermode for us */ | |
1087 | .swapgs = paravirt_nop, | |
1088 | ||
1089 | .start_context_switch = paravirt_start_context_switch, | |
1090 | .end_context_switch = xen_end_context_switch, | |
1091 | }; | |
1092 | ||
1093 | static void xen_restart(char *msg) | |
1094 | { | |
1095 | xen_reboot(SHUTDOWN_reboot); | |
1096 | } | |
1097 | ||
1098 | static void xen_machine_halt(void) | |
1099 | { | |
1100 | xen_reboot(SHUTDOWN_poweroff); | |
1101 | } | |
1102 | ||
1103 | static void xen_machine_power_off(void) | |
1104 | { | |
1105 | if (pm_power_off) | |
1106 | pm_power_off(); | |
1107 | xen_reboot(SHUTDOWN_poweroff); | |
1108 | } | |
1109 | ||
1110 | static void xen_crash_shutdown(struct pt_regs *regs) | |
1111 | { | |
1112 | xen_reboot(SHUTDOWN_crash); | |
1113 | } | |
1114 | ||
1115 | static const struct machine_ops xen_machine_ops __initconst = { | |
1116 | .restart = xen_restart, | |
1117 | .halt = xen_machine_halt, | |
1118 | .power_off = xen_machine_power_off, | |
1119 | .shutdown = xen_machine_halt, | |
1120 | .crash_shutdown = xen_crash_shutdown, | |
1121 | .emergency_restart = xen_emergency_restart, | |
1122 | }; | |
1123 | ||
1124 | static unsigned char xen_get_nmi_reason(void) | |
1125 | { | |
1126 | unsigned char reason = 0; | |
1127 | ||
1128 | /* Construct a value which looks like it came from port 0x61. */ | |
1129 | if (test_bit(_XEN_NMIREASON_io_error, | |
1130 | &HYPERVISOR_shared_info->arch.nmi_reason)) | |
1131 | reason |= NMI_REASON_IOCHK; | |
1132 | if (test_bit(_XEN_NMIREASON_pci_serr, | |
1133 | &HYPERVISOR_shared_info->arch.nmi_reason)) | |
1134 | reason |= NMI_REASON_SERR; | |
1135 | ||
1136 | return reason; | |
1137 | } | |
1138 | ||
1139 | static void __init xen_boot_params_init_edd(void) | |
1140 | { | |
1141 | #if IS_ENABLED(CONFIG_EDD) | |
1142 | struct xen_platform_op op; | |
1143 | struct edd_info *edd_info; | |
1144 | u32 *mbr_signature; | |
1145 | unsigned nr; | |
1146 | int ret; | |
1147 | ||
1148 | edd_info = boot_params.eddbuf; | |
1149 | mbr_signature = boot_params.edd_mbr_sig_buffer; | |
1150 | ||
1151 | op.cmd = XENPF_firmware_info; | |
1152 | ||
1153 | op.u.firmware_info.type = XEN_FW_DISK_INFO; | |
1154 | for (nr = 0; nr < EDDMAXNR; nr++) { | |
1155 | struct edd_info *info = edd_info + nr; | |
1156 | ||
1157 | op.u.firmware_info.index = nr; | |
1158 | info->params.length = sizeof(info->params); | |
1159 | set_xen_guest_handle(op.u.firmware_info.u.disk_info.edd_params, | |
1160 | &info->params); | |
1161 | ret = HYPERVISOR_platform_op(&op); | |
1162 | if (ret) | |
1163 | break; | |
1164 | ||
1165 | #define C(x) info->x = op.u.firmware_info.u.disk_info.x | |
1166 | C(device); | |
1167 | C(version); | |
1168 | C(interface_support); | |
1169 | C(legacy_max_cylinder); | |
1170 | C(legacy_max_head); | |
1171 | C(legacy_sectors_per_track); | |
1172 | #undef C | |
1173 | } | |
1174 | boot_params.eddbuf_entries = nr; | |
1175 | ||
1176 | op.u.firmware_info.type = XEN_FW_DISK_MBR_SIGNATURE; | |
1177 | for (nr = 0; nr < EDD_MBR_SIG_MAX; nr++) { | |
1178 | op.u.firmware_info.index = nr; | |
1179 | ret = HYPERVISOR_platform_op(&op); | |
1180 | if (ret) | |
1181 | break; | |
1182 | mbr_signature[nr] = op.u.firmware_info.u.disk_mbr_signature.mbr_signature; | |
1183 | } | |
1184 | boot_params.edd_mbr_sig_buf_entries = nr; | |
1185 | #endif | |
1186 | } | |
1187 | ||
1188 | /* | |
1189 | * Set up the GDT and segment registers for -fstack-protector. Until | |
1190 | * we do this, we have to be careful not to call any stack-protected | |
1191 | * function, which is most of the kernel. | |
1192 | */ | |
1193 | static void xen_setup_gdt(int cpu) | |
1194 | { | |
1195 | pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry_boot; | |
1196 | pv_cpu_ops.load_gdt = xen_load_gdt_boot; | |
1197 | ||
1198 | setup_stack_canary_segment(0); | |
1199 | switch_to_new_gdt(0); | |
1200 | ||
1201 | pv_cpu_ops.write_gdt_entry = xen_write_gdt_entry; | |
1202 | pv_cpu_ops.load_gdt = xen_load_gdt; | |
1203 | } | |
1204 | ||
1205 | static void __init xen_dom0_set_legacy_features(void) | |
1206 | { | |
1207 | x86_platform.legacy.rtc = 1; | |
1208 | } | |
1209 | ||
1210 | /* First C function to be called on Xen boot */ | |
1211 | asmlinkage __visible void __init xen_start_kernel(void) | |
1212 | { | |
1213 | struct physdev_set_iopl set_iopl; | |
1214 | unsigned long initrd_start = 0; | |
1215 | int rc; | |
1216 | ||
1217 | if (!xen_start_info) | |
1218 | return; | |
1219 | ||
1220 | xen_domain_type = XEN_PV_DOMAIN; | |
1221 | ||
1222 | xen_setup_features(); | |
1223 | ||
1224 | xen_setup_machphys_mapping(); | |
1225 | ||
1226 | /* Install Xen paravirt ops */ | |
1227 | pv_info = xen_info; | |
edcb5cf8 | 1228 | pv_init_ops.patch = paravirt_patch_default; |
e1dab14c VK |
1229 | pv_cpu_ops = xen_cpu_ops; |
1230 | ||
1231 | x86_platform.get_nmi_reason = xen_get_nmi_reason; | |
1232 | ||
1233 | x86_init.resources.memory_setup = xen_memory_setup; | |
1234 | x86_init.oem.arch_setup = xen_arch_setup; | |
1235 | x86_init.oem.banner = xen_banner; | |
1236 | ||
e1dab14c VK |
1237 | /* |
1238 | * Set up some pagetable state before starting to set any ptes. | |
1239 | */ | |
1240 | ||
1241 | xen_init_mmu_ops(); | |
1242 | ||
1243 | /* Prevent unwanted bits from being set in PTEs. */ | |
1244 | __supported_pte_mask &= ~_PAGE_GLOBAL; | |
1245 | ||
1246 | /* | |
1247 | * Prevent page tables from being allocated in highmem, even | |
1248 | * if CONFIG_HIGHPTE is enabled. | |
1249 | */ | |
1250 | __userpte_alloc_gfp &= ~__GFP_HIGHMEM; | |
1251 | ||
1252 | /* Work out if we support NX */ | |
1253 | x86_configure_nx(); | |
1254 | ||
1255 | /* Get mfn list */ | |
1256 | xen_build_dynamic_phys_to_machine(); | |
1257 | ||
1258 | /* | |
1259 | * Set up kernel GDT and segment registers, mainly so that | |
1260 | * -fstack-protector code can be executed. | |
1261 | */ | |
1262 | xen_setup_gdt(0); | |
1263 | ||
1264 | xen_init_irq_ops(); | |
0808e80c | 1265 | xen_init_capabilities(); |
e1dab14c VK |
1266 | |
1267 | #ifdef CONFIG_X86_LOCAL_APIC | |
1268 | /* | |
1269 | * set up the basic apic ops. | |
1270 | */ | |
1271 | xen_init_apic(); | |
1272 | #endif | |
1273 | ||
1274 | if (xen_feature(XENFEAT_mmu_pt_update_preserve_ad)) { | |
1275 | pv_mmu_ops.ptep_modify_prot_start = xen_ptep_modify_prot_start; | |
1276 | pv_mmu_ops.ptep_modify_prot_commit = xen_ptep_modify_prot_commit; | |
1277 | } | |
1278 | ||
1279 | machine_ops = xen_machine_ops; | |
1280 | ||
1281 | /* | |
1282 | * The only reliable way to retain the initial address of the | |
1283 | * percpu gdt_page is to remember it here, so we can go and | |
1284 | * mark it RW later, when the initial percpu area is freed. | |
1285 | */ | |
1286 | xen_initial_gdt = &per_cpu(gdt_page, 0); | |
1287 | ||
1288 | xen_smp_init(); | |
1289 | ||
1290 | #ifdef CONFIG_ACPI_NUMA | |
1291 | /* | |
1292 | * The pages we from Xen are not related to machine pages, so | |
1293 | * any NUMA information the kernel tries to get from ACPI will | |
1294 | * be meaningless. Prevent it from trying. | |
1295 | */ | |
1296 | acpi_numa = -1; | |
1297 | #endif | |
ad73fd59 AA |
1298 | /* Let's presume PV guests always boot on vCPU with id 0. */ |
1299 | per_cpu(xen_vcpu_id, 0) = 0; | |
1300 | ||
1301 | /* | |
1302 | * Setup xen_vcpu early because start_kernel needs it for | |
1303 | * local_irq_disable(), irqs_disabled(). | |
1304 | * | |
1305 | * Don't do the full vcpu_info placement stuff until we have | |
1306 | * the cpu_possible_mask and a non-dummy shared_info. | |
1307 | */ | |
1308 | xen_vcpu_info_reset(0); | |
e1dab14c VK |
1309 | |
1310 | WARN_ON(xen_cpuhp_setup(xen_cpu_up_prepare_pv, xen_cpu_dead_pv)); | |
1311 | ||
1312 | local_irq_disable(); | |
1313 | early_boot_irqs_disabled = true; | |
1314 | ||
1315 | xen_raw_console_write("mapping kernel into physical memory\n"); | |
1316 | xen_setup_kernel_pagetable((pgd_t *)xen_start_info->pt_base, | |
1317 | xen_start_info->nr_pages); | |
1318 | xen_reserve_special_pages(); | |
1319 | ||
1320 | /* keep using Xen gdt for now; no urgent need to change it */ | |
1321 | ||
1322 | #ifdef CONFIG_X86_32 | |
1323 | pv_info.kernel_rpl = 1; | |
1324 | if (xen_feature(XENFEAT_supervisor_mode_kernel)) | |
1325 | pv_info.kernel_rpl = 0; | |
1326 | #else | |
1327 | pv_info.kernel_rpl = 0; | |
1328 | #endif | |
1329 | /* set the limit of our address space */ | |
1330 | xen_reserve_top(); | |
1331 | ||
1332 | /* | |
1333 | * We used to do this in xen_arch_setup, but that is too late | |
1334 | * on AMD were early_cpu_init (run before ->arch_setup()) calls | |
1335 | * early_amd_init which pokes 0xcf8 port. | |
1336 | */ | |
1337 | set_iopl.iopl = 1; | |
1338 | rc = HYPERVISOR_physdev_op(PHYSDEVOP_set_iopl, &set_iopl); | |
1339 | if (rc != 0) | |
1340 | xen_raw_printk("physdev_op failed %d\n", rc); | |
1341 | ||
1342 | #ifdef CONFIG_X86_32 | |
1343 | /* set up basic CPUID stuff */ | |
1344 | cpu_detect(&new_cpu_data); | |
1345 | set_cpu_cap(&new_cpu_data, X86_FEATURE_FPU); | |
1346 | new_cpu_data.x86_capability[CPUID_1_EDX] = cpuid_edx(1); | |
1347 | #endif | |
1348 | ||
1349 | if (xen_start_info->mod_start) { | |
1350 | if (xen_start_info->flags & SIF_MOD_START_PFN) | |
1351 | initrd_start = PFN_PHYS(xen_start_info->mod_start); | |
1352 | else | |
1353 | initrd_start = __pa(xen_start_info->mod_start); | |
1354 | } | |
1355 | ||
1356 | /* Poke various useful things into boot_params */ | |
1357 | boot_params.hdr.type_of_loader = (9 << 4) | 0; | |
1358 | boot_params.hdr.ramdisk_image = initrd_start; | |
1359 | boot_params.hdr.ramdisk_size = xen_start_info->mod_len; | |
1360 | boot_params.hdr.cmd_line_ptr = __pa(xen_start_info->cmd_line); | |
1361 | boot_params.hdr.hardware_subarch = X86_SUBARCH_XEN; | |
1362 | ||
1363 | if (!xen_initial_domain()) { | |
1364 | add_preferred_console("xenboot", 0, NULL); | |
1365 | add_preferred_console("tty", 0, NULL); | |
1366 | add_preferred_console("hvc", 0, NULL); | |
1367 | if (pci_xen) | |
1368 | x86_init.pci.arch_init = pci_xen_init; | |
1369 | } else { | |
1370 | const struct dom0_vga_console_info *info = | |
1371 | (void *)((char *)xen_start_info + | |
1372 | xen_start_info->console.dom0.info_off); | |
1373 | struct xen_platform_op op = { | |
1374 | .cmd = XENPF_firmware_info, | |
1375 | .interface_version = XENPF_INTERFACE_VERSION, | |
1376 | .u.firmware_info.type = XEN_FW_KBD_SHIFT_FLAGS, | |
1377 | }; | |
1378 | ||
1379 | x86_platform.set_legacy_features = | |
1380 | xen_dom0_set_legacy_features; | |
1381 | xen_init_vga(info, xen_start_info->console.dom0.info_size); | |
1382 | xen_start_info->console.domU.mfn = 0; | |
1383 | xen_start_info->console.domU.evtchn = 0; | |
1384 | ||
1385 | if (HYPERVISOR_platform_op(&op) == 0) | |
1386 | boot_params.kbd_status = op.u.firmware_info.u.kbd_shift_flags; | |
1387 | ||
1388 | /* Make sure ACS will be enabled */ | |
1389 | pci_request_acs(); | |
1390 | ||
1391 | xen_acpi_sleep_register(); | |
1392 | ||
1393 | /* Avoid searching for BIOS MP tables */ | |
1394 | x86_init.mpparse.find_smp_config = x86_init_noop; | |
1395 | x86_init.mpparse.get_smp_config = x86_init_uint_noop; | |
1396 | ||
1397 | xen_boot_params_init_edd(); | |
1398 | } | |
1399 | #ifdef CONFIG_PCI | |
1400 | /* PCI BIOS service won't work from a PV guest. */ | |
1401 | pci_probe &= ~PCI_PROBE_BIOS; | |
1402 | #endif | |
1403 | xen_raw_console_write("about to get started...\n"); | |
1404 | ||
ad73fd59 | 1405 | /* We need this for printk timestamps */ |
e1dab14c VK |
1406 | xen_setup_runstate_info(0); |
1407 | ||
1408 | xen_efi_init(); | |
1409 | ||
1410 | /* Start the world */ | |
1411 | #ifdef CONFIG_X86_32 | |
1412 | i386_start_kernel(); | |
1413 | #else | |
1414 | cr4_init_shadow(); /* 32b kernel does this in i386_start_kernel() */ | |
1415 | x86_64_start_reservations((char *)__pa_symbol(&boot_params)); | |
1416 | #endif | |
1417 | } | |
1418 | ||
1419 | static int xen_cpu_up_prepare_pv(unsigned int cpu) | |
1420 | { | |
1421 | int rc; | |
1422 | ||
c9b5d98b AA |
1423 | if (per_cpu(xen_vcpu, cpu) == NULL) |
1424 | return -ENODEV; | |
1425 | ||
e1dab14c VK |
1426 | xen_setup_timer(cpu); |
1427 | ||
1428 | rc = xen_smp_intr_init(cpu); | |
1429 | if (rc) { | |
1430 | WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n", | |
1431 | cpu, rc); | |
1432 | return rc; | |
1433 | } | |
04e95761 VK |
1434 | |
1435 | rc = xen_smp_intr_init_pv(cpu); | |
1436 | if (rc) { | |
1437 | WARN(1, "xen_smp_intr_init_pv() for CPU %d failed: %d\n", | |
1438 | cpu, rc); | |
1439 | return rc; | |
1440 | } | |
1441 | ||
e1dab14c VK |
1442 | return 0; |
1443 | } | |
1444 | ||
1445 | static int xen_cpu_dead_pv(unsigned int cpu) | |
1446 | { | |
1447 | xen_smp_intr_free(cpu); | |
04e95761 | 1448 | xen_smp_intr_free_pv(cpu); |
e1dab14c VK |
1449 | |
1450 | xen_teardown_timer(cpu); | |
1451 | ||
1452 | return 0; | |
1453 | } | |
1454 | ||
1455 | static uint32_t __init xen_platform_pv(void) | |
1456 | { | |
1457 | if (xen_pv_domain()) | |
1458 | return xen_cpuid_base(); | |
1459 | ||
1460 | return 0; | |
1461 | } | |
1462 | ||
e1dab14c VK |
1463 | const struct hypervisor_x86 x86_hyper_xen_pv = { |
1464 | .name = "Xen PV", | |
1465 | .detect = xen_platform_pv, | |
e1dab14c VK |
1466 | .pin_vcpu = xen_pin_vcpu, |
1467 | }; | |
1468 | EXPORT_SYMBOL(x86_hyper_xen_pv); |