xref: /netbsd-src/sys/arch/xen/x86/xen_pmap.c (revision 77901f27b5c1718b6fffc02836239da3b0093093)
1 /*	$NetBSD: xen_pmap.c,v 1.41 2023/08/25 08:05:19 riastradh Exp $	*/
2 
3 /*
4  * Copyright (c) 2007 Manuel Bouyer.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  */
26 
27 /*
28  * Copyright (c) 2006 Mathieu Ropert <mro@adviseo.fr>
29  *
30  * Permission to use, copy, modify, and distribute this software for any
31  * purpose with or without fee is hereby granted, provided that the above
32  * copyright notice and this permission notice appear in all copies.
33  *
34  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
35  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
36  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
37  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
38  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
39  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
40  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
41  */
42 
43 /*
44  * Copyright (c) 1997 Charles D. Cranor and Washington University.
45  * All rights reserved.
46  *
47  * Redistribution and use in source and binary forms, with or without
48  * modification, are permitted provided that the following conditions
49  * are met:
50  * 1. Redistributions of source code must retain the above copyright
51  *    notice, this list of conditions and the following disclaimer.
52  * 2. Redistributions in binary form must reproduce the above copyright
53  *    notice, this list of conditions and the following disclaimer in the
54  *    documentation and/or other materials provided with the distribution.
55  *
56  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
57  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
58  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
59  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
60  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
61  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
62  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
63  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
64  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
65  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
66  */
67 
68 /*
69  * Copyright 2001 (c) Wasabi Systems, Inc.
70  * All rights reserved.
71  *
72  * Written by Frank van der Linden for Wasabi Systems, Inc.
73  *
74  * Redistribution and use in source and binary forms, with or without
75  * modification, are permitted provided that the following conditions
76  * are met:
77  * 1. Redistributions of source code must retain the above copyright
78  *    notice, this list of conditions and the following disclaimer.
79  * 2. Redistributions in binary form must reproduce the above copyright
80  *    notice, this list of conditions and the following disclaimer in the
81  *    documentation and/or other materials provided with the distribution.
82  * 3. All advertising materials mentioning features or use of this software
83  *    must display the following acknowledgement:
84  *      This product includes software developed for the NetBSD Project by
85  *      Wasabi Systems, Inc.
86  * 4. The name of Wasabi Systems, Inc. may not be used to endorse
87  *    or promote products derived from this software without specific prior
88  *    written permission.
89  *
90  * THIS SOFTWARE IS PROVIDED BY WASABI SYSTEMS, INC. ``AS IS'' AND
91  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
92  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
93  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL WASABI SYSTEMS, INC
94  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
95  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
96  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
97  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
98  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
99  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
100  * POSSIBILITY OF SUCH DAMAGE.
101  */
102 
103 #include <sys/cdefs.h>
104 __KERNEL_RCSID(0, "$NetBSD: xen_pmap.c,v 1.41 2023/08/25 08:05:19 riastradh Exp $");
105 
106 #include "opt_lockdebug.h"
107 #include "opt_multiprocessor.h"
108 #include "opt_xen.h"
109 
110 #include <sys/param.h>
111 #include <sys/systm.h>
112 #include <sys/proc.h>
113 #include <sys/pool.h>
114 #include <sys/kernel.h>
115 #include <sys/atomic.h>
116 #include <sys/cpu.h>
117 #include <sys/intr.h>
118 
119 #include <uvm/uvm.h>
120 
121 #include <dev/isa/isareg.h>
122 
123 #include <machine/specialreg.h>
124 #include <machine/gdt.h>
125 #include <machine/isa_machdep.h>
126 #include <machine/cpuvar.h>
127 #include <machine/pmap_private.h>
128 
129 #include <x86/machdep.h>
130 #include <x86/pmap_pv.h>
131 
132 #include <x86/i82489reg.h>
133 #include <x86/i82489var.h>
134 
135 #include <xen/include/public/xen.h>
136 #include <xen/hypervisor.h>
137 #include <xen/xenpmap.h>
138 
139 #define COUNT(x)	/* nothing */
140 
141 extern pd_entry_t * const normal_pdes[];
142 
143 extern paddr_t pmap_pa_start; /* PA of first physical page for this domain */
144 extern paddr_t pmap_pa_end;   /* PA of last physical page for this domain */
145 
146 int
pmap_enter(struct pmap * pmap,vaddr_t va,paddr_t pa,vm_prot_t prot,u_int flags)147 pmap_enter(struct pmap *pmap, vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
148 {
149 	paddr_t ma;
150 
151 	if (__predict_false(pa < pmap_pa_start || pmap_pa_end <= pa)) {
152 		ma = pa; /* XXX hack */
153 	} else {
154 		ma = xpmap_ptom(pa);
155 	}
156 
157 	return pmap_enter_ma(pmap, va, ma, pa, prot, flags, DOMID_SELF);
158 }
159 
160 /*
161  * pmap_kenter_ma: enter a kernel mapping without R/M (pv_entry) tracking
162  *
163  * => no need to lock anything, assume va is already allocated
164  * => should be faster than normal pmap enter function
165  * => we expect a MACHINE address
166  */
167 
168 void
pmap_kenter_ma(vaddr_t va,paddr_t ma,vm_prot_t prot,u_int flags)169 pmap_kenter_ma(vaddr_t va, paddr_t ma, vm_prot_t prot, u_int flags)
170 {
171 	pt_entry_t *pte, opte, npte;
172 
173 	if (va < VM_MIN_KERNEL_ADDRESS)
174 		pte = vtopte(va);
175 	else
176 		pte = kvtopte(va);
177 
178 	npte = ma | ((prot & VM_PROT_WRITE) ? PTE_W : 0) | PTE_P;
179 	if (flags & PMAP_NOCACHE)
180 		npte |= PTE_PCD;
181 
182 	if ((cpu_feature[2] & CPUID_NOX) && !(prot & VM_PROT_EXECUTE))
183 		npte |= PTE_NX;
184 
185 	opte = pmap_pte_testset(pte, npte); /* zap! */
186 
187 	if (pmap_valid_entry(opte)) {
188 #if defined(MULTIPROCESSOR)
189 		if (__predict_false(x86_mp_online == false)) {
190 			pmap_update_pg(va);
191 		} else {
192 			kpreempt_disable();
193 			pmap_tlb_shootdown(pmap_kernel(), va, opte,
194 			    TLBSHOOT_KENTER);
195 			kpreempt_enable();
196 		}
197 #else
198 		/* Don't bother deferring in the single CPU case. */
199 		pmap_update_pg(va);
200 #endif
201 	}
202 }
203 
204 /*
205  * pmap_extract_ma: extract a MA for the given VA
206  */
207 
208 bool
pmap_extract_ma(struct pmap * pmap,vaddr_t va,paddr_t * pap)209 pmap_extract_ma(struct pmap *pmap, vaddr_t va, paddr_t *pap)
210 {
211 	pt_entry_t *ptes, pte;
212 	pd_entry_t pde;
213 	pd_entry_t * const *pdes;
214 	struct pmap *pmap2;
215 	int lvl;
216 
217 	if (pmap != pmap_kernel()) {
218 		mutex_enter(&pmap->pm_lock);
219 	}
220 	pmap_map_ptes(pmap, &pmap2, &ptes, &pdes);
221 	if (!pmap_pdes_valid(va, pdes, &pde, &lvl)) {
222 		pmap_unmap_ptes(pmap, pmap2);
223 		if (pmap != pmap_kernel()) {
224 			mutex_exit(&pmap->pm_lock);
225 		}
226 		return false;
227 	}
228 
229 	KASSERT(lvl == 1);
230 	pte = ptes[pl1_i(va)];
231 	pmap_unmap_ptes(pmap, pmap2);
232 	if (pmap != pmap_kernel()) {
233 		mutex_exit(&pmap->pm_lock);
234 	}
235 
236 	if (__predict_true((pte & PTE_P) != 0)) {
237 		if (pap != NULL)
238 			*pap = (pte & PTE_4KFRAME) | (va & (NBPD_L1 - 1));
239 		return true;
240 	}
241 
242 	return false;
243 }
244 
245 /*
246  * Xen pmap's handlers for save/restore
247  */
248 void
pmap_xen_suspend(void)249 pmap_xen_suspend(void)
250 {
251 	pmap_unmap_recursive_entries();
252 
253 	xpq_flush_queue();
254 }
255 
256 void
pmap_xen_resume(void)257 pmap_xen_resume(void)
258 {
259 	pmap_map_recursive_entries();
260 
261 	xpq_flush_queue();
262 }
263 
264 /*
265  * NetBSD uses L2 shadow pages to support PAE with Xen. However, Xen does not
266  * handle them correctly during save/restore, leading to incorrect page
267  * tracking and pinning during restore.
268  * For save/restore to succeed, two functions are introduced:
269  * - pmap_map_recursive_entries(), used by resume code to set the recursive
270  *   mapping entries to their correct value
271  * - pmap_unmap_recursive_entries(), used by suspend code to clear all
272  *   PDIR_SLOT_PTE entries
273  */
274 void
pmap_map_recursive_entries(void)275 pmap_map_recursive_entries(void)
276 {
277 	int i;
278 	struct pmap *pm;
279 
280 	mutex_enter(&pmaps_lock);
281 	LIST_FOREACH(pm, &pmaps, pm_list) {
282 		for (i = 0; i < PDP_SIZE; i++) {
283 			xpq_queue_pte_update(
284 			    xpmap_ptom(pmap_pdirpa(pm, PDIR_SLOT_PTE + i)),
285 			    xpmap_ptom((pm)->pm_pdirpa[i]) | PTE_P);
286 		}
287 	}
288 	mutex_exit(&pmaps_lock);
289 
290 	for (i = 0; i < PDP_SIZE; i++) {
291 		xpq_queue_pte_update(
292 		    xpmap_ptom(pmap_pdirpa(pmap_kernel(), PDIR_SLOT_PTE + i)),
293 		    xpmap_ptom(pmap_kernel()->pm_pdirpa[i]) | PTE_P);
294 	}
295 }
296 
297 /*
298  * Unmap recursive entries found in pmaps. Required during Xen
299  * save/restore operations, as Xen does not handle recursive mappings
300  * properly.
301  */
302 void
pmap_unmap_recursive_entries(void)303 pmap_unmap_recursive_entries(void)
304 {
305 	int i;
306 	struct pmap *pm;
307 
308 	/*
309 	 * Invalidate pmap_pdp_cache as it contains L2-pinned objects with
310 	 * recursive entries.
311 	 * XXX jym@ : find a way to drain per-CPU caches to. pool_cache_inv
312 	 * does not do that.
313 	 */
314 	pool_cache_invalidate(&pmap_cache);
315 
316 	mutex_enter(&pmaps_lock);
317 	LIST_FOREACH(pm, &pmaps, pm_list) {
318 		for (i = 0; i < PDP_SIZE; i++) {
319 			xpq_queue_pte_update(
320 			    xpmap_ptom(pmap_pdirpa(pm, PDIR_SLOT_PTE + i)), 0);
321 		}
322 	}
323 	mutex_exit(&pmaps_lock);
324 
325 	/* do it for pmap_kernel() too! */
326 	for (i = 0; i < PDP_SIZE; i++) {
327 		xpq_queue_pte_update(
328 		    xpmap_ptom(pmap_pdirpa(pmap_kernel(), PDIR_SLOT_PTE + i)),
329 		    0);
330 	}
331 }
332 
333 static __inline void
pmap_kpm_setpte(struct cpu_info * ci,struct pmap * pmap,int index)334 pmap_kpm_setpte(struct cpu_info *ci, struct pmap *pmap, int index)
335 {
336 	KASSERT(mutex_owned(&pmap->pm_lock));
337 	KASSERT(mutex_owned(&ci->ci_kpm_mtx));
338 	if (pmap == pmap_kernel()) {
339 		KASSERT(index >= PDIR_SLOT_KERN);
340 	}
341 
342 #ifdef __x86_64__
343 	xpq_queue_pte_update(
344 	    xpmap_ptetomach(&ci->ci_kpm_pdir[index]),
345 	    pmap->pm_pdir[index]);
346 #else
347 	xpq_queue_pte_update(
348 	    xpmap_ptetomach(&ci->ci_kpm_pdir[l2tol2(index)]),
349 	    pmap->pm_pdir[index]);
350 #endif
351 
352 	xpq_flush_queue();
353 }
354 
355 /*
356  * Synchronise shadow pdir with the pmap on all cpus on which it is
357  * loaded.
358  */
359 void
xen_kpm_sync(struct pmap * pmap,int index)360 xen_kpm_sync(struct pmap *pmap, int index)
361 {
362 	CPU_INFO_ITERATOR cii;
363 	struct cpu_info *ci;
364 
365 	KASSERT(pmap != NULL);
366 	KASSERT(kpreempt_disabled());
367 
368 	pmap_pte_flush();
369 
370 	for (CPU_INFO_FOREACH(cii, ci)) {
371 		if (ci == NULL) {
372 			continue;
373 		}
374 		cpuid_t cid = cpu_index(ci);
375 		if (pmap != pmap_kernel() &&
376 		    !kcpuset_isset(pmap->pm_xen_ptp_cpus, cid))
377 			continue;
378 
379 		/* take the lock and check again */
380 		mutex_enter(&ci->ci_kpm_mtx);
381 		if (pmap == pmap_kernel() ||
382 		    kcpuset_isset(pmap->pm_xen_ptp_cpus, cid)) {
383 			pmap_kpm_setpte(ci, pmap, index);
384 		}
385 		mutex_exit(&ci->ci_kpm_mtx);
386 	}
387 }
388 
389 int
x86_select_freelist(uint64_t maxaddr)390 x86_select_freelist(uint64_t maxaddr)
391 {
392 	return VM_FREELIST_DEFAULT;
393 }
394