1 /*-
2 * Copyright (c) 2004 Marcel Moolenaar
3 * Copyright (c) 2001 Doug Rabson
4 * Copyright (c) 2016 The FreeBSD Foundation
5 * All rights reserved.
6 *
7 * Portions of this software were developed by Konstantin Belousov
8 * under sponsorship from the FreeBSD Foundation.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34
35 #include <sys/param.h>
36 #include <sys/efi.h>
37 #include <sys/kernel.h>
38 #include <sys/linker.h>
39 #include <sys/lock.h>
40 #include <sys/module.h>
41 #include <sys/mutex.h>
42 #include <sys/clock.h>
43 #include <sys/proc.h>
44 #include <sys/rwlock.h>
45 #include <sys/sched.h>
46 #include <sys/sysctl.h>
47 #include <sys/systm.h>
48 #include <sys/vmmeter.h>
49 #include <isa/rtc.h>
50 #include <machine/fpu.h>
51 #include <machine/efi.h>
52 #include <machine/metadata.h>
53 #include <machine/md_var.h>
54 #include <machine/smp.h>
55 #include <machine/vmparam.h>
56 #include <vm/vm.h>
57 #include <vm/pmap.h>
58 #include <vm/vm_extern.h>
59 #include <vm/vm_map.h>
60 #include <vm/vm_object.h>
61 #include <vm/vm_page.h>
62 #include <vm/vm_pager.h>
63
64 static pml5_entry_t *efi_pml5;
65 static pml4_entry_t *efi_pml4;
66 static vm_object_t obj_1t1_pt;
67 static vm_page_t efi_pmltop_page;
68 static vm_pindex_t efi_1t1_idx;
69
70 void
efi_destroy_1t1_map(void)71 efi_destroy_1t1_map(void)
72 {
73 vm_page_t m;
74
75 if (obj_1t1_pt != NULL) {
76 VM_OBJECT_RLOCK(obj_1t1_pt);
77 TAILQ_FOREACH(m, &obj_1t1_pt->memq, listq)
78 m->ref_count = VPRC_OBJREF;
79 vm_wire_sub(obj_1t1_pt->resident_page_count);
80 VM_OBJECT_RUNLOCK(obj_1t1_pt);
81 vm_object_deallocate(obj_1t1_pt);
82 }
83
84 obj_1t1_pt = NULL;
85 efi_pml4 = NULL;
86 efi_pml5 = NULL;
87 efi_pmltop_page = NULL;
88 }
89
90 /*
91 * Map a physical address from EFI runtime space into KVA space. Returns 0 to
92 * indicate a failed mapping so that the caller may handle error.
93 */
94 vm_offset_t
efi_phys_to_kva(vm_paddr_t paddr)95 efi_phys_to_kva(vm_paddr_t paddr)
96 {
97
98 if (paddr >= dmaplimit)
99 return (0);
100 return (PHYS_TO_DMAP(paddr));
101 }
102
103 static vm_page_t
efi_1t1_page(void)104 efi_1t1_page(void)
105 {
106
107 return (vm_page_grab(obj_1t1_pt, efi_1t1_idx++, VM_ALLOC_NOBUSY |
108 VM_ALLOC_WIRED | VM_ALLOC_ZERO));
109 }
110
111 static pt_entry_t *
efi_1t1_pte(vm_offset_t va)112 efi_1t1_pte(vm_offset_t va)
113 {
114 pml5_entry_t *pml5e;
115 pml4_entry_t *pml4e;
116 pdp_entry_t *pdpe;
117 pd_entry_t *pde;
118 pt_entry_t *pte;
119 vm_page_t m;
120 vm_pindex_t pml5_idx, pml4_idx, pdp_idx, pd_idx;
121 vm_paddr_t mphys;
122
123 pml4_idx = pmap_pml4e_index(va);
124 if (la57) {
125 pml5_idx = pmap_pml5e_index(va);
126 pml5e = &efi_pml5[pml5_idx];
127 if (*pml5e == 0) {
128 m = efi_1t1_page();
129 mphys = VM_PAGE_TO_PHYS(m);
130 *pml5e = mphys | X86_PG_RW | X86_PG_V;
131 } else {
132 mphys = *pml5e & PG_FRAME;
133 }
134 pml4e = (pml4_entry_t *)PHYS_TO_DMAP(mphys);
135 pml4e = &pml4e[pml4_idx];
136 } else {
137 pml4e = &efi_pml4[pml4_idx];
138 }
139
140 if (*pml4e == 0) {
141 m = efi_1t1_page();
142 mphys = VM_PAGE_TO_PHYS(m);
143 *pml4e = mphys | X86_PG_RW | X86_PG_V;
144 } else {
145 mphys = *pml4e & PG_FRAME;
146 }
147
148 pdpe = (pdp_entry_t *)PHYS_TO_DMAP(mphys);
149 pdp_idx = pmap_pdpe_index(va);
150 pdpe += pdp_idx;
151 if (*pdpe == 0) {
152 m = efi_1t1_page();
153 mphys = VM_PAGE_TO_PHYS(m);
154 *pdpe = mphys | X86_PG_RW | X86_PG_V;
155 } else {
156 mphys = *pdpe & PG_FRAME;
157 }
158
159 pde = (pd_entry_t *)PHYS_TO_DMAP(mphys);
160 pd_idx = pmap_pde_index(va);
161 pde += pd_idx;
162 if (*pde == 0) {
163 m = efi_1t1_page();
164 mphys = VM_PAGE_TO_PHYS(m);
165 *pde = mphys | X86_PG_RW | X86_PG_V;
166 } else {
167 mphys = *pde & PG_FRAME;
168 }
169
170 pte = (pt_entry_t *)PHYS_TO_DMAP(mphys);
171 pte += pmap_pte_index(va);
172 KASSERT(*pte == 0, ("va %#jx *pt %#jx", va, *pte));
173
174 return (pte);
175 }
176
177 bool
efi_create_1t1_map(struct efi_md * map,int ndesc,int descsz)178 efi_create_1t1_map(struct efi_md *map, int ndesc, int descsz)
179 {
180 struct efi_md *p;
181 pt_entry_t *pte;
182 void *pml;
183 vm_offset_t va;
184 uint64_t idx;
185 int bits, i, mode;
186
187 obj_1t1_pt = vm_pager_allocate(OBJT_PHYS, NULL, ptoa(1 +
188 NPML4EPG + NPML4EPG * NPDPEPG + NPML4EPG * NPDPEPG * NPDEPG),
189 VM_PROT_ALL, 0, NULL);
190 efi_1t1_idx = 0;
191 VM_OBJECT_WLOCK(obj_1t1_pt);
192 efi_pmltop_page = efi_1t1_page();
193 VM_OBJECT_WUNLOCK(obj_1t1_pt);
194 pml = (void *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(efi_pmltop_page));
195 if (la57) {
196 efi_pml5 = pml;
197 pmap_pinit_pml5(efi_pmltop_page);
198 } else {
199 efi_pml4 = pml;
200 pmap_pinit_pml4(efi_pmltop_page);
201 }
202
203 for (i = 0, p = map; i < ndesc; i++, p = efi_next_descriptor(p,
204 descsz)) {
205 if ((p->md_attr & EFI_MD_ATTR_RT) == 0)
206 continue;
207 if (p->md_virt != NULL && (uint64_t)p->md_virt != p->md_phys) {
208 if (bootverbose)
209 printf("EFI Runtime entry %d is mapped\n", i);
210 goto fail;
211 }
212 if ((p->md_phys & EFI_PAGE_MASK) != 0) {
213 if (bootverbose)
214 printf("EFI Runtime entry %d is not aligned\n",
215 i);
216 goto fail;
217 }
218 if (p->md_phys + p->md_pages * EFI_PAGE_SIZE < p->md_phys ||
219 p->md_phys + p->md_pages * EFI_PAGE_SIZE >=
220 VM_MAXUSER_ADDRESS) {
221 printf("EFI Runtime entry %d is not in mappable for RT:"
222 "base %#016jx %#jx pages\n",
223 i, (uintmax_t)p->md_phys,
224 (uintmax_t)p->md_pages);
225 goto fail;
226 }
227 if ((p->md_attr & EFI_MD_ATTR_WB) != 0)
228 mode = VM_MEMATTR_WRITE_BACK;
229 else if ((p->md_attr & EFI_MD_ATTR_WT) != 0)
230 mode = VM_MEMATTR_WRITE_THROUGH;
231 else if ((p->md_attr & EFI_MD_ATTR_WC) != 0)
232 mode = VM_MEMATTR_WRITE_COMBINING;
233 else if ((p->md_attr & EFI_MD_ATTR_WP) != 0)
234 mode = VM_MEMATTR_WRITE_PROTECTED;
235 else if ((p->md_attr & EFI_MD_ATTR_UC) != 0)
236 mode = VM_MEMATTR_UNCACHEABLE;
237 else {
238 if (bootverbose)
239 printf("EFI Runtime entry %d mapping "
240 "attributes unsupported\n", i);
241 mode = VM_MEMATTR_UNCACHEABLE;
242 }
243 bits = pmap_cache_bits(kernel_pmap, mode, FALSE) | X86_PG_RW |
244 X86_PG_V;
245 VM_OBJECT_WLOCK(obj_1t1_pt);
246 for (va = p->md_phys, idx = 0; idx < p->md_pages; idx++,
247 va += PAGE_SIZE) {
248 pte = efi_1t1_pte(va);
249 pte_store(pte, va | bits);
250 }
251 VM_OBJECT_WUNLOCK(obj_1t1_pt);
252 }
253
254 return (true);
255
256 fail:
257 efi_destroy_1t1_map();
258 return (false);
259 }
260
261 /*
262 * Create an environment for the EFI runtime code call. The most
263 * important part is creating the required 1:1 physical->virtual
264 * mappings for the runtime segments. To do that, we manually create
265 * page table which unmap userspace but gives correct kernel mapping.
266 * The 1:1 mappings for runtime segments usually occupy low 4G of the
267 * physical address map.
268 *
269 * The 1:1 mappings were chosen over the SetVirtualAddressMap() EFI RT
270 * service, because there are some BIOSes which fail to correctly
271 * relocate itself on the call, requiring both 1:1 and virtual
272 * mapping. As result, we must provide 1:1 mapping anyway, so no
273 * reason to bother with the virtual map, and no need to add a
274 * complexity into loader.
275 *
276 * The fpu_kern_enter() call allows firmware to use FPU, as mandated
277 * by the specification. In particular, CR0.TS bit is cleared. Also
278 * it enters critical section, giving us neccessary protection against
279 * context switch.
280 *
281 * There is no need to disable interrupts around the change of %cr3,
282 * the kernel mappings are correct, while we only grabbed the
283 * userspace portion of VA. Interrupts handlers must not access
284 * userspace. Having interrupts enabled fixes the issue with
285 * firmware/SMM long operation, which would negatively affect IPIs,
286 * esp. TLB shootdown requests.
287 */
288 int
efi_arch_enter(void)289 efi_arch_enter(void)
290 {
291 pmap_t curpmap;
292
293 curpmap = PCPU_GET(curpmap);
294 PMAP_LOCK_ASSERT(curpmap, MA_OWNED);
295 curthread->td_md.md_efirt_dis_pf = vm_fault_disable_pagefaults();
296
297 /*
298 * IPI TLB shootdown handler invltlb_pcid_handler() reloads
299 * %cr3 from the curpmap->pm_cr3, which would disable runtime
300 * segments mappings. Block the handler's action by setting
301 * curpmap to impossible value. See also comment in
302 * pmap.c:pmap_activate_sw().
303 */
304 if (pmap_pcid_enabled && !invpcid_works)
305 PCPU_SET(curpmap, NULL);
306
307 load_cr3(VM_PAGE_TO_PHYS(efi_pmltop_page) | (pmap_pcid_enabled ?
308 curpmap->pm_pcids[PCPU_GET(cpuid)].pm_pcid : 0));
309 /*
310 * If PCID is enabled, the clear CR3_PCID_SAVE bit in the loaded %cr3
311 * causes TLB invalidation.
312 */
313 if (!pmap_pcid_enabled)
314 invltlb();
315 return (0);
316 }
317
318 void
efi_arch_leave(void)319 efi_arch_leave(void)
320 {
321 pmap_t curpmap;
322
323 curpmap = &curproc->p_vmspace->vm_pmap;
324 if (pmap_pcid_enabled && !invpcid_works)
325 PCPU_SET(curpmap, curpmap);
326 load_cr3(curpmap->pm_cr3 | (pmap_pcid_enabled ?
327 curpmap->pm_pcids[PCPU_GET(cpuid)].pm_pcid : 0));
328 if (!pmap_pcid_enabled)
329 invltlb();
330 vm_fault_enable_pagefaults(curthread->td_md.md_efirt_dis_pf);
331 }
332
333 /* XXX debug stuff */
334 static int
efi_time_sysctl_handler(SYSCTL_HANDLER_ARGS)335 efi_time_sysctl_handler(SYSCTL_HANDLER_ARGS)
336 {
337 struct efi_tm tm;
338 int error, val;
339
340 val = 0;
341 error = sysctl_handle_int(oidp, &val, 0, req);
342 if (error != 0 || req->newptr == NULL)
343 return (error);
344 error = efi_get_time(&tm);
345 if (error == 0) {
346 uprintf("EFI reports: Year %d Month %d Day %d Hour %d Min %d "
347 "Sec %d\n", tm.tm_year, tm.tm_mon, tm.tm_mday, tm.tm_hour,
348 tm.tm_min, tm.tm_sec);
349 }
350 return (error);
351 }
352
353 SYSCTL_PROC(_debug, OID_AUTO, efi_time,
354 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 0,
355 efi_time_sysctl_handler, "I",
356 "");
357