1 /* $NetBSD: pmap.c,v 1.121 2023/12/15 09:42:33 rin Exp $ */
2 /*-
3 * Copyright (c) 2001 The NetBSD Foundation, Inc.
4 * All rights reserved.
5 *
6 * This code is derived from software contributed to The NetBSD Foundation
7 * by Matt Thomas <matt@3am-software.com> of Allegro Networks, Inc.
8 *
9 * Support for PPC64 Bridge mode added by Sanjay Lal <sanjayl@kymasys.com>
10 * of Kyma Systems LLC.
11 *
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
14 * are met:
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
23 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
24 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
25 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
28 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
29 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
30 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31 * POSSIBILITY OF SUCH DAMAGE.
32 */
33
34 /*
35 * Copyright (C) 1995, 1996 Wolfgang Solfrank.
36 * Copyright (C) 1995, 1996 TooLs GmbH.
37 * All rights reserved.
38 *
39 * Redistribution and use in source and binary forms, with or without
40 * modification, are permitted provided that the following conditions
41 * are met:
42 * 1. Redistributions of source code must retain the above copyright
43 * notice, this list of conditions and the following disclaimer.
44 * 2. Redistributions in binary form must reproduce the above copyright
45 * notice, this list of conditions and the following disclaimer in the
46 * documentation and/or other materials provided with the distribution.
47 * 3. All advertising materials mentioning features or use of this software
48 * must display the following acknowledgement:
49 * This product includes software developed by TooLs GmbH.
50 * 4. The name of TooLs GmbH may not be used to endorse or promote products
51 * derived from this software without specific prior written permission.
52 *
53 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR
54 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
55 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
56 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
57 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
58 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
59 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
60 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
61 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
62 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
63 */
64
65 #include <sys/cdefs.h>
66 __KERNEL_RCSID(0, "$NetBSD: pmap.c,v 1.121 2023/12/15 09:42:33 rin Exp $");
67
68 #define PMAP_NOOPNAMES
69
70 #ifdef _KERNEL_OPT
71 #include "opt_altivec.h"
72 #include "opt_multiprocessor.h"
73 #include "opt_pmap.h"
74 #include "opt_ppcarch.h"
75 #endif
76
77 #include <sys/param.h>
78 #include <sys/proc.h>
79 #include <sys/pool.h>
80 #include <sys/queue.h>
81 #include <sys/device.h> /* for evcnt */
82 #include <sys/systm.h>
83 #include <sys/atomic.h>
84
85 #include <uvm/uvm.h>
86 #include <uvm/uvm_physseg.h>
87
88 #include <machine/powerpc.h>
89 #include <powerpc/bat.h>
90 #include <powerpc/pcb.h>
91 #include <powerpc/psl.h>
92 #include <powerpc/spr.h>
93 #include <powerpc/oea/spr.h>
94 #include <powerpc/oea/sr_601.h>
95
96 #ifdef ALTIVEC
97 extern int pmap_use_altivec;
98 #endif
99
100 #ifdef PMAP_MEMLIMIT
101 static paddr_t pmap_memlimit = PMAP_MEMLIMIT;
102 #else
103 static paddr_t pmap_memlimit = -PAGE_SIZE; /* there is no limit */
104 #endif
105
106 extern struct pmap kernel_pmap_;
107 static unsigned int pmap_pages_stolen;
108 static u_long pmap_pte_valid;
109 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
110 static u_long pmap_pvo_enter_depth;
111 static u_long pmap_pvo_remove_depth;
112 #endif
113
114 #ifndef MSGBUFADDR
115 extern paddr_t msgbuf_paddr;
116 #endif
117
118 static struct mem_region *mem, *avail;
119 static u_int mem_cnt, avail_cnt;
120
121 #if !defined(PMAP_OEA64) && !defined(PMAP_OEA64_BRIDGE)
122 # define PMAP_OEA 1
123 #endif
124
125 #if defined(PMAP_OEA)
126 #define _PRIxpte "lx"
127 #else
128 #define _PRIxpte PRIx64
129 #endif
130 #define _PRIxpa "lx"
131 #define _PRIxva "lx"
132 #define _PRIsr "lx"
133
134 #ifdef PMAP_NEEDS_FIXUP
135 #if defined(PMAP_OEA)
136 #define PMAPNAME(name) pmap32_##name
137 #elif defined(PMAP_OEA64)
138 #define PMAPNAME(name) pmap64_##name
139 #elif defined(PMAP_OEA64_BRIDGE)
140 #define PMAPNAME(name) pmap64bridge_##name
141 #else
142 #error unknown variant for pmap
143 #endif
144 #endif /* PMAP_NEEDS_FIXUP */
145
146 #ifdef PMAPNAME
147 #define STATIC static
148 #define pmap_pte_spill PMAPNAME(pte_spill)
149 #define pmap_real_memory PMAPNAME(real_memory)
150 #define pmap_init PMAPNAME(init)
151 #define pmap_virtual_space PMAPNAME(virtual_space)
152 #define pmap_create PMAPNAME(create)
153 #define pmap_reference PMAPNAME(reference)
154 #define pmap_destroy PMAPNAME(destroy)
155 #define pmap_copy PMAPNAME(copy)
156 #define pmap_update PMAPNAME(update)
157 #define pmap_enter PMAPNAME(enter)
158 #define pmap_remove PMAPNAME(remove)
159 #define pmap_kenter_pa PMAPNAME(kenter_pa)
160 #define pmap_kremove PMAPNAME(kremove)
161 #define pmap_extract PMAPNAME(extract)
162 #define pmap_protect PMAPNAME(protect)
163 #define pmap_unwire PMAPNAME(unwire)
164 #define pmap_page_protect PMAPNAME(page_protect)
165 #define pmap_pv_protect PMAPNAME(pv_protect)
166 #define pmap_query_bit PMAPNAME(query_bit)
167 #define pmap_clear_bit PMAPNAME(clear_bit)
168
169 #define pmap_activate PMAPNAME(activate)
170 #define pmap_deactivate PMAPNAME(deactivate)
171
172 #define pmap_pinit PMAPNAME(pinit)
173 #define pmap_procwr PMAPNAME(procwr)
174
175 #define pmap_pool PMAPNAME(pool)
176 #define pmap_pvo_pool PMAPNAME(pvo_pool)
177 #define pmap_pvo_table PMAPNAME(pvo_table)
178 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
179 #define pmap_pte_print PMAPNAME(pte_print)
180 #define pmap_pteg_check PMAPNAME(pteg_check)
181 #define pmap_print_mmruregs PMAPNAME(print_mmuregs)
182 #define pmap_print_pte PMAPNAME(print_pte)
183 #define pmap_pteg_dist PMAPNAME(pteg_dist)
184 #endif
185 #if defined(DEBUG) || defined(PMAPCHECK)
186 #define pmap_pvo_verify PMAPNAME(pvo_verify)
187 #define pmapcheck PMAPNAME(check)
188 #endif
189 #if defined(DEBUG) || defined(PMAPDEBUG)
190 #define pmapdebug PMAPNAME(debug)
191 #endif
192 #define pmap_steal_memory PMAPNAME(steal_memory)
193 #define pmap_bootstrap PMAPNAME(bootstrap)
194 #define pmap_bootstrap1 PMAPNAME(bootstrap1)
195 #define pmap_bootstrap2 PMAPNAME(bootstrap2)
196 #else
197 #define STATIC /* nothing */
198 #endif /* PMAPNAME */
199
200 STATIC int pmap_pte_spill(struct pmap *, vaddr_t, bool);
201 STATIC void pmap_real_memory(paddr_t *, psize_t *);
202 STATIC void pmap_init(void);
203 STATIC void pmap_virtual_space(vaddr_t *, vaddr_t *);
204 STATIC pmap_t pmap_create(void);
205 STATIC void pmap_reference(pmap_t);
206 STATIC void pmap_destroy(pmap_t);
207 STATIC void pmap_copy(pmap_t, pmap_t, vaddr_t, vsize_t, vaddr_t);
208 STATIC void pmap_update(pmap_t);
209 STATIC int pmap_enter(pmap_t, vaddr_t, paddr_t, vm_prot_t, u_int);
210 STATIC void pmap_remove(pmap_t, vaddr_t, vaddr_t);
211 STATIC void pmap_kenter_pa(vaddr_t, paddr_t, vm_prot_t, u_int);
212 STATIC void pmap_kremove(vaddr_t, vsize_t);
213 STATIC bool pmap_extract(pmap_t, vaddr_t, paddr_t *);
214
215 STATIC void pmap_protect(pmap_t, vaddr_t, vaddr_t, vm_prot_t);
216 STATIC void pmap_unwire(pmap_t, vaddr_t);
217 STATIC void pmap_page_protect(struct vm_page *, vm_prot_t);
218 STATIC void pmap_pv_protect(paddr_t, vm_prot_t);
219 STATIC bool pmap_query_bit(struct vm_page *, int);
220 STATIC bool pmap_clear_bit(struct vm_page *, int);
221
222 STATIC void pmap_activate(struct lwp *);
223 STATIC void pmap_deactivate(struct lwp *);
224
225 STATIC void pmap_pinit(pmap_t pm);
226 STATIC void pmap_procwr(struct proc *, vaddr_t, size_t);
227
228 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
229 STATIC void pmap_pte_print(volatile struct pte *);
230 STATIC void pmap_pteg_check(void);
231 STATIC void pmap_print_mmuregs(void);
232 STATIC void pmap_print_pte(pmap_t, vaddr_t);
233 STATIC void pmap_pteg_dist(void);
234 #endif
235 #if defined(DEBUG) || defined(PMAPCHECK)
236 STATIC void pmap_pvo_verify(void);
237 #endif
238 STATIC vaddr_t pmap_steal_memory(vsize_t, vaddr_t *, vaddr_t *);
239 STATIC void pmap_bootstrap(paddr_t, paddr_t);
240 STATIC void pmap_bootstrap1(paddr_t, paddr_t);
241 STATIC void pmap_bootstrap2(void);
242
243 #ifdef PMAPNAME
244 const struct pmap_ops PMAPNAME(ops) = {
245 .pmapop_pte_spill = pmap_pte_spill,
246 .pmapop_real_memory = pmap_real_memory,
247 .pmapop_init = pmap_init,
248 .pmapop_virtual_space = pmap_virtual_space,
249 .pmapop_create = pmap_create,
250 .pmapop_reference = pmap_reference,
251 .pmapop_destroy = pmap_destroy,
252 .pmapop_copy = pmap_copy,
253 .pmapop_update = pmap_update,
254 .pmapop_enter = pmap_enter,
255 .pmapop_remove = pmap_remove,
256 .pmapop_kenter_pa = pmap_kenter_pa,
257 .pmapop_kremove = pmap_kremove,
258 .pmapop_extract = pmap_extract,
259 .pmapop_protect = pmap_protect,
260 .pmapop_unwire = pmap_unwire,
261 .pmapop_page_protect = pmap_page_protect,
262 .pmapop_pv_protect = pmap_pv_protect,
263 .pmapop_query_bit = pmap_query_bit,
264 .pmapop_clear_bit = pmap_clear_bit,
265 .pmapop_activate = pmap_activate,
266 .pmapop_deactivate = pmap_deactivate,
267 .pmapop_pinit = pmap_pinit,
268 .pmapop_procwr = pmap_procwr,
269 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
270 .pmapop_pte_print = pmap_pte_print,
271 .pmapop_pteg_check = pmap_pteg_check,
272 .pmapop_print_mmuregs = pmap_print_mmuregs,
273 .pmapop_print_pte = pmap_print_pte,
274 .pmapop_pteg_dist = pmap_pteg_dist,
275 #else
276 .pmapop_pte_print = NULL,
277 .pmapop_pteg_check = NULL,
278 .pmapop_print_mmuregs = NULL,
279 .pmapop_print_pte = NULL,
280 .pmapop_pteg_dist = NULL,
281 #endif
282 #if defined(DEBUG) || defined(PMAPCHECK)
283 .pmapop_pvo_verify = pmap_pvo_verify,
284 #else
285 .pmapop_pvo_verify = NULL,
286 #endif
287 .pmapop_steal_memory = pmap_steal_memory,
288 .pmapop_bootstrap = pmap_bootstrap,
289 .pmapop_bootstrap1 = pmap_bootstrap1,
290 .pmapop_bootstrap2 = pmap_bootstrap2,
291 };
292 #endif /* !PMAPNAME */
293
294 /*
295 * The following structure is aligned to 32 bytes, if reasonably possible.
296 */
297 struct pvo_entry {
298 LIST_ENTRY(pvo_entry) pvo_vlink; /* Link to common virt page */
299 TAILQ_ENTRY(pvo_entry) pvo_olink; /* Link to overflow entry */
300 struct pte pvo_pte; /* Prebuilt PTE */
301 pmap_t pvo_pmap; /* ptr to owning pmap */
302 vaddr_t pvo_vaddr; /* VA of entry */
303 #define PVO_PTEGIDX_MASK 0x0007 /* which PTEG slot */
304 #define PVO_PTEGIDX_VALID 0x0008 /* slot is valid */
305 #define PVO_WIRED 0x0010 /* PVO entry is wired */
306 #define PVO_MANAGED 0x0020 /* PVO e. for managed page */
307 #define PVO_EXECUTABLE 0x0040 /* PVO e. for executable page */
308 #define PVO_WIRED_P(pvo) ((pvo)->pvo_vaddr & PVO_WIRED)
309 #define PVO_MANAGED_P(pvo) ((pvo)->pvo_vaddr & PVO_MANAGED)
310 #define PVO_EXECUTABLE_P(pvo) ((pvo)->pvo_vaddr & PVO_EXECUTABLE)
311 #define PVO_ENTER_INSERT 0 /* PVO has been removed */
312 #define PVO_SPILL_UNSET 1 /* PVO has been evicted */
313 #define PVO_SPILL_SET 2 /* PVO has been spilled */
314 #define PVO_SPILL_INSERT 3 /* PVO has been inserted */
315 #define PVO_PMAP_PAGE_PROTECT 4 /* PVO has changed */
316 #define PVO_PMAP_PROTECT 5 /* PVO has changed */
317 #define PVO_REMOVE 6 /* PVO has been removed */
318 #define PVO_WHERE_MASK 15
319 #define PVO_WHERE_SHFT 8
320 };
321
322 #if defined(PMAP_OEA) && !defined(DIAGNOSTIC)
323 #define PMAP_PVO_ENTRY_ALIGN 32
324 #else
325 #define PMAP_PVO_ENTRY_ALIGN __alignof(struct pvo_entry)
326 #endif
327
328 #define PVO_VADDR(pvo) ((pvo)->pvo_vaddr & ~ADDR_POFF)
329 #define PVO_PTEGIDX_GET(pvo) ((pvo)->pvo_vaddr & PVO_PTEGIDX_MASK)
330 #define PVO_PTEGIDX_ISSET(pvo) ((pvo)->pvo_vaddr & PVO_PTEGIDX_VALID)
331 #define PVO_PTEGIDX_CLR(pvo) \
332 ((void)((pvo)->pvo_vaddr &= ~(PVO_PTEGIDX_VALID|PVO_PTEGIDX_MASK)))
333 #define PVO_PTEGIDX_SET(pvo,i) \
334 ((void)((pvo)->pvo_vaddr |= (i)|PVO_PTEGIDX_VALID))
335 #define PVO_WHERE(pvo,w) \
336 ((pvo)->pvo_vaddr &= ~(PVO_WHERE_MASK << PVO_WHERE_SHFT), \
337 (pvo)->pvo_vaddr |= ((PVO_ ## w) << PVO_WHERE_SHFT))
338
339 TAILQ_HEAD(pvo_tqhead, pvo_entry);
340 struct pvo_tqhead *pmap_pvo_table; /* pvo entries by ptegroup index */
341
342 struct pool pmap_pool; /* pool for pmap structures */
343 struct pool pmap_pvo_pool; /* pool for pvo entries */
344
345 static void *pmap_pool_alloc(struct pool *, int);
346 static void pmap_pool_free(struct pool *, void *);
347
348 static struct pool_allocator pmap_pool_allocator = {
349 .pa_alloc = pmap_pool_alloc,
350 .pa_free = pmap_pool_free,
351 .pa_pagesz = 0,
352 };
353
354 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
355 void pmap_pte_print(volatile struct pte *);
356 void pmap_pteg_check(void);
357 void pmap_pteg_dist(void);
358 void pmap_print_pte(pmap_t, vaddr_t);
359 void pmap_print_mmuregs(void);
360 #endif
361
362 #if defined(DEBUG) || defined(PMAPCHECK)
363 #ifdef PMAPCHECK
364 int pmapcheck = 1;
365 #else
366 int pmapcheck = 0;
367 #endif
368 void pmap_pvo_verify(void);
369 static void pmap_pvo_check(const struct pvo_entry *);
370 #define PMAP_PVO_CHECK(pvo) \
371 do { \
372 if (pmapcheck) \
373 pmap_pvo_check(pvo); \
374 } while (0)
375 #else
376 #define PMAP_PVO_CHECK(pvo) do { } while (/*CONSTCOND*/0)
377 #endif
378 static int pmap_pte_insert(int, struct pte *);
379 static int pmap_pvo_enter(pmap_t, struct pool *, struct pvo_head *,
380 vaddr_t, paddr_t, register_t, int);
381 static void pmap_pvo_remove(struct pvo_entry *, int, struct pvo_head *);
382 static void pmap_pvo_free(struct pvo_entry *);
383 static void pmap_pvo_free_list(struct pvo_head *);
384 static struct pvo_entry *pmap_pvo_find_va(pmap_t, vaddr_t, int *);
385 static volatile struct pte *pmap_pvo_to_pte(const struct pvo_entry *, int);
386 static struct pvo_entry *pmap_pvo_reclaim(void);
387 static void pvo_set_exec(struct pvo_entry *);
388 static void pvo_clear_exec(struct pvo_entry *);
389
390 static void tlbia(void);
391
392 static void pmap_release(pmap_t);
393 static paddr_t pmap_boot_find_memory(psize_t, psize_t, int);
394
395 static uint32_t pmap_pvo_reclaim_nextidx;
396 #ifdef DEBUG
397 static int pmap_pvo_reclaim_debugctr;
398 #endif
399
400 #define VSID_NBPW (sizeof(uint32_t) * 8)
401 static uint32_t pmap_vsid_bitmap[NPMAPS / VSID_NBPW];
402
403 static int pmap_initialized;
404
405 #if defined(DEBUG) || defined(PMAPDEBUG)
406 #define PMAPDEBUG_BOOT 0x0001
407 #define PMAPDEBUG_PTE 0x0002
408 #define PMAPDEBUG_EXEC 0x0008
409 #define PMAPDEBUG_PVOENTER 0x0010
410 #define PMAPDEBUG_PVOREMOVE 0x0020
411 #define PMAPDEBUG_ACTIVATE 0x0100
412 #define PMAPDEBUG_CREATE 0x0200
413 #define PMAPDEBUG_ENTER 0x1000
414 #define PMAPDEBUG_KENTER 0x2000
415 #define PMAPDEBUG_KREMOVE 0x4000
416 #define PMAPDEBUG_REMOVE 0x8000
417
418 unsigned int pmapdebug = 0;
419
420 # define DPRINTF(x, ...) printf(x, __VA_ARGS__)
421 # define DPRINTFN(n, x, ...) do if (pmapdebug & PMAPDEBUG_ ## n) printf(x, __VA_ARGS__); while (0)
422 #else
423 # define DPRINTF(x, ...) do { } while (0)
424 # define DPRINTFN(n, x, ...) do { } while (0)
425 #endif
426
427
428 #ifdef PMAPCOUNTERS
429 /*
430 * From pmap_subr.c
431 */
432 extern struct evcnt pmap_evcnt_mappings;
433 extern struct evcnt pmap_evcnt_unmappings;
434
435 extern struct evcnt pmap_evcnt_kernel_mappings;
436 extern struct evcnt pmap_evcnt_kernel_unmappings;
437
438 extern struct evcnt pmap_evcnt_mappings_replaced;
439
440 extern struct evcnt pmap_evcnt_exec_mappings;
441 extern struct evcnt pmap_evcnt_exec_cached;
442
443 extern struct evcnt pmap_evcnt_exec_synced;
444 extern struct evcnt pmap_evcnt_exec_synced_clear_modify;
445 extern struct evcnt pmap_evcnt_exec_synced_pvo_remove;
446
447 extern struct evcnt pmap_evcnt_exec_uncached_page_protect;
448 extern struct evcnt pmap_evcnt_exec_uncached_clear_modify;
449 extern struct evcnt pmap_evcnt_exec_uncached_zero_page;
450 extern struct evcnt pmap_evcnt_exec_uncached_copy_page;
451 extern struct evcnt pmap_evcnt_exec_uncached_pvo_remove;
452
453 extern struct evcnt pmap_evcnt_updates;
454 extern struct evcnt pmap_evcnt_collects;
455 extern struct evcnt pmap_evcnt_copies;
456
457 extern struct evcnt pmap_evcnt_ptes_spilled;
458 extern struct evcnt pmap_evcnt_ptes_unspilled;
459 extern struct evcnt pmap_evcnt_ptes_evicted;
460
461 extern struct evcnt pmap_evcnt_ptes_primary[8];
462 extern struct evcnt pmap_evcnt_ptes_secondary[8];
463 extern struct evcnt pmap_evcnt_ptes_removed;
464 extern struct evcnt pmap_evcnt_ptes_changed;
465 extern struct evcnt pmap_evcnt_pvos_reclaimed;
466 extern struct evcnt pmap_evcnt_pvos_failed;
467
468 extern struct evcnt pmap_evcnt_zeroed_pages;
469 extern struct evcnt pmap_evcnt_copied_pages;
470 extern struct evcnt pmap_evcnt_idlezeroed_pages;
471
472 #define PMAPCOUNT(ev) ((pmap_evcnt_ ## ev).ev_count++)
473 #define PMAPCOUNT2(ev) ((ev).ev_count++)
474 #else
475 #define PMAPCOUNT(ev) ((void) 0)
476 #define PMAPCOUNT2(ev) ((void) 0)
477 #endif
478
479 #define TLBIE(va) __asm volatile("tlbie %0" :: "r"(va) : "memory")
480
481 /* XXXSL: this needs to be moved to assembler */
482 #define TLBIEL(va) __asm volatile("tlbie %0" :: "r"(va) : "memory")
483
484 #ifdef MD_TLBSYNC
485 #define TLBSYNC() MD_TLBSYNC()
486 #else
487 #define TLBSYNC() __asm volatile("tlbsync" ::: "memory")
488 #endif
489 #define SYNC() __asm volatile("sync" ::: "memory")
490 #define EIEIO() __asm volatile("eieio" ::: "memory")
491 #define DCBST(va) __asm volatile("dcbst 0,%0" :: "r"(va) : "memory")
492 #define MFMSR() mfmsr()
493 #define MTMSR(psl) mtmsr(psl)
494 #define MFPVR() mfpvr()
495 #define MFSRIN(va) mfsrin(va)
496 #define MFTB() mfrtcltbl()
497
498 #if defined(DDB) && !defined(PMAP_OEA64)
499 static inline register_t
mfsrin(vaddr_t va)500 mfsrin(vaddr_t va)
501 {
502 register_t sr;
503 __asm volatile ("mfsrin %0,%1" : "=r"(sr) : "r"(va));
504 return sr;
505 }
506 #endif /* DDB && !PMAP_OEA64 */
507
508 #if defined (PMAP_OEA64_BRIDGE)
509 extern void mfmsr64 (register64_t *result);
510 #endif /* PMAP_OEA64_BRIDGE */
511
512 #define PMAP_LOCK() KERNEL_LOCK(1, NULL)
513 #define PMAP_UNLOCK() KERNEL_UNLOCK_ONE(NULL)
514
515 static inline register_t
pmap_interrupts_off(void)516 pmap_interrupts_off(void)
517 {
518 register_t msr = MFMSR();
519 if (msr & PSL_EE)
520 MTMSR(msr & ~PSL_EE);
521 return msr;
522 }
523
524 static void
pmap_interrupts_restore(register_t msr)525 pmap_interrupts_restore(register_t msr)
526 {
527 if (msr & PSL_EE)
528 MTMSR(msr);
529 }
530
531 static inline u_int32_t
mfrtcltbl(void)532 mfrtcltbl(void)
533 {
534 #ifdef PPC_OEA601
535 if ((MFPVR() >> 16) == MPC601)
536 return (mfrtcl() >> 7);
537 else
538 #endif
539 return (mftbl());
540 }
541
542 /*
543 * These small routines may have to be replaced,
544 * if/when we support processors other that the 604.
545 */
546
547 void
tlbia(void)548 tlbia(void)
549 {
550 char *i;
551
552 SYNC();
553 #if defined(PMAP_OEA)
554 /*
555 * Why not use "tlbia"? Because not all processors implement it.
556 *
557 * This needs to be a per-CPU callback to do the appropriate thing
558 * for the CPU. XXX
559 */
560 for (i = 0; i < (char *)0x00040000; i += 0x00001000) {
561 TLBIE(i);
562 EIEIO();
563 SYNC();
564 }
565 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
566 /* This is specifically for the 970, 970UM v1.6 pp. 140. */
567 for (i = 0; i <= (char *)0xFF000; i += 0x00001000) {
568 TLBIEL(i);
569 EIEIO();
570 SYNC();
571 }
572 #endif
573 TLBSYNC();
574 SYNC();
575 }
576
577 static inline register_t
va_to_vsid(const struct pmap * pm,vaddr_t addr)578 va_to_vsid(const struct pmap *pm, vaddr_t addr)
579 {
580 /*
581 * Rather than searching the STE groups for the VSID or extracting
582 * it from the SR, we know how we generate that from the ESID and
583 * so do that.
584 *
585 * This makes the code the same for OEA and OEA64, and also allows
586 * us to generate a correct-for-that-address-space VSID even if the
587 * pmap contains a different SR value at any given moment (e.g.
588 * kernel pmap on a 601 that is using I/O segments).
589 */
590 return VSID_MAKE(addr >> ADDR_SR_SHFT, pm->pm_vsid) >> SR_VSID_SHFT;
591 }
592
593 static inline register_t
va_to_pteg(const struct pmap * pm,vaddr_t addr)594 va_to_pteg(const struct pmap *pm, vaddr_t addr)
595 {
596 register_t hash;
597
598 hash = va_to_vsid(pm, addr) ^ ((addr & ADDR_PIDX) >> ADDR_PIDX_SHFT);
599 return hash & pmap_pteg_mask;
600 }
601
602 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
603 /*
604 * Given a PTE in the page table, calculate the VADDR that hashes to it.
605 * The only bit of magic is that the top 4 bits of the address doesn't
606 * technically exist in the PTE. But we know we reserved 4 bits of the
607 * VSID for it so that's how we get it.
608 */
609 static vaddr_t
pmap_pte_to_va(volatile const struct pte * pt)610 pmap_pte_to_va(volatile const struct pte *pt)
611 {
612 vaddr_t va;
613 uintptr_t ptaddr = (uintptr_t) pt;
614
615 if (pt->pte_hi & PTE_HID)
616 ptaddr ^= (pmap_pteg_mask * sizeof(struct pteg));
617
618 /* PPC Bits 10-19 PPC64 Bits 42-51 */
619 #if defined(PMAP_OEA)
620 va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x3ff;
621 #elif defined (PMAP_OEA64) || defined (PMAP_OEA64_BRIDGE)
622 va = ((pt->pte_hi >> PTE_VSID_SHFT) ^ (ptaddr / sizeof(struct pteg))) & 0x7ff;
623 #endif
624 va <<= ADDR_PIDX_SHFT;
625
626 /* PPC Bits 4-9 PPC64 Bits 36-41 */
627 va |= (pt->pte_hi & PTE_API) << ADDR_API_SHFT;
628
629 #if defined(PMAP_OEA64)
630 /* PPC63 Bits 0-35 */
631 /* va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT; */
632 #elif defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
633 /* PPC Bits 0-3 */
634 va |= VSID_TO_SR(pt->pte_hi >> PTE_VSID_SHFT) << ADDR_SR_SHFT;
635 #endif
636
637 return va;
638 }
639 #endif
640
641 static inline struct pvo_head *
pa_to_pvoh(paddr_t pa,struct vm_page ** pg_p)642 pa_to_pvoh(paddr_t pa, struct vm_page **pg_p)
643 {
644 struct vm_page *pg;
645 struct vm_page_md *md;
646 struct pmap_page *pp;
647
648 pg = PHYS_TO_VM_PAGE(pa);
649 if (pg_p != NULL)
650 *pg_p = pg;
651 if (pg == NULL) {
652 if ((pp = pmap_pv_tracked(pa)) != NULL)
653 return &pp->pp_pvoh;
654 return NULL;
655 }
656 md = VM_PAGE_TO_MD(pg);
657 return &md->mdpg_pvoh;
658 }
659
660 static inline struct pvo_head *
vm_page_to_pvoh(struct vm_page * pg)661 vm_page_to_pvoh(struct vm_page *pg)
662 {
663 struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
664
665 return &md->mdpg_pvoh;
666 }
667
668 static inline void
pmap_pp_attr_clear(struct pmap_page * pp,int ptebit)669 pmap_pp_attr_clear(struct pmap_page *pp, int ptebit)
670 {
671
672 pp->pp_attrs &= ~ptebit;
673 }
674
675 static inline void
pmap_attr_clear(struct vm_page * pg,int ptebit)676 pmap_attr_clear(struct vm_page *pg, int ptebit)
677 {
678 struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
679
680 pmap_pp_attr_clear(&md->mdpg_pp, ptebit);
681 }
682
683 static inline int
pmap_pp_attr_fetch(struct pmap_page * pp)684 pmap_pp_attr_fetch(struct pmap_page *pp)
685 {
686
687 return pp->pp_attrs;
688 }
689
690 static inline int
pmap_attr_fetch(struct vm_page * pg)691 pmap_attr_fetch(struct vm_page *pg)
692 {
693 struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
694
695 return pmap_pp_attr_fetch(&md->mdpg_pp);
696 }
697
698 static inline void
pmap_attr_save(struct vm_page * pg,int ptebit)699 pmap_attr_save(struct vm_page *pg, int ptebit)
700 {
701 struct vm_page_md * const md = VM_PAGE_TO_MD(pg);
702
703 md->mdpg_attrs |= ptebit;
704 }
705
706 static inline int
pmap_pte_compare(const volatile struct pte * pt,const struct pte * pvo_pt)707 pmap_pte_compare(const volatile struct pte *pt, const struct pte *pvo_pt)
708 {
709 if (pt->pte_hi == pvo_pt->pte_hi
710 #if 0
711 && ((pt->pte_lo ^ pvo_pt->pte_lo) &
712 ~(PTE_REF|PTE_CHG)) == 0
713 #endif
714 )
715 return 1;
716 return 0;
717 }
718
719 static inline void
pmap_pte_create(struct pte * pt,const struct pmap * pm,vaddr_t va,register_t pte_lo)720 pmap_pte_create(struct pte *pt, const struct pmap *pm, vaddr_t va, register_t pte_lo)
721 {
722 /*
723 * Construct the PTE. Default to IMB initially. Valid bit
724 * only gets set when the real pte is set in memory.
725 *
726 * Note: Don't set the valid bit for correct operation of tlb update.
727 */
728 #if defined(PMAP_OEA)
729 pt->pte_hi = (va_to_vsid(pm, va) << PTE_VSID_SHFT)
730 | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API);
731 pt->pte_lo = pte_lo;
732 #elif defined (PMAP_OEA64_BRIDGE) || defined (PMAP_OEA64)
733 pt->pte_hi = ((u_int64_t)va_to_vsid(pm, va) << PTE_VSID_SHFT)
734 | (((va & ADDR_PIDX) >> (ADDR_API_SHFT - PTE_API_SHFT)) & PTE_API);
735 pt->pte_lo = (u_int64_t) pte_lo;
736 #endif /* PMAP_OEA */
737 }
738
739 static inline void
pmap_pte_synch(volatile struct pte * pt,struct pte * pvo_pt)740 pmap_pte_synch(volatile struct pte *pt, struct pte *pvo_pt)
741 {
742 pvo_pt->pte_lo |= pt->pte_lo & (PTE_REF|PTE_CHG);
743 }
744
745 static inline void
pmap_pte_clear(volatile struct pte * pt,vaddr_t va,int ptebit)746 pmap_pte_clear(volatile struct pte *pt, vaddr_t va, int ptebit)
747 {
748 /*
749 * As shown in Section 7.6.3.2.3
750 */
751 pt->pte_lo &= ~ptebit;
752 TLBIE(va);
753 SYNC();
754 EIEIO();
755 TLBSYNC();
756 SYNC();
757 #ifdef MULTIPROCESSOR
758 DCBST(pt);
759 #endif
760 }
761
762 static inline void
pmap_pte_set(volatile struct pte * pt,struct pte * pvo_pt)763 pmap_pte_set(volatile struct pte *pt, struct pte *pvo_pt)
764 {
765 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
766 if (pvo_pt->pte_hi & PTE_VALID)
767 panic("pte_set: setting an already valid pte %p", pvo_pt);
768 #endif
769 pvo_pt->pte_hi |= PTE_VALID;
770
771 /*
772 * Update the PTE as defined in section 7.6.3.1
773 * Note that the REF/CHG bits are from pvo_pt and thus should
774 * have been saved so this routine can restore them (if desired).
775 */
776 pt->pte_lo = pvo_pt->pte_lo;
777 EIEIO();
778 pt->pte_hi = pvo_pt->pte_hi;
779 TLBSYNC();
780 SYNC();
781 #ifdef MULTIPROCESSOR
782 DCBST(pt);
783 #endif
784 pmap_pte_valid++;
785 }
786
787 static inline void
pmap_pte_unset(volatile struct pte * pt,struct pte * pvo_pt,vaddr_t va)788 pmap_pte_unset(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va)
789 {
790 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
791 if ((pvo_pt->pte_hi & PTE_VALID) == 0)
792 panic("pte_unset: attempt to unset an inactive pte#1 %p/%p", pvo_pt, pt);
793 if ((pt->pte_hi & PTE_VALID) == 0)
794 panic("pte_unset: attempt to unset an inactive pte#2 %p/%p", pvo_pt, pt);
795 #endif
796
797 pvo_pt->pte_hi &= ~PTE_VALID;
798 /*
799 * Force the ref & chg bits back into the PTEs.
800 */
801 SYNC();
802 /*
803 * Invalidate the pte ... (Section 7.6.3.3)
804 */
805 pt->pte_hi &= ~PTE_VALID;
806 SYNC();
807 TLBIE(va);
808 SYNC();
809 EIEIO();
810 TLBSYNC();
811 SYNC();
812 /*
813 * Save the ref & chg bits ...
814 */
815 pmap_pte_synch(pt, pvo_pt);
816 pmap_pte_valid--;
817 }
818
819 static inline void
pmap_pte_change(volatile struct pte * pt,struct pte * pvo_pt,vaddr_t va)820 pmap_pte_change(volatile struct pte *pt, struct pte *pvo_pt, vaddr_t va)
821 {
822 /*
823 * Invalidate the PTE
824 */
825 pmap_pte_unset(pt, pvo_pt, va);
826 pmap_pte_set(pt, pvo_pt);
827 }
828
829 /*
830 * Try to insert the PTE @ *pvo_pt into the pmap_pteg_table at ptegidx
831 * (either primary or secondary location).
832 *
833 * Note: both the destination and source PTEs must not have PTE_VALID set.
834 */
835
836 static int
pmap_pte_insert(int ptegidx,struct pte * pvo_pt)837 pmap_pte_insert(int ptegidx, struct pte *pvo_pt)
838 {
839 volatile struct pte *pt;
840 int i;
841
842 #if defined(DEBUG)
843 DPRINTFN(PTE, "pmap_pte_insert: idx %#x, pte %#" _PRIxpte " %#" _PRIxpte "\n",
844 ptegidx, pvo_pt->pte_hi, pvo_pt->pte_lo);
845 #endif
846 /*
847 * First try primary hash.
848 */
849 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
850 if ((pt->pte_hi & PTE_VALID) == 0) {
851 pvo_pt->pte_hi &= ~PTE_HID;
852 pmap_pte_set(pt, pvo_pt);
853 return i;
854 }
855 }
856
857 /*
858 * Now try secondary hash.
859 */
860 ptegidx ^= pmap_pteg_mask;
861 for (pt = pmap_pteg_table[ptegidx].pt, i = 0; i < 8; i++, pt++) {
862 if ((pt->pte_hi & PTE_VALID) == 0) {
863 pvo_pt->pte_hi |= PTE_HID;
864 pmap_pte_set(pt, pvo_pt);
865 return i;
866 }
867 }
868 return -1;
869 }
870
871 /*
872 * Spill handler.
873 *
874 * Tries to spill a page table entry from the overflow area.
875 * This runs in either real mode (if dealing with a exception spill)
876 * or virtual mode when dealing with manually spilling one of the
877 * kernel's pte entries.
878 */
879
880 int
pmap_pte_spill(struct pmap * pm,vaddr_t addr,bool isi_p)881 pmap_pte_spill(struct pmap *pm, vaddr_t addr, bool isi_p)
882 {
883 struct pvo_tqhead *spvoh, *vpvoh;
884 struct pvo_entry *pvo, *source_pvo, *victim_pvo;
885 volatile struct pteg *pteg;
886 volatile struct pte *pt;
887 register_t msr, vsid, hash;
888 int ptegidx, hid, i, j;
889 int done = 0;
890
891 PMAP_LOCK();
892 msr = pmap_interrupts_off();
893
894 /* XXXRO paranoid? */
895 if (pm->pm_evictions == 0)
896 goto out;
897
898 ptegidx = va_to_pteg(pm, addr);
899
900 /*
901 * Find source pvo.
902 */
903 spvoh = &pmap_pvo_table[ptegidx];
904 source_pvo = NULL;
905 TAILQ_FOREACH(pvo, spvoh, pvo_olink) {
906 /*
907 * We need to find pvo entry for this address...
908 */
909 PMAP_PVO_CHECK(pvo); /* sanity check */
910
911 /*
912 * If we haven't found the source and we come to a PVO with
913 * a valid PTE, then we know we can't find it because all
914 * evicted PVOs always are first in the list.
915 */
916 if ((pvo->pvo_pte.pte_hi & PTE_VALID) != 0)
917 break;
918
919 if (pm == pvo->pvo_pmap && addr == PVO_VADDR(pvo)) {
920 if (isi_p) {
921 if (!PVO_EXECUTABLE_P(pvo))
922 goto out;
923 #if defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
924 int sr __diagused =
925 PVO_VADDR(pvo) >> ADDR_SR_SHFT;
926 KASSERT((pm->pm_sr[sr] & SR_NOEXEC) == 0);
927 #endif
928 }
929 KASSERT(!PVO_PTEGIDX_ISSET(pvo));
930 /* XXXRO where check */
931 source_pvo = pvo;
932 break;
933 }
934 }
935 if (source_pvo == NULL) {
936 PMAPCOUNT(ptes_unspilled);
937 goto out;
938 }
939
940 /*
941 * Now we have found the entry to be spilled into the
942 * pteg. Attempt to insert it into the page table.
943 */
944 i = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
945 if (i >= 0) {
946 PVO_PTEGIDX_SET(pvo, i);
947 PMAP_PVO_CHECK(pvo); /* sanity check */
948 PVO_WHERE(pvo, SPILL_INSERT);
949 pvo->pvo_pmap->pm_evictions--;
950 PMAPCOUNT(ptes_spilled);
951 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID) != 0
952 ? pmap_evcnt_ptes_secondary
953 : pmap_evcnt_ptes_primary)[i]);
954
955 TAILQ_REMOVE(spvoh, pvo, pvo_olink);
956 TAILQ_INSERT_TAIL(spvoh, pvo, pvo_olink);
957
958 done = 1;
959 goto out;
960 }
961
962 /*
963 * Have to substitute some entry. Use the primary hash for this.
964 * Use low bits of timebase as random generator.
965 *
966 * XXX:
967 * Make sure we are not picking a kernel pte for replacement.
968 */
969 hid = 0;
970 i = MFTB() & 7;
971 pteg = &pmap_pteg_table[ptegidx];
972 retry:
973 for (j = 0; j < 8; j++, i = (i + 1) & 7) {
974 pt = &pteg->pt[i];
975
976 if ((pt->pte_hi & PTE_VALID) == 0)
977 break;
978
979 vsid = (pt->pte_hi & PTE_VSID) >> PTE_VSID_SHFT;
980 hash = VSID_TO_HASH(vsid);
981 if (hash < PHYSMAP_VSIDBITS)
982 break;
983 }
984 if (j == 8) {
985 if (hid != 0)
986 panic("%s: no victim\n", __func__);
987 hid = PTE_HID;
988 pteg = &pmap_pteg_table[ptegidx ^ pmap_pteg_mask];
989 goto retry;
990 }
991
992 /*
993 * We also need the pvo entry of the victim we are replacing
994 * so save the R & C bits of the PTE.
995 */
996 if ((pt->pte_hi & PTE_HID) == hid)
997 vpvoh = spvoh;
998 else
999 vpvoh = &pmap_pvo_table[ptegidx ^ pmap_pteg_mask];
1000 victim_pvo = NULL;
1001 TAILQ_FOREACH(pvo, vpvoh, pvo_olink) {
1002 PMAP_PVO_CHECK(pvo); /* sanity check */
1003
1004 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0)
1005 continue;
1006
1007 if (pmap_pte_compare(pt, &pvo->pvo_pte)) {
1008 victim_pvo = pvo;
1009 break;
1010 }
1011 }
1012 if (victim_pvo == NULL) {
1013 panic("%s: victim p-pte (%p) has no pvo entry!",
1014 __func__, pt);
1015 }
1016
1017 /*
1018 * The victim should be not be a kernel PVO/PTE entry.
1019 */
1020 KASSERT(victim_pvo->pvo_pmap != pmap_kernel());
1021 KASSERT(PVO_PTEGIDX_ISSET(victim_pvo));
1022 KASSERT(PVO_PTEGIDX_GET(victim_pvo) == i);
1023
1024 /*
1025 * We are invalidating the TLB entry for the EA for the
1026 * we are replacing even though its valid; If we don't
1027 * we lose any ref/chg bit changes contained in the TLB
1028 * entry.
1029 */
1030 if (hid == 0)
1031 source_pvo->pvo_pte.pte_hi &= ~PTE_HID;
1032 else
1033 source_pvo->pvo_pte.pte_hi |= PTE_HID;
1034
1035 /*
1036 * To enforce the PVO list ordering constraint that all
1037 * evicted entries should come before all valid entries,
1038 * move the source PVO to the tail of its list and the
1039 * victim PVO to the head of its list (which might not be
1040 * the same list, if the victim was using the secondary hash).
1041 */
1042 TAILQ_REMOVE(spvoh, source_pvo, pvo_olink);
1043 TAILQ_INSERT_TAIL(spvoh, source_pvo, pvo_olink);
1044 TAILQ_REMOVE(vpvoh, victim_pvo, pvo_olink);
1045 TAILQ_INSERT_HEAD(vpvoh, victim_pvo, pvo_olink);
1046 pmap_pte_unset(pt, &victim_pvo->pvo_pte, victim_pvo->pvo_vaddr);
1047 pmap_pte_set(pt, &source_pvo->pvo_pte);
1048 victim_pvo->pvo_pmap->pm_evictions++;
1049 source_pvo->pvo_pmap->pm_evictions--;
1050 PVO_WHERE(victim_pvo, SPILL_UNSET);
1051 PVO_WHERE(source_pvo, SPILL_SET);
1052
1053 PVO_PTEGIDX_CLR(victim_pvo);
1054 PVO_PTEGIDX_SET(source_pvo, i);
1055 PMAPCOUNT2(pmap_evcnt_ptes_primary[i]);
1056 PMAPCOUNT(ptes_spilled);
1057 PMAPCOUNT(ptes_evicted);
1058 PMAPCOUNT(ptes_removed);
1059
1060 PMAP_PVO_CHECK(victim_pvo);
1061 PMAP_PVO_CHECK(source_pvo);
1062
1063 done = 1;
1064
1065 out:
1066 pmap_interrupts_restore(msr);
1067 PMAP_UNLOCK();
1068 return done;
1069 }
1070
1071 /*
1072 * Restrict given range to physical memory
1073 */
1074 void
pmap_real_memory(paddr_t * start,psize_t * size)1075 pmap_real_memory(paddr_t *start, psize_t *size)
1076 {
1077 struct mem_region *mp;
1078
1079 for (mp = mem; mp->size; mp++) {
1080 if (*start + *size > mp->start
1081 && *start < mp->start + mp->size) {
1082 if (*start < mp->start) {
1083 *size -= mp->start - *start;
1084 *start = mp->start;
1085 }
1086 if (*start + *size > mp->start + mp->size)
1087 *size = mp->start + mp->size - *start;
1088 return;
1089 }
1090 }
1091 *size = 0;
1092 }
1093
1094 /*
1095 * Initialize anything else for pmap handling.
1096 * Called during vm_init().
1097 */
1098 void
pmap_init(void)1099 pmap_init(void)
1100 {
1101
1102 pmap_initialized = 1;
1103 }
1104
1105 /*
1106 * How much virtual space does the kernel get?
1107 */
1108 void
pmap_virtual_space(vaddr_t * start,vaddr_t * end)1109 pmap_virtual_space(vaddr_t *start, vaddr_t *end)
1110 {
1111 /*
1112 * For now, reserve one segment (minus some overhead) for kernel
1113 * virtual memory
1114 */
1115 *start = VM_MIN_KERNEL_ADDRESS;
1116 *end = VM_MAX_KERNEL_ADDRESS;
1117 }
1118
1119 /*
1120 * Allocate, initialize, and return a new physical map.
1121 */
1122 pmap_t
pmap_create(void)1123 pmap_create(void)
1124 {
1125 pmap_t pm;
1126
1127 pm = pool_get(&pmap_pool, PR_WAITOK | PR_ZERO);
1128 KASSERT((vaddr_t)pm < PMAP_DIRECT_MAPPED_LEN);
1129 pmap_pinit(pm);
1130
1131 DPRINTFN(CREATE, "pmap_create: pm %p:\n"
1132 "\t%#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr
1133 " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr "\n"
1134 "\t%#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr
1135 " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr " %#" _PRIsr "\n",
1136 pm,
1137 pm->pm_sr[0], pm->pm_sr[1],
1138 pm->pm_sr[2], pm->pm_sr[3],
1139 pm->pm_sr[4], pm->pm_sr[5],
1140 pm->pm_sr[6], pm->pm_sr[7],
1141 pm->pm_sr[8], pm->pm_sr[9],
1142 pm->pm_sr[10], pm->pm_sr[11],
1143 pm->pm_sr[12], pm->pm_sr[13],
1144 pm->pm_sr[14], pm->pm_sr[15]);
1145 return pm;
1146 }
1147
1148 /*
1149 * Initialize a preallocated and zeroed pmap structure.
1150 */
1151 void
pmap_pinit(pmap_t pm)1152 pmap_pinit(pmap_t pm)
1153 {
1154 register_t entropy = MFTB();
1155 register_t mask;
1156 int i;
1157
1158 /*
1159 * Allocate some segment registers for this pmap.
1160 */
1161 pm->pm_refs = 1;
1162 PMAP_LOCK();
1163 for (i = 0; i < NPMAPS; i += VSID_NBPW) {
1164 static register_t pmap_vsidcontext;
1165 register_t hash;
1166 unsigned int n;
1167
1168 /* Create a new value by multiplying by a prime adding in
1169 * entropy from the timebase register. This is to make the
1170 * VSID more random so that the PT Hash function collides
1171 * less often. (note that the prime causes gcc to do shifts
1172 * instead of a multiply)
1173 */
1174 pmap_vsidcontext = (pmap_vsidcontext * 0x1105) + entropy;
1175 hash = pmap_vsidcontext & (NPMAPS - 1);
1176 if (hash == 0) { /* 0 is special, avoid it */
1177 entropy += 0xbadf00d;
1178 continue;
1179 }
1180 n = hash >> 5;
1181 mask = 1L << (hash & (VSID_NBPW-1));
1182 hash = pmap_vsidcontext;
1183 if (pmap_vsid_bitmap[n] & mask) { /* collision? */
1184 /* anything free in this bucket? */
1185 if (~pmap_vsid_bitmap[n] == 0) {
1186 entropy = hash ^ (hash >> 16);
1187 continue;
1188 }
1189 i = ffs(~pmap_vsid_bitmap[n]) - 1;
1190 mask = 1L << i;
1191 hash &= ~(VSID_NBPW-1);
1192 hash |= i;
1193 }
1194 hash &= PTE_VSID >> PTE_VSID_SHFT;
1195 pmap_vsid_bitmap[n] |= mask;
1196 pm->pm_vsid = hash;
1197 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1198 for (i = 0; i < 16; i++)
1199 pm->pm_sr[i] = VSID_MAKE(i, hash) | SR_PRKEY |
1200 SR_NOEXEC;
1201 #endif
1202 PMAP_UNLOCK();
1203 return;
1204 }
1205 PMAP_UNLOCK();
1206 panic("pmap_pinit: out of segments");
1207 }
1208
1209 /*
1210 * Add a reference to the given pmap.
1211 */
1212 void
pmap_reference(pmap_t pm)1213 pmap_reference(pmap_t pm)
1214 {
1215 atomic_inc_uint(&pm->pm_refs);
1216 }
1217
1218 /*
1219 * Retire the given pmap from service.
1220 * Should only be called if the map contains no valid mappings.
1221 */
1222 void
pmap_destroy(pmap_t pm)1223 pmap_destroy(pmap_t pm)
1224 {
1225 membar_release();
1226 if (atomic_dec_uint_nv(&pm->pm_refs) == 0) {
1227 membar_acquire();
1228 pmap_release(pm);
1229 pool_put(&pmap_pool, pm);
1230 }
1231 }
1232
1233 /*
1234 * Release any resources held by the given physical map.
1235 * Called when a pmap initialized by pmap_pinit is being released.
1236 */
1237 void
pmap_release(pmap_t pm)1238 pmap_release(pmap_t pm)
1239 {
1240 int idx, mask;
1241
1242 KASSERT(pm->pm_stats.resident_count == 0);
1243 KASSERT(pm->pm_stats.wired_count == 0);
1244
1245 PMAP_LOCK();
1246 if (pm->pm_sr[0] == 0)
1247 panic("pmap_release");
1248 idx = pm->pm_vsid & (NPMAPS-1);
1249 mask = 1 << (idx % VSID_NBPW);
1250 idx /= VSID_NBPW;
1251
1252 KASSERT(pmap_vsid_bitmap[idx] & mask);
1253 pmap_vsid_bitmap[idx] &= ~mask;
1254 PMAP_UNLOCK();
1255 }
1256
1257 /*
1258 * Copy the range specified by src_addr/len
1259 * from the source map to the range dst_addr/len
1260 * in the destination map.
1261 *
1262 * This routine is only advisory and need not do anything.
1263 */
1264 void
pmap_copy(pmap_t dst_pmap,pmap_t src_pmap,vaddr_t dst_addr,vsize_t len,vaddr_t src_addr)1265 pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr,
1266 vsize_t len, vaddr_t src_addr)
1267 {
1268 PMAPCOUNT(copies);
1269 }
1270
1271 /*
1272 * Require that all active physical maps contain no
1273 * incorrect entries NOW.
1274 */
1275 void
pmap_update(struct pmap * pmap)1276 pmap_update(struct pmap *pmap)
1277 {
1278 PMAPCOUNT(updates);
1279 TLBSYNC();
1280 }
1281
1282 static inline int
pmap_pvo_pte_index(const struct pvo_entry * pvo,int ptegidx)1283 pmap_pvo_pte_index(const struct pvo_entry *pvo, int ptegidx)
1284 {
1285 int pteidx;
1286 /*
1287 * We can find the actual pte entry without searching by
1288 * grabbing the PTEG index from 3 unused bits in pte_lo[11:9]
1289 * and by noticing the HID bit.
1290 */
1291 pteidx = ptegidx * 8 + PVO_PTEGIDX_GET(pvo);
1292 if (pvo->pvo_pte.pte_hi & PTE_HID)
1293 pteidx ^= pmap_pteg_mask * 8;
1294 return pteidx;
1295 }
1296
1297 volatile struct pte *
pmap_pvo_to_pte(const struct pvo_entry * pvo,int pteidx)1298 pmap_pvo_to_pte(const struct pvo_entry *pvo, int pteidx)
1299 {
1300 volatile struct pte *pt;
1301
1302 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK)
1303 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0)
1304 return NULL;
1305 #endif
1306
1307 /*
1308 * If we haven't been supplied the ptegidx, calculate it.
1309 */
1310 if (pteidx == -1) {
1311 int ptegidx;
1312 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr);
1313 pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1314 }
1315
1316 pt = &pmap_pteg_table[pteidx >> 3].pt[pteidx & 7];
1317
1318 #if !defined(DIAGNOSTIC) && !defined(DEBUG) && !defined(PMAPCHECK)
1319 return pt;
1320 #else
1321 if ((pvo->pvo_pte.pte_hi & PTE_VALID) && !PVO_PTEGIDX_ISSET(pvo)) {
1322 panic("pmap_pvo_to_pte: pvo %p: has valid pte in "
1323 "pvo but no valid pte index", pvo);
1324 }
1325 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0 && PVO_PTEGIDX_ISSET(pvo)) {
1326 panic("pmap_pvo_to_pte: pvo %p: has valid pte index in "
1327 "pvo but no valid pte", pvo);
1328 }
1329
1330 if ((pt->pte_hi ^ (pvo->pvo_pte.pte_hi & ~PTE_VALID)) == PTE_VALID) {
1331 if ((pvo->pvo_pte.pte_hi & PTE_VALID) == 0) {
1332 #if defined(DEBUG) || defined(PMAPCHECK)
1333 pmap_pte_print(pt);
1334 #endif
1335 panic("pmap_pvo_to_pte: pvo %p: has valid pte in "
1336 "pmap_pteg_table %p but invalid in pvo",
1337 pvo, pt);
1338 }
1339 if (((pt->pte_lo ^ pvo->pvo_pte.pte_lo) & ~(PTE_CHG|PTE_REF)) != 0) {
1340 #if defined(DEBUG) || defined(PMAPCHECK)
1341 pmap_pte_print(pt);
1342 #endif
1343 panic("pmap_pvo_to_pte: pvo %p: pvo pte does "
1344 "not match pte %p in pmap_pteg_table",
1345 pvo, pt);
1346 }
1347 return pt;
1348 }
1349
1350 if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1351 #if defined(DEBUG) || defined(PMAPCHECK)
1352 pmap_pte_print(pt);
1353 #endif
1354 panic("pmap_pvo_to_pte: pvo %p: has nomatching pte %p in "
1355 "pmap_pteg_table but valid in pvo", pvo, pt);
1356 }
1357 return NULL;
1358 #endif /* !(!DIAGNOSTIC && !DEBUG && !PMAPCHECK) */
1359 }
1360
1361 struct pvo_entry *
pmap_pvo_find_va(pmap_t pm,vaddr_t va,int * pteidx_p)1362 pmap_pvo_find_va(pmap_t pm, vaddr_t va, int *pteidx_p)
1363 {
1364 struct pvo_entry *pvo;
1365 int ptegidx;
1366
1367 va &= ~ADDR_POFF;
1368 ptegidx = va_to_pteg(pm, va);
1369
1370 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1371 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1372 if ((uintptr_t) pvo >= PMAP_DIRECT_MAPPED_LEN)
1373 panic("pmap_pvo_find_va: invalid pvo %p on "
1374 "list %#x (%p)", pvo, ptegidx,
1375 &pmap_pvo_table[ptegidx]);
1376 #endif
1377 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1378 if (pteidx_p)
1379 *pteidx_p = pmap_pvo_pte_index(pvo, ptegidx);
1380 return pvo;
1381 }
1382 }
1383 if ((pm == pmap_kernel()) && (va < PMAP_DIRECT_MAPPED_LEN))
1384 panic("%s: returning NULL for %s pmap, va: %#" _PRIxva "\n",
1385 __func__, (pm == pmap_kernel() ? "kernel" : "user"), va);
1386 return NULL;
1387 }
1388
1389 #if defined(DEBUG) || defined(PMAPCHECK)
1390 void
pmap_pvo_check(const struct pvo_entry * pvo)1391 pmap_pvo_check(const struct pvo_entry *pvo)
1392 {
1393 struct pvo_head *pvo_head;
1394 struct pvo_entry *pvo0;
1395 volatile struct pte *pt;
1396 int failed = 0;
1397
1398 PMAP_LOCK();
1399
1400 if ((uintptr_t)(pvo+1) >= PMAP_DIRECT_MAPPED_LEN)
1401 panic("pmap_pvo_check: pvo %p: invalid address", pvo);
1402
1403 if ((uintptr_t)(pvo->pvo_pmap+1) >= PMAP_DIRECT_MAPPED_LEN) {
1404 printf("pmap_pvo_check: pvo %p: invalid pmap address %p\n",
1405 pvo, pvo->pvo_pmap);
1406 failed = 1;
1407 }
1408
1409 if ((uintptr_t)TAILQ_NEXT(pvo, pvo_olink) >= PMAP_DIRECT_MAPPED_LEN ||
1410 (((uintptr_t)TAILQ_NEXT(pvo, pvo_olink)) & 0x1f) != 0) {
1411 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n",
1412 pvo, TAILQ_NEXT(pvo, pvo_olink));
1413 failed = 1;
1414 }
1415
1416 if ((uintptr_t)LIST_NEXT(pvo, pvo_vlink) >= PMAP_DIRECT_MAPPED_LEN ||
1417 (((uintptr_t)LIST_NEXT(pvo, pvo_vlink)) & 0x1f) != 0) {
1418 printf("pmap_pvo_check: pvo %p: invalid ovlink address %p\n",
1419 pvo, LIST_NEXT(pvo, pvo_vlink));
1420 failed = 1;
1421 }
1422
1423 if (PVO_MANAGED_P(pvo)) {
1424 pvo_head = pa_to_pvoh(pvo->pvo_pte.pte_lo & PTE_RPGN, NULL);
1425 LIST_FOREACH(pvo0, pvo_head, pvo_vlink) {
1426 if (pvo0 == pvo)
1427 break;
1428 }
1429 if (pvo0 == NULL) {
1430 printf("pmap_pvo_check: pvo %p: not present "
1431 "on its vlist head %p\n", pvo, pvo_head);
1432 failed = 1;
1433 }
1434 } else {
1435 KASSERT(pvo->pvo_vaddr >= VM_MIN_KERNEL_ADDRESS);
1436 if (__predict_false(pvo->pvo_vaddr < VM_MIN_KERNEL_ADDRESS))
1437 failed = 1;
1438 }
1439 if (pvo != pmap_pvo_find_va(pvo->pvo_pmap, pvo->pvo_vaddr, NULL)) {
1440 printf("pmap_pvo_check: pvo %p: not present "
1441 "on its olist head\n", pvo);
1442 failed = 1;
1443 }
1444 pt = pmap_pvo_to_pte(pvo, -1);
1445 if (pt == NULL) {
1446 if (pvo->pvo_pte.pte_hi & PTE_VALID) {
1447 printf("pmap_pvo_check: pvo %p: pte_hi VALID but "
1448 "no PTE\n", pvo);
1449 failed = 1;
1450 }
1451 } else {
1452 if ((uintptr_t) pt < (uintptr_t) &pmap_pteg_table[0] ||
1453 (uintptr_t) pt >=
1454 (uintptr_t) &pmap_pteg_table[pmap_pteg_cnt]) {
1455 printf("pmap_pvo_check: pvo %p: pte %p not in "
1456 "pteg table\n", pvo, pt);
1457 failed = 1;
1458 }
1459 if (((((uintptr_t) pt) >> 3) & 7) != PVO_PTEGIDX_GET(pvo)) {
1460 printf("pmap_pvo_check: pvo %p: pte_hi VALID but "
1461 "no PTE\n", pvo);
1462 failed = 1;
1463 }
1464 if (pvo->pvo_pte.pte_hi != pt->pte_hi) {
1465 printf("pmap_pvo_check: pvo %p: pte_hi differ: "
1466 "%#" _PRIxpte "/%#" _PRIxpte "\n", pvo,
1467 pvo->pvo_pte.pte_hi,
1468 pt->pte_hi);
1469 failed = 1;
1470 }
1471 if (((pvo->pvo_pte.pte_lo ^ pt->pte_lo) &
1472 (PTE_PP|PTE_WIMG|PTE_RPGN)) != 0) {
1473 printf("pmap_pvo_check: pvo %p: pte_lo differ: "
1474 "%#" _PRIxpte "/%#" _PRIxpte "\n", pvo,
1475 (pvo->pvo_pte.pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)),
1476 (pt->pte_lo & (PTE_PP|PTE_WIMG|PTE_RPGN)));
1477 failed = 1;
1478 }
1479 if ((pmap_pte_to_va(pt) ^ PVO_VADDR(pvo)) & 0x0fffffff) {
1480 printf("pmap_pvo_check: pvo %p: PTE %p derived VA %#" _PRIxva ""
1481 " doesn't not match PVO's VA %#" _PRIxva "\n",
1482 pvo, pt, pmap_pte_to_va(pt), PVO_VADDR(pvo));
1483 failed = 1;
1484 }
1485 if (failed)
1486 pmap_pte_print(pt);
1487 }
1488 if (failed)
1489 panic("pmap_pvo_check: pvo %p, pm %p: bugcheck!", pvo,
1490 pvo->pvo_pmap);
1491
1492 PMAP_UNLOCK();
1493 }
1494 #endif /* DEBUG || PMAPCHECK */
1495
1496 /*
1497 * Search the PVO table looking for a non-wired entry.
1498 * If we find one, remove it and return it.
1499 */
1500
1501 struct pvo_entry *
pmap_pvo_reclaim(void)1502 pmap_pvo_reclaim(void)
1503 {
1504 struct pvo_tqhead *pvoh;
1505 struct pvo_entry *pvo;
1506 uint32_t idx, endidx;
1507
1508 endidx = pmap_pvo_reclaim_nextidx;
1509 for (idx = (endidx + 1) & pmap_pteg_mask; idx != endidx;
1510 idx = (idx + 1) & pmap_pteg_mask) {
1511 pvoh = &pmap_pvo_table[idx];
1512 TAILQ_FOREACH(pvo, pvoh, pvo_olink) {
1513 if (!PVO_WIRED_P(pvo)) {
1514 pmap_pvo_remove(pvo, -1, NULL);
1515 pmap_pvo_reclaim_nextidx = idx;
1516 PMAPCOUNT(pvos_reclaimed);
1517 return pvo;
1518 }
1519 }
1520 }
1521 return NULL;
1522 }
1523
1524 /*
1525 * This returns whether this is the first mapping of a page.
1526 */
1527 int
pmap_pvo_enter(pmap_t pm,struct pool * pl,struct pvo_head * pvo_head,vaddr_t va,paddr_t pa,register_t pte_lo,int flags)1528 pmap_pvo_enter(pmap_t pm, struct pool *pl, struct pvo_head *pvo_head,
1529 vaddr_t va, paddr_t pa, register_t pte_lo, int flags)
1530 {
1531 struct pvo_entry *pvo;
1532 struct pvo_tqhead *pvoh;
1533 register_t msr;
1534 int ptegidx;
1535 int i;
1536 int poolflags = PR_NOWAIT;
1537
1538 /*
1539 * Compute the PTE Group index.
1540 */
1541 va &= ~ADDR_POFF;
1542 ptegidx = va_to_pteg(pm, va);
1543
1544 msr = pmap_interrupts_off();
1545
1546 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1547 if (pmap_pvo_remove_depth > 0)
1548 panic("pmap_pvo_enter: called while pmap_pvo_remove active!");
1549 if (++pmap_pvo_enter_depth > 1)
1550 panic("pmap_pvo_enter: called recursively!");
1551 #endif
1552
1553 /*
1554 * Remove any existing mapping for this page. Reuse the
1555 * pvo entry if there a mapping.
1556 */
1557 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
1558 if (pvo->pvo_pmap == pm && PVO_VADDR(pvo) == va) {
1559 #ifdef DEBUG
1560 if ((pmapdebug & PMAPDEBUG_PVOENTER) &&
1561 ((pvo->pvo_pte.pte_lo ^ (pa|pte_lo)) &
1562 ~(PTE_REF|PTE_CHG)) == 0 &&
1563 va < VM_MIN_KERNEL_ADDRESS) {
1564 printf("pmap_pvo_enter: pvo %p: dup %#" _PRIxpte "/%#" _PRIxpa "\n",
1565 pvo, pvo->pvo_pte.pte_lo, pte_lo|pa);
1566 printf("pmap_pvo_enter: pte_hi=%#" _PRIxpte " sr=%#" _PRIsr "\n",
1567 pvo->pvo_pte.pte_hi,
1568 pm->pm_sr[va >> ADDR_SR_SHFT]);
1569 pmap_pte_print(pmap_pvo_to_pte(pvo, -1));
1570 #ifdef DDBX
1571 Debugger();
1572 #endif
1573 }
1574 #endif
1575 PMAPCOUNT(mappings_replaced);
1576 pmap_pvo_remove(pvo, -1, NULL);
1577 break;
1578 }
1579 }
1580
1581 /*
1582 * If we aren't overwriting an mapping, try to allocate
1583 */
1584 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1585 --pmap_pvo_enter_depth;
1586 #endif
1587 pmap_interrupts_restore(msr);
1588 if (pvo == NULL) {
1589 pvo = pool_get(pl, poolflags);
1590 }
1591 KASSERT((vaddr_t)pvo < VM_MIN_KERNEL_ADDRESS);
1592
1593 #ifdef DEBUG
1594 /*
1595 * Exercise pmap_pvo_reclaim() a little.
1596 */
1597 if (pvo && (flags & PMAP_CANFAIL) != 0 &&
1598 pmap_pvo_reclaim_debugctr++ > 0x1000 &&
1599 (pmap_pvo_reclaim_debugctr & 0xff) == 0) {
1600 pool_put(pl, pvo);
1601 pvo = NULL;
1602 }
1603 #endif
1604
1605 msr = pmap_interrupts_off();
1606 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1607 ++pmap_pvo_enter_depth;
1608 #endif
1609 if (pvo == NULL) {
1610 pvo = pmap_pvo_reclaim();
1611 if (pvo == NULL) {
1612 if ((flags & PMAP_CANFAIL) == 0)
1613 panic("pmap_pvo_enter: failed");
1614 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1615 pmap_pvo_enter_depth--;
1616 #endif
1617 PMAPCOUNT(pvos_failed);
1618 pmap_interrupts_restore(msr);
1619 return ENOMEM;
1620 }
1621 }
1622
1623 pvo->pvo_vaddr = va;
1624 pvo->pvo_pmap = pm;
1625 pvo->pvo_vaddr &= ~ADDR_POFF;
1626 if (flags & VM_PROT_EXECUTE) {
1627 PMAPCOUNT(exec_mappings);
1628 pvo_set_exec(pvo);
1629 }
1630 if (flags & PMAP_WIRED)
1631 pvo->pvo_vaddr |= PVO_WIRED;
1632 if (pvo_head != NULL) {
1633 pvo->pvo_vaddr |= PVO_MANAGED;
1634 PMAPCOUNT(mappings);
1635 } else {
1636 PMAPCOUNT(kernel_mappings);
1637 }
1638 pmap_pte_create(&pvo->pvo_pte, pm, va, pa | pte_lo);
1639
1640 if (pvo_head != NULL)
1641 LIST_INSERT_HEAD(pvo_head, pvo, pvo_vlink);
1642 if (PVO_WIRED_P(pvo))
1643 pvo->pvo_pmap->pm_stats.wired_count++;
1644 pvo->pvo_pmap->pm_stats.resident_count++;
1645 #if defined(DEBUG)
1646 /* if (pm != pmap_kernel() && va < VM_MIN_KERNEL_ADDRESS) */
1647 DPRINTFN(PVOENTER,
1648 "pmap_pvo_enter: pvo %p: pm %p va %#" _PRIxva " pa %#" _PRIxpa "\n",
1649 pvo, pm, va, pa);
1650 #endif
1651
1652 /*
1653 * We hope this succeeds but it isn't required.
1654 */
1655 pvoh = &pmap_pvo_table[ptegidx];
1656 i = pmap_pte_insert(ptegidx, &pvo->pvo_pte);
1657 if (i >= 0) {
1658 PVO_PTEGIDX_SET(pvo, i);
1659 PVO_WHERE(pvo, ENTER_INSERT);
1660 PMAPCOUNT2(((pvo->pvo_pte.pte_hi & PTE_HID)
1661 ? pmap_evcnt_ptes_secondary : pmap_evcnt_ptes_primary)[i]);
1662 TAILQ_INSERT_TAIL(pvoh, pvo, pvo_olink);
1663
1664 } else {
1665 /*
1666 * Since we didn't have room for this entry (which makes it
1667 * and evicted entry), place it at the head of the list.
1668 */
1669 TAILQ_INSERT_HEAD(pvoh, pvo, pvo_olink);
1670 PMAPCOUNT(ptes_evicted);
1671 pm->pm_evictions++;
1672 /*
1673 * If this is a kernel page, make sure it's active.
1674 */
1675 if (pm == pmap_kernel()) {
1676 i = pmap_pte_spill(pm, va, false);
1677 KASSERT(i);
1678 }
1679 }
1680 PMAP_PVO_CHECK(pvo); /* sanity check */
1681 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1682 pmap_pvo_enter_depth--;
1683 #endif
1684 pmap_interrupts_restore(msr);
1685 return 0;
1686 }
1687
1688 static void
pmap_pvo_remove(struct pvo_entry * pvo,int pteidx,struct pvo_head * pvol)1689 pmap_pvo_remove(struct pvo_entry *pvo, int pteidx, struct pvo_head *pvol)
1690 {
1691 volatile struct pte *pt;
1692 int ptegidx;
1693
1694 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1695 if (++pmap_pvo_remove_depth > 1)
1696 panic("pmap_pvo_remove: called recursively!");
1697 #endif
1698
1699 /*
1700 * If we haven't been supplied the ptegidx, calculate it.
1701 */
1702 if (pteidx == -1) {
1703 ptegidx = va_to_pteg(pvo->pvo_pmap, pvo->pvo_vaddr);
1704 pteidx = pmap_pvo_pte_index(pvo, ptegidx);
1705 } else {
1706 ptegidx = pteidx >> 3;
1707 if (pvo->pvo_pte.pte_hi & PTE_HID)
1708 ptegidx ^= pmap_pteg_mask;
1709 }
1710 PMAP_PVO_CHECK(pvo); /* sanity check */
1711
1712 /*
1713 * If there is an active pte entry, we need to deactivate it
1714 * (and save the ref & chg bits).
1715 */
1716 pt = pmap_pvo_to_pte(pvo, pteidx);
1717 if (pt != NULL) {
1718 pmap_pte_unset(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
1719 PVO_WHERE(pvo, REMOVE);
1720 PVO_PTEGIDX_CLR(pvo);
1721 PMAPCOUNT(ptes_removed);
1722 } else {
1723 KASSERT(pvo->pvo_pmap->pm_evictions > 0);
1724 pvo->pvo_pmap->pm_evictions--;
1725 }
1726
1727 /*
1728 * Account for executable mappings.
1729 */
1730 if (PVO_EXECUTABLE_P(pvo))
1731 pvo_clear_exec(pvo);
1732
1733 /*
1734 * Update our statistics.
1735 */
1736 pvo->pvo_pmap->pm_stats.resident_count--;
1737 if (PVO_WIRED_P(pvo))
1738 pvo->pvo_pmap->pm_stats.wired_count--;
1739
1740 /*
1741 * If the page is managed:
1742 * Save the REF/CHG bits into their cache.
1743 * Remove the PVO from the P/V list.
1744 */
1745 if (PVO_MANAGED_P(pvo)) {
1746 register_t ptelo = pvo->pvo_pte.pte_lo;
1747 struct vm_page *pg = PHYS_TO_VM_PAGE(ptelo & PTE_RPGN);
1748
1749 if (pg != NULL) {
1750 /*
1751 * If this page was changed and it is mapped exec,
1752 * invalidate it.
1753 */
1754 if ((ptelo & PTE_CHG) &&
1755 (pmap_attr_fetch(pg) & PTE_EXEC)) {
1756 struct pvo_head *pvoh = vm_page_to_pvoh(pg);
1757 if (LIST_EMPTY(pvoh)) {
1758 DPRINTFN(EXEC, "[pmap_pvo_remove: "
1759 "%#" _PRIxpa ": clear-exec]\n",
1760 VM_PAGE_TO_PHYS(pg));
1761 pmap_attr_clear(pg, PTE_EXEC);
1762 PMAPCOUNT(exec_uncached_pvo_remove);
1763 } else {
1764 DPRINTFN(EXEC, "[pmap_pvo_remove: "
1765 "%#" _PRIxpa ": syncicache]\n",
1766 VM_PAGE_TO_PHYS(pg));
1767 pmap_syncicache(VM_PAGE_TO_PHYS(pg),
1768 PAGE_SIZE);
1769 PMAPCOUNT(exec_synced_pvo_remove);
1770 }
1771 }
1772
1773 pmap_attr_save(pg, ptelo & (PTE_REF|PTE_CHG));
1774 }
1775 LIST_REMOVE(pvo, pvo_vlink);
1776 PMAPCOUNT(unmappings);
1777 } else {
1778 PMAPCOUNT(kernel_unmappings);
1779 }
1780
1781 /*
1782 * Remove the PVO from its list and return it to the pool.
1783 */
1784 TAILQ_REMOVE(&pmap_pvo_table[ptegidx], pvo, pvo_olink);
1785 if (pvol) {
1786 LIST_INSERT_HEAD(pvol, pvo, pvo_vlink);
1787 }
1788 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
1789 pmap_pvo_remove_depth--;
1790 #endif
1791 }
1792
1793 void
pmap_pvo_free(struct pvo_entry * pvo)1794 pmap_pvo_free(struct pvo_entry *pvo)
1795 {
1796
1797 pool_put(&pmap_pvo_pool, pvo);
1798 }
1799
1800 void
pmap_pvo_free_list(struct pvo_head * pvol)1801 pmap_pvo_free_list(struct pvo_head *pvol)
1802 {
1803 struct pvo_entry *pvo, *npvo;
1804
1805 for (pvo = LIST_FIRST(pvol); pvo != NULL; pvo = npvo) {
1806 npvo = LIST_NEXT(pvo, pvo_vlink);
1807 LIST_REMOVE(pvo, pvo_vlink);
1808 pmap_pvo_free(pvo);
1809 }
1810 }
1811
1812 /*
1813 * Mark a mapping as executable.
1814 * If this is the first executable mapping in the segment,
1815 * clear the noexec flag.
1816 */
1817 static void
pvo_set_exec(struct pvo_entry * pvo)1818 pvo_set_exec(struct pvo_entry *pvo)
1819 {
1820 struct pmap *pm = pvo->pvo_pmap;
1821
1822 if (pm == pmap_kernel() || PVO_EXECUTABLE_P(pvo)) {
1823 return;
1824 }
1825 pvo->pvo_vaddr |= PVO_EXECUTABLE;
1826 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1827 {
1828 int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT;
1829 if (pm->pm_exec[sr]++ == 0) {
1830 pm->pm_sr[sr] &= ~SR_NOEXEC;
1831 }
1832 }
1833 #endif
1834 }
1835
1836 /*
1837 * Mark a mapping as non-executable.
1838 * If this was the last executable mapping in the segment,
1839 * set the noexec flag.
1840 */
1841 static void
pvo_clear_exec(struct pvo_entry * pvo)1842 pvo_clear_exec(struct pvo_entry *pvo)
1843 {
1844 struct pmap *pm = pvo->pvo_pmap;
1845
1846 if (pm == pmap_kernel() || !PVO_EXECUTABLE_P(pvo)) {
1847 return;
1848 }
1849 pvo->pvo_vaddr &= ~PVO_EXECUTABLE;
1850 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
1851 {
1852 int sr = PVO_VADDR(pvo) >> ADDR_SR_SHFT;
1853 if (--pm->pm_exec[sr] == 0) {
1854 pm->pm_sr[sr] |= SR_NOEXEC;
1855 }
1856 }
1857 #endif
1858 }
1859
1860 /*
1861 * Insert physical page at pa into the given pmap at virtual address va.
1862 */
1863 int
pmap_enter(pmap_t pm,vaddr_t va,paddr_t pa,vm_prot_t prot,u_int flags)1864 pmap_enter(pmap_t pm, vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
1865 {
1866 struct mem_region *mp;
1867 struct pvo_head *pvo_head;
1868 struct vm_page *pg;
1869 register_t pte_lo;
1870 int error;
1871 u_int was_exec = 0;
1872
1873 PMAP_LOCK();
1874
1875 if (__predict_false(!pmap_initialized)) {
1876 pvo_head = NULL;
1877 pg = NULL;
1878 was_exec = PTE_EXEC;
1879
1880 } else {
1881 pvo_head = pa_to_pvoh(pa, &pg);
1882 }
1883
1884 DPRINTFN(ENTER,
1885 "pmap_enter(%p, %#" _PRIxva ", %#" _PRIxpa ", 0x%x, 0x%x):",
1886 pm, va, pa, prot, flags);
1887
1888 /*
1889 * If this is a managed page, and it's the first reference to the
1890 * page clear the execness of the page. Otherwise fetch the execness.
1891 */
1892 if (pg != NULL)
1893 was_exec = pmap_attr_fetch(pg) & PTE_EXEC;
1894
1895 DPRINTFN(ENTER, " was_exec=%d", was_exec);
1896
1897 /*
1898 * Assume the page is cache inhibited and access is guarded unless
1899 * it's in our available memory array. If it is in the memory array,
1900 * assume it's in memory coherent memory.
1901 */
1902 if (flags & PMAP_MD_PREFETCHABLE) {
1903 pte_lo = 0;
1904 } else
1905 pte_lo = PTE_G;
1906
1907 if ((flags & PMAP_NOCACHE) == 0) {
1908 for (mp = mem; mp->size; mp++) {
1909 if (pa >= mp->start && pa < mp->start + mp->size) {
1910 pte_lo = PTE_M;
1911 break;
1912 }
1913 }
1914 #ifdef MULTIPROCESSOR
1915 if (((mfpvr() >> 16) & 0xffff) == MPC603e)
1916 pte_lo = PTE_M;
1917 #endif
1918 } else {
1919 pte_lo |= PTE_I;
1920 }
1921
1922 if (prot & VM_PROT_WRITE)
1923 pte_lo |= PTE_BW;
1924 else
1925 pte_lo |= PTE_BR;
1926
1927 /*
1928 * If this was in response to a fault, "pre-fault" the PTE's
1929 * changed/referenced bit appropriately.
1930 */
1931 if (flags & VM_PROT_WRITE)
1932 pte_lo |= PTE_CHG;
1933 if (flags & VM_PROT_ALL)
1934 pte_lo |= PTE_REF;
1935
1936 /*
1937 * We need to know if this page can be executable
1938 */
1939 flags |= (prot & VM_PROT_EXECUTE);
1940
1941 /*
1942 * Record mapping for later back-translation and pte spilling.
1943 * This will overwrite any existing mapping.
1944 */
1945 error = pmap_pvo_enter(pm, &pmap_pvo_pool, pvo_head, va, pa, pte_lo, flags);
1946
1947 /*
1948 * Flush the real page from the instruction cache if this page is
1949 * mapped executable and cacheable and has not been flushed since
1950 * the last time it was modified.
1951 */
1952 if (error == 0 &&
1953 (flags & VM_PROT_EXECUTE) &&
1954 (pte_lo & PTE_I) == 0 &&
1955 was_exec == 0) {
1956 DPRINTFN(ENTER, " %s", "syncicache");
1957 PMAPCOUNT(exec_synced);
1958 pmap_syncicache(pa, PAGE_SIZE);
1959 if (pg != NULL) {
1960 pmap_attr_save(pg, PTE_EXEC);
1961 PMAPCOUNT(exec_cached);
1962 #if defined(DEBUG) || defined(PMAPDEBUG)
1963 if (pmapdebug & PMAPDEBUG_ENTER)
1964 printf(" marked-as-exec");
1965 else if (pmapdebug & PMAPDEBUG_EXEC)
1966 printf("[pmap_enter: %#" _PRIxpa ": marked-as-exec]\n",
1967 VM_PAGE_TO_PHYS(pg));
1968 #endif
1969 }
1970 }
1971
1972 DPRINTFN(ENTER, ": error=%d\n", error);
1973
1974 PMAP_UNLOCK();
1975
1976 return error;
1977 }
1978
1979 void
pmap_kenter_pa(vaddr_t va,paddr_t pa,vm_prot_t prot,u_int flags)1980 pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot, u_int flags)
1981 {
1982 struct mem_region *mp;
1983 register_t pte_lo;
1984 int error;
1985
1986 #if defined (PMAP_OEA64_BRIDGE) || defined (PMAP_OEA)
1987 if (va < VM_MIN_KERNEL_ADDRESS)
1988 panic("pmap_kenter_pa: attempt to enter "
1989 "non-kernel address %#" _PRIxva "!", va);
1990 #endif
1991
1992 DPRINTFN(KENTER,
1993 "pmap_kenter_pa(%#" _PRIxva ",%#" _PRIxpa ",%#x)\n", va, pa, prot);
1994
1995 PMAP_LOCK();
1996
1997 /*
1998 * Assume the page is cache inhibited and access is guarded unless
1999 * it's in our available memory array. If it is in the memory array,
2000 * assume it's in memory coherent memory.
2001 */
2002 pte_lo = PTE_IG;
2003 if ((flags & PMAP_NOCACHE) == 0) {
2004 for (mp = mem; mp->size; mp++) {
2005 if (pa >= mp->start && pa < mp->start + mp->size) {
2006 pte_lo = PTE_M;
2007 break;
2008 }
2009 }
2010 #ifdef MULTIPROCESSOR
2011 if (((mfpvr() >> 16) & 0xffff) == MPC603e)
2012 pte_lo = PTE_M;
2013 #endif
2014 }
2015
2016 if (prot & VM_PROT_WRITE)
2017 pte_lo |= PTE_BW;
2018 else
2019 pte_lo |= PTE_BR;
2020
2021 /*
2022 * We don't care about REF/CHG on PVOs on the unmanaged list.
2023 */
2024 error = pmap_pvo_enter(pmap_kernel(), &pmap_pvo_pool,
2025 NULL, va, pa, pte_lo, prot|PMAP_WIRED);
2026
2027 if (error != 0)
2028 panic("pmap_kenter_pa: failed to enter va %#" _PRIxva " pa %#" _PRIxpa ": %d",
2029 va, pa, error);
2030
2031 PMAP_UNLOCK();
2032 }
2033
2034 void
pmap_kremove(vaddr_t va,vsize_t len)2035 pmap_kremove(vaddr_t va, vsize_t len)
2036 {
2037 if (va < VM_MIN_KERNEL_ADDRESS)
2038 panic("pmap_kremove: attempt to remove "
2039 "non-kernel address %#" _PRIxva "!", va);
2040
2041 DPRINTFN(KREMOVE, "pmap_kremove(%#" _PRIxva ",%#" _PRIxva ")\n", va, len);
2042 pmap_remove(pmap_kernel(), va, va + len);
2043 }
2044
2045 /*
2046 * Remove the given range of mapping entries.
2047 */
2048 void
pmap_remove(pmap_t pm,vaddr_t va,vaddr_t endva)2049 pmap_remove(pmap_t pm, vaddr_t va, vaddr_t endva)
2050 {
2051 struct pvo_head pvol;
2052 struct pvo_entry *pvo;
2053 register_t msr;
2054 int pteidx;
2055
2056 PMAP_LOCK();
2057 LIST_INIT(&pvol);
2058 msr = pmap_interrupts_off();
2059 for (; va < endva; va += PAGE_SIZE) {
2060 pvo = pmap_pvo_find_va(pm, va, &pteidx);
2061 if (pvo != NULL) {
2062 pmap_pvo_remove(pvo, pteidx, &pvol);
2063 }
2064 }
2065 pmap_interrupts_restore(msr);
2066 pmap_pvo_free_list(&pvol);
2067 PMAP_UNLOCK();
2068 }
2069
2070 #if defined(PMAP_OEA)
2071 #ifdef PPC_OEA601
2072 bool
pmap_extract_ioseg601(vaddr_t va,paddr_t * pap)2073 pmap_extract_ioseg601(vaddr_t va, paddr_t *pap)
2074 {
2075 if ((MFPVR() >> 16) != MPC601)
2076 return false;
2077
2078 const register_t sr = iosrtable[va >> ADDR_SR_SHFT];
2079
2080 if (SR601_VALID_P(sr) && SR601_PA_MATCH_P(sr, va)) {
2081 if (pap)
2082 *pap = va;
2083 return true;
2084 }
2085 return false;
2086 }
2087
2088 static bool
pmap_extract_battable601(vaddr_t va,paddr_t * pap)2089 pmap_extract_battable601(vaddr_t va, paddr_t *pap)
2090 {
2091 const register_t batu = battable[va >> 23].batu;
2092 const register_t batl = battable[va >> 23].batl;
2093
2094 if (BAT601_VALID_P(batl) && BAT601_VA_MATCH_P(batu, batl, va)) {
2095 const register_t mask =
2096 (~(batl & BAT601_BSM) << 17) & ~0x1ffffL;
2097 if (pap)
2098 *pap = (batl & mask) | (va & ~mask);
2099 return true;
2100 }
2101 return false;
2102 }
2103 #endif /* PPC_OEA601 */
2104
2105 bool
pmap_extract_battable(vaddr_t va,paddr_t * pap)2106 pmap_extract_battable(vaddr_t va, paddr_t *pap)
2107 {
2108 #ifdef PPC_OEA601
2109 if ((MFPVR() >> 16) == MPC601)
2110 return pmap_extract_battable601(va, pap);
2111 #endif /* PPC_OEA601 */
2112
2113 if (oeacpufeat & OEACPU_NOBAT)
2114 return false;
2115
2116 const register_t batu = battable[BAT_VA2IDX(va)].batu;
2117
2118 if (BAT_VALID_P(batu, 0) && BAT_VA_MATCH_P(batu, va)) {
2119 const register_t batl = battable[BAT_VA2IDX(va)].batl;
2120 const register_t mask =
2121 (~(batu & (BAT_XBL|BAT_BL)) << 15) & ~0x1ffffL;
2122 if (pap)
2123 *pap = (batl & mask) | (va & ~mask);
2124 return true;
2125 }
2126 return false;
2127 }
2128 #endif /* PMAP_OEA */
2129
2130 /*
2131 * Get the physical page address for the given pmap/virtual address.
2132 */
2133 bool
pmap_extract(pmap_t pm,vaddr_t va,paddr_t * pap)2134 pmap_extract(pmap_t pm, vaddr_t va, paddr_t *pap)
2135 {
2136 struct pvo_entry *pvo;
2137 register_t msr;
2138
2139 PMAP_LOCK();
2140
2141 /*
2142 * If this is the kernel pmap, check the battable and I/O
2143 * segments for a hit. This is done only for regions outside
2144 * VM_MIN_KERNEL_ADDRESS-VM_MAX_KERNEL_ADDRESS.
2145 *
2146 * Be careful when checking VM_MAX_KERNEL_ADDRESS; you don't
2147 * want to wrap around to 0.
2148 */
2149 if (pm == pmap_kernel() &&
2150 (va < VM_MIN_KERNEL_ADDRESS ||
2151 (KERNEL2_SR < 15 && VM_MAX_KERNEL_ADDRESS <= va))) {
2152 KASSERT((va >> ADDR_SR_SHFT) != USER_SR);
2153 #if defined(PMAP_OEA)
2154 #ifdef PPC_OEA601
2155 if (pmap_extract_ioseg601(va, pap)) {
2156 PMAP_UNLOCK();
2157 return true;
2158 }
2159 #endif /* PPC_OEA601 */
2160 if (pmap_extract_battable(va, pap)) {
2161 PMAP_UNLOCK();
2162 return true;
2163 }
2164 /*
2165 * We still check the HTAB...
2166 */
2167 #elif defined(PMAP_OEA64_BRIDGE)
2168 if (va < PMAP_DIRECT_MAPPED_LEN) {
2169 if (pap)
2170 *pap = va;
2171 PMAP_UNLOCK();
2172 return true;
2173 }
2174 /*
2175 * We still check the HTAB...
2176 */
2177 #elif defined(PMAP_OEA64)
2178 #error PPC_OEA64 not supported
2179 #endif /* PPC_OEA */
2180 }
2181
2182 msr = pmap_interrupts_off();
2183 pvo = pmap_pvo_find_va(pm, va & ~ADDR_POFF, NULL);
2184 if (pvo != NULL) {
2185 PMAP_PVO_CHECK(pvo); /* sanity check */
2186 if (pap)
2187 *pap = (pvo->pvo_pte.pte_lo & PTE_RPGN)
2188 | (va & ADDR_POFF);
2189 }
2190 pmap_interrupts_restore(msr);
2191 PMAP_UNLOCK();
2192 return pvo != NULL;
2193 }
2194
2195 /*
2196 * Lower the protection on the specified range of this pmap.
2197 */
2198 void
pmap_protect(pmap_t pm,vaddr_t va,vaddr_t endva,vm_prot_t prot)2199 pmap_protect(pmap_t pm, vaddr_t va, vaddr_t endva, vm_prot_t prot)
2200 {
2201 struct pvo_entry *pvo;
2202 volatile struct pte *pt;
2203 register_t msr;
2204 int pteidx;
2205
2206 /*
2207 * Since this routine only downgrades protection, we should
2208 * always be called with at least one bit not set.
2209 */
2210 KASSERT(prot != VM_PROT_ALL);
2211
2212 /*
2213 * If there is no protection, this is equivalent to
2214 * remove the pmap from the pmap.
2215 */
2216 if ((prot & VM_PROT_READ) == 0) {
2217 pmap_remove(pm, va, endva);
2218 return;
2219 }
2220
2221 PMAP_LOCK();
2222
2223 msr = pmap_interrupts_off();
2224 for (; va < endva; va += PAGE_SIZE) {
2225 pvo = pmap_pvo_find_va(pm, va, &pteidx);
2226 if (pvo == NULL)
2227 continue;
2228 PMAP_PVO_CHECK(pvo); /* sanity check */
2229
2230 /*
2231 * Revoke executable if asked to do so.
2232 */
2233 if ((prot & VM_PROT_EXECUTE) == 0)
2234 pvo_clear_exec(pvo);
2235
2236 #if 0
2237 /*
2238 * If the page is already read-only, no change
2239 * needs to be made.
2240 */
2241 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR)
2242 continue;
2243 #endif
2244 /*
2245 * Grab the PTE pointer before we diddle with
2246 * the cached PTE copy.
2247 */
2248 pt = pmap_pvo_to_pte(pvo, pteidx);
2249 /*
2250 * Change the protection of the page.
2251 */
2252 pvo->pvo_pte.pte_lo &= ~PTE_PP;
2253 pvo->pvo_pte.pte_lo |= PTE_BR;
2254
2255 /*
2256 * If the PVO is in the page table, update
2257 * that pte at well.
2258 */
2259 if (pt != NULL) {
2260 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
2261 PVO_WHERE(pvo, PMAP_PROTECT);
2262 PMAPCOUNT(ptes_changed);
2263 }
2264
2265 PMAP_PVO_CHECK(pvo); /* sanity check */
2266 }
2267 pmap_interrupts_restore(msr);
2268 PMAP_UNLOCK();
2269 }
2270
2271 void
pmap_unwire(pmap_t pm,vaddr_t va)2272 pmap_unwire(pmap_t pm, vaddr_t va)
2273 {
2274 struct pvo_entry *pvo;
2275 register_t msr;
2276
2277 PMAP_LOCK();
2278 msr = pmap_interrupts_off();
2279 pvo = pmap_pvo_find_va(pm, va, NULL);
2280 if (pvo != NULL) {
2281 if (PVO_WIRED_P(pvo)) {
2282 pvo->pvo_vaddr &= ~PVO_WIRED;
2283 pm->pm_stats.wired_count--;
2284 }
2285 PMAP_PVO_CHECK(pvo); /* sanity check */
2286 }
2287 pmap_interrupts_restore(msr);
2288 PMAP_UNLOCK();
2289 }
2290
2291 static void
pmap_pp_protect(struct pmap_page * pp,paddr_t pa,vm_prot_t prot)2292 pmap_pp_protect(struct pmap_page *pp, paddr_t pa, vm_prot_t prot)
2293 {
2294 struct pvo_head *pvo_head, pvol;
2295 struct pvo_entry *pvo, *next_pvo;
2296 volatile struct pte *pt;
2297 register_t msr;
2298
2299 PMAP_LOCK();
2300
2301 KASSERT(prot != VM_PROT_ALL);
2302 LIST_INIT(&pvol);
2303 msr = pmap_interrupts_off();
2304
2305 /*
2306 * When UVM reuses a page, it does a pmap_page_protect with
2307 * VM_PROT_NONE. At that point, we can clear the exec flag
2308 * since we know the page will have different contents.
2309 */
2310 if ((prot & VM_PROT_READ) == 0) {
2311 DPRINTFN(EXEC, "[pmap_page_protect: %#" _PRIxpa ": clear-exec]\n",
2312 pa);
2313 if (pmap_pp_attr_fetch(pp) & PTE_EXEC) {
2314 PMAPCOUNT(exec_uncached_page_protect);
2315 pmap_pp_attr_clear(pp, PTE_EXEC);
2316 }
2317 }
2318
2319 pvo_head = &pp->pp_pvoh;
2320 for (pvo = LIST_FIRST(pvo_head); pvo != NULL; pvo = next_pvo) {
2321 next_pvo = LIST_NEXT(pvo, pvo_vlink);
2322 PMAP_PVO_CHECK(pvo); /* sanity check */
2323
2324 /*
2325 * Downgrading to no mapping at all, we just remove the entry.
2326 */
2327 if ((prot & VM_PROT_READ) == 0) {
2328 pmap_pvo_remove(pvo, -1, &pvol);
2329 continue;
2330 }
2331
2332 /*
2333 * If EXEC permission is being revoked, just clear the
2334 * flag in the PVO.
2335 */
2336 if ((prot & VM_PROT_EXECUTE) == 0)
2337 pvo_clear_exec(pvo);
2338
2339 /*
2340 * If this entry is already RO, don't diddle with the
2341 * page table.
2342 */
2343 if ((pvo->pvo_pte.pte_lo & PTE_PP) == PTE_BR) {
2344 PMAP_PVO_CHECK(pvo);
2345 continue;
2346 }
2347
2348 /*
2349 * Grab the PTE before the we diddle the bits so
2350 * pvo_to_pte can verify the pte contents are as
2351 * expected.
2352 */
2353 pt = pmap_pvo_to_pte(pvo, -1);
2354 pvo->pvo_pte.pte_lo &= ~PTE_PP;
2355 pvo->pvo_pte.pte_lo |= PTE_BR;
2356 if (pt != NULL) {
2357 pmap_pte_change(pt, &pvo->pvo_pte, pvo->pvo_vaddr);
2358 PVO_WHERE(pvo, PMAP_PAGE_PROTECT);
2359 PMAPCOUNT(ptes_changed);
2360 }
2361 PMAP_PVO_CHECK(pvo); /* sanity check */
2362 }
2363 pmap_interrupts_restore(msr);
2364 pmap_pvo_free_list(&pvol);
2365
2366 PMAP_UNLOCK();
2367 }
2368
2369 /*
2370 * Lower the protection on the specified physical page.
2371 */
2372 void
pmap_page_protect(struct vm_page * pg,vm_prot_t prot)2373 pmap_page_protect(struct vm_page *pg, vm_prot_t prot)
2374 {
2375 struct vm_page_md *md = VM_PAGE_TO_MD(pg);
2376
2377 pmap_pp_protect(&md->mdpg_pp, VM_PAGE_TO_PHYS(pg), prot);
2378 }
2379
2380 /*
2381 * Lower the protection on the physical page at the specified physical
2382 * address, which may not be managed and so may not have a struct
2383 * vm_page.
2384 */
2385 void
pmap_pv_protect(paddr_t pa,vm_prot_t prot)2386 pmap_pv_protect(paddr_t pa, vm_prot_t prot)
2387 {
2388 struct pmap_page *pp;
2389
2390 if ((pp = pmap_pv_tracked(pa)) == NULL)
2391 return;
2392 pmap_pp_protect(pp, pa, prot);
2393 }
2394
2395 /*
2396 * Activate the address space for the specified process. If the process
2397 * is the current process, load the new MMU context.
2398 */
2399 void
pmap_activate(struct lwp * l)2400 pmap_activate(struct lwp *l)
2401 {
2402 struct pcb *pcb = lwp_getpcb(l);
2403 pmap_t pmap = l->l_proc->p_vmspace->vm_map.pmap;
2404
2405 DPRINTFN(ACTIVATE,
2406 "pmap_activate: lwp %p (curlwp %p)\n", l, curlwp);
2407
2408 /*
2409 * XXX Normally performed in cpu_lwp_fork().
2410 */
2411 pcb->pcb_pm = pmap;
2412
2413 /*
2414 * In theory, the SR registers need only be valid on return
2415 * to user space wait to do them there.
2416 */
2417 if (l == curlwp) {
2418 /* Store pointer to new current pmap. */
2419 curpm = pmap;
2420 }
2421 }
2422
2423 /*
2424 * Deactivate the specified process's address space.
2425 */
2426 void
pmap_deactivate(struct lwp * l)2427 pmap_deactivate(struct lwp *l)
2428 {
2429 }
2430
2431 bool
pmap_query_bit(struct vm_page * pg,int ptebit)2432 pmap_query_bit(struct vm_page *pg, int ptebit)
2433 {
2434 struct pvo_entry *pvo;
2435 volatile struct pte *pt;
2436 register_t msr;
2437
2438 PMAP_LOCK();
2439
2440 if (pmap_attr_fetch(pg) & ptebit) {
2441 PMAP_UNLOCK();
2442 return true;
2443 }
2444
2445 msr = pmap_interrupts_off();
2446 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) {
2447 PMAP_PVO_CHECK(pvo); /* sanity check */
2448 /*
2449 * See if we saved the bit off. If so cache, it and return
2450 * success.
2451 */
2452 if (pvo->pvo_pte.pte_lo & ptebit) {
2453 pmap_attr_save(pg, ptebit);
2454 PMAP_PVO_CHECK(pvo); /* sanity check */
2455 pmap_interrupts_restore(msr);
2456 PMAP_UNLOCK();
2457 return true;
2458 }
2459 }
2460 /*
2461 * No luck, now go thru the hard part of looking at the ptes
2462 * themselves. Sync so any pending REF/CHG bits are flushed
2463 * to the PTEs.
2464 */
2465 SYNC();
2466 LIST_FOREACH(pvo, vm_page_to_pvoh(pg), pvo_vlink) {
2467 PMAP_PVO_CHECK(pvo); /* sanity check */
2468 /*
2469 * See if this pvo have a valid PTE. If so, fetch the
2470 * REF/CHG bits from the valid PTE. If the appropriate
2471 * ptebit is set, cache, it and return success.
2472 */
2473 pt = pmap_pvo_to_pte(pvo, -1);
2474 if (pt != NULL) {
2475 pmap_pte_synch(pt, &pvo->pvo_pte);
2476 if (pvo->pvo_pte.pte_lo & ptebit) {
2477 pmap_attr_save(pg, ptebit);
2478 PMAP_PVO_CHECK(pvo); /* sanity check */
2479 pmap_interrupts_restore(msr);
2480 PMAP_UNLOCK();
2481 return true;
2482 }
2483 }
2484 }
2485 pmap_interrupts_restore(msr);
2486 PMAP_UNLOCK();
2487 return false;
2488 }
2489
2490 bool
pmap_clear_bit(struct vm_page * pg,int ptebit)2491 pmap_clear_bit(struct vm_page *pg, int ptebit)
2492 {
2493 struct pvo_head *pvoh = vm_page_to_pvoh(pg);
2494 struct pvo_entry *pvo;
2495 volatile struct pte *pt;
2496 register_t msr;
2497 int rv = 0;
2498
2499 PMAP_LOCK();
2500 msr = pmap_interrupts_off();
2501
2502 /*
2503 * Fetch the cache value
2504 */
2505 rv |= pmap_attr_fetch(pg);
2506
2507 /*
2508 * Clear the cached value.
2509 */
2510 pmap_attr_clear(pg, ptebit);
2511
2512 /*
2513 * Sync so any pending REF/CHG bits are flushed to the PTEs (so we
2514 * can reset the right ones). Note that since the pvo entries and
2515 * list heads are accessed via BAT0 and are never placed in the
2516 * page table, we don't have to worry about further accesses setting
2517 * the REF/CHG bits.
2518 */
2519 SYNC();
2520
2521 /*
2522 * For each pvo entry, clear pvo's ptebit. If this pvo have a
2523 * valid PTE. If so, clear the ptebit from the valid PTE.
2524 */
2525 LIST_FOREACH(pvo, pvoh, pvo_vlink) {
2526 PMAP_PVO_CHECK(pvo); /* sanity check */
2527 pt = pmap_pvo_to_pte(pvo, -1);
2528 if (pt != NULL) {
2529 /*
2530 * Only sync the PTE if the bit we are looking
2531 * for is not already set.
2532 */
2533 if ((pvo->pvo_pte.pte_lo & ptebit) == 0)
2534 pmap_pte_synch(pt, &pvo->pvo_pte);
2535 /*
2536 * If the bit we are looking for was already set,
2537 * clear that bit in the pte.
2538 */
2539 if (pvo->pvo_pte.pte_lo & ptebit)
2540 pmap_pte_clear(pt, PVO_VADDR(pvo), ptebit);
2541 }
2542 rv |= pvo->pvo_pte.pte_lo & (PTE_CHG|PTE_REF);
2543 pvo->pvo_pte.pte_lo &= ~ptebit;
2544 PMAP_PVO_CHECK(pvo); /* sanity check */
2545 }
2546 pmap_interrupts_restore(msr);
2547
2548 /*
2549 * If we are clearing the modify bit and this page was marked EXEC
2550 * and the user of the page thinks the page was modified, then we
2551 * need to clean it from the icache if it's mapped or clear the EXEC
2552 * bit if it's not mapped. The page itself might not have the CHG
2553 * bit set if the modification was done via DMA to the page.
2554 */
2555 if ((ptebit & PTE_CHG) && (rv & PTE_EXEC)) {
2556 if (LIST_EMPTY(pvoh)) {
2557 DPRINTFN(EXEC, "[pmap_clear_bit: %#" _PRIxpa ": clear-exec]\n",
2558 VM_PAGE_TO_PHYS(pg));
2559 pmap_attr_clear(pg, PTE_EXEC);
2560 PMAPCOUNT(exec_uncached_clear_modify);
2561 } else {
2562 DPRINTFN(EXEC, "[pmap_clear_bit: %#" _PRIxpa ": syncicache]\n",
2563 VM_PAGE_TO_PHYS(pg));
2564 pmap_syncicache(VM_PAGE_TO_PHYS(pg), PAGE_SIZE);
2565 PMAPCOUNT(exec_synced_clear_modify);
2566 }
2567 }
2568 PMAP_UNLOCK();
2569 return (rv & ptebit) != 0;
2570 }
2571
2572 void
pmap_procwr(struct proc * p,vaddr_t va,size_t len)2573 pmap_procwr(struct proc *p, vaddr_t va, size_t len)
2574 {
2575 struct pvo_entry *pvo;
2576 size_t offset = va & ADDR_POFF;
2577 int s;
2578
2579 PMAP_LOCK();
2580 s = splvm();
2581 while (len > 0) {
2582 size_t seglen = PAGE_SIZE - offset;
2583 if (seglen > len)
2584 seglen = len;
2585 pvo = pmap_pvo_find_va(p->p_vmspace->vm_map.pmap, va, NULL);
2586 if (pvo != NULL && PVO_EXECUTABLE_P(pvo)) {
2587 pmap_syncicache(
2588 (pvo->pvo_pte.pte_lo & PTE_RPGN) | offset, seglen);
2589 PMAP_PVO_CHECK(pvo);
2590 }
2591 va += seglen;
2592 len -= seglen;
2593 offset = 0;
2594 }
2595 splx(s);
2596 PMAP_UNLOCK();
2597 }
2598
2599 #if defined(DEBUG) || defined(PMAPCHECK) || defined(DDB)
2600 void
pmap_pte_print(volatile struct pte * pt)2601 pmap_pte_print(volatile struct pte *pt)
2602 {
2603 printf("PTE %p: ", pt);
2604
2605 #if defined(PMAP_OEA)
2606 /* High word: */
2607 printf("%#" _PRIxpte ": [", pt->pte_hi);
2608 #else
2609 printf("%#" _PRIxpte ": [", pt->pte_hi);
2610 #endif /* PMAP_OEA */
2611
2612 printf("%c ", (pt->pte_hi & PTE_VALID) ? 'v' : 'i');
2613 printf("%c ", (pt->pte_hi & PTE_HID) ? 'h' : '-');
2614
2615 printf("%#" _PRIxpte " %#" _PRIxpte "",
2616 (pt->pte_hi &~ PTE_VALID)>>PTE_VSID_SHFT,
2617 pt->pte_hi & PTE_API);
2618 #if defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
2619 printf(" (va %#" _PRIxva ")] ", pmap_pte_to_va(pt));
2620 #else
2621 printf(" (va %#" _PRIxva ")] ", pmap_pte_to_va(pt));
2622 #endif /* PMAP_OEA */
2623
2624 /* Low word: */
2625 #if defined (PMAP_OEA)
2626 printf(" %#" _PRIxpte ": [", pt->pte_lo);
2627 printf("%#" _PRIxpte "... ", pt->pte_lo >> 12);
2628 #else
2629 printf(" %#" _PRIxpte ": [", pt->pte_lo);
2630 printf("%#" _PRIxpte "... ", pt->pte_lo >> 12);
2631 #endif
2632 printf("%c ", (pt->pte_lo & PTE_REF) ? 'r' : 'u');
2633 printf("%c ", (pt->pte_lo & PTE_CHG) ? 'c' : 'n');
2634 printf("%c", (pt->pte_lo & PTE_W) ? 'w' : '.');
2635 printf("%c", (pt->pte_lo & PTE_I) ? 'i' : '.');
2636 printf("%c", (pt->pte_lo & PTE_M) ? 'm' : '.');
2637 printf("%c ", (pt->pte_lo & PTE_G) ? 'g' : '.');
2638 switch (pt->pte_lo & PTE_PP) {
2639 case PTE_BR: printf("br]\n"); break;
2640 case PTE_BW: printf("bw]\n"); break;
2641 case PTE_SO: printf("so]\n"); break;
2642 case PTE_SW: printf("sw]\n"); break;
2643 }
2644 }
2645 #endif
2646
2647 #if defined(DDB)
2648 void
pmap_pteg_check(void)2649 pmap_pteg_check(void)
2650 {
2651 volatile struct pte *pt;
2652 int i;
2653 int ptegidx;
2654 u_int p_valid = 0;
2655 u_int s_valid = 0;
2656 u_int invalid = 0;
2657
2658 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2659 for (pt = pmap_pteg_table[ptegidx].pt, i = 8; --i >= 0; pt++) {
2660 if (pt->pte_hi & PTE_VALID) {
2661 if (pt->pte_hi & PTE_HID)
2662 s_valid++;
2663 else
2664 {
2665 p_valid++;
2666 }
2667 } else
2668 invalid++;
2669 }
2670 }
2671 printf("pteg_check: v(p) %#x (%d), v(s) %#x (%d), i %#x (%d)\n",
2672 p_valid, p_valid, s_valid, s_valid,
2673 invalid, invalid);
2674 }
2675
2676 void
pmap_print_mmuregs(void)2677 pmap_print_mmuregs(void)
2678 {
2679 int i;
2680 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
2681 u_int cpuvers;
2682 #endif
2683 #ifndef PMAP_OEA64
2684 vaddr_t addr;
2685 register_t soft_sr[16];
2686 #endif
2687 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
2688 struct bat soft_ibat[4];
2689 struct bat soft_dbat[4];
2690 #endif
2691 paddr_t sdr1;
2692
2693 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
2694 cpuvers = MFPVR() >> 16;
2695 #endif
2696 __asm volatile ("mfsdr1 %0" : "=r"(sdr1));
2697 #ifndef PMAP_OEA64
2698 addr = 0;
2699 for (i = 0; i < 16; i++) {
2700 soft_sr[i] = MFSRIN(addr);
2701 addr += (1 << ADDR_SR_SHFT);
2702 }
2703 #endif
2704
2705 #if defined (PMAP_OEA) || defined (PMAP_OEA64_BRIDGE)
2706 /* read iBAT (601: uBAT) registers */
2707 __asm volatile ("mfibatu %0,0" : "=r"(soft_ibat[0].batu));
2708 __asm volatile ("mfibatl %0,0" : "=r"(soft_ibat[0].batl));
2709 __asm volatile ("mfibatu %0,1" : "=r"(soft_ibat[1].batu));
2710 __asm volatile ("mfibatl %0,1" : "=r"(soft_ibat[1].batl));
2711 __asm volatile ("mfibatu %0,2" : "=r"(soft_ibat[2].batu));
2712 __asm volatile ("mfibatl %0,2" : "=r"(soft_ibat[2].batl));
2713 __asm volatile ("mfibatu %0,3" : "=r"(soft_ibat[3].batu));
2714 __asm volatile ("mfibatl %0,3" : "=r"(soft_ibat[3].batl));
2715
2716
2717 if (cpuvers != MPC601) {
2718 /* read dBAT registers */
2719 __asm volatile ("mfdbatu %0,0" : "=r"(soft_dbat[0].batu));
2720 __asm volatile ("mfdbatl %0,0" : "=r"(soft_dbat[0].batl));
2721 __asm volatile ("mfdbatu %0,1" : "=r"(soft_dbat[1].batu));
2722 __asm volatile ("mfdbatl %0,1" : "=r"(soft_dbat[1].batl));
2723 __asm volatile ("mfdbatu %0,2" : "=r"(soft_dbat[2].batu));
2724 __asm volatile ("mfdbatl %0,2" : "=r"(soft_dbat[2].batl));
2725 __asm volatile ("mfdbatu %0,3" : "=r"(soft_dbat[3].batu));
2726 __asm volatile ("mfdbatl %0,3" : "=r"(soft_dbat[3].batl));
2727 }
2728 #endif
2729
2730 printf("SDR1:\t%#" _PRIxpa "\n", sdr1);
2731 #ifndef PMAP_OEA64
2732 printf("SR[]:\t");
2733 for (i = 0; i < 4; i++)
2734 printf("0x%08lx, ", soft_sr[i]);
2735 printf("\n\t");
2736 for ( ; i < 8; i++)
2737 printf("0x%08lx, ", soft_sr[i]);
2738 printf("\n\t");
2739 for ( ; i < 12; i++)
2740 printf("0x%08lx, ", soft_sr[i]);
2741 printf("\n\t");
2742 for ( ; i < 16; i++)
2743 printf("0x%08lx, ", soft_sr[i]);
2744 printf("\n");
2745 #endif
2746
2747 #if defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
2748 printf("%cBAT[]:\t", cpuvers == MPC601 ? 'u' : 'i');
2749 for (i = 0; i < 4; i++) {
2750 printf("0x%08lx 0x%08lx, ",
2751 soft_ibat[i].batu, soft_ibat[i].batl);
2752 if (i == 1)
2753 printf("\n\t");
2754 }
2755 if (cpuvers != MPC601) {
2756 printf("\ndBAT[]:\t");
2757 for (i = 0; i < 4; i++) {
2758 printf("0x%08lx 0x%08lx, ",
2759 soft_dbat[i].batu, soft_dbat[i].batl);
2760 if (i == 1)
2761 printf("\n\t");
2762 }
2763 }
2764 printf("\n");
2765 #endif /* PMAP_OEA... */
2766 }
2767
2768 void
pmap_print_pte(pmap_t pm,vaddr_t va)2769 pmap_print_pte(pmap_t pm, vaddr_t va)
2770 {
2771 struct pvo_entry *pvo;
2772 volatile struct pte *pt;
2773 int pteidx;
2774
2775 pvo = pmap_pvo_find_va(pm, va, &pteidx);
2776 if (pvo != NULL) {
2777 pt = pmap_pvo_to_pte(pvo, pteidx);
2778 if (pt != NULL) {
2779 printf("VA %#" _PRIxva " -> %p -> %s %#" _PRIxpte ", %#" _PRIxpte "\n",
2780 va, pt,
2781 pt->pte_hi & PTE_HID ? "(sec)" : "(pri)",
2782 pt->pte_hi, pt->pte_lo);
2783 } else {
2784 printf("No valid PTE found\n");
2785 }
2786 } else {
2787 printf("Address not in pmap\n");
2788 }
2789 }
2790
2791 void
pmap_pteg_dist(void)2792 pmap_pteg_dist(void)
2793 {
2794 struct pvo_entry *pvo;
2795 int ptegidx;
2796 int depth;
2797 int max_depth = 0;
2798 unsigned int depths[64];
2799
2800 memset(depths, 0, sizeof(depths));
2801 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2802 depth = 0;
2803 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
2804 depth++;
2805 }
2806 if (depth > max_depth)
2807 max_depth = depth;
2808 if (depth > 63)
2809 depth = 63;
2810 depths[depth]++;
2811 }
2812
2813 for (depth = 0; depth < 64; depth++) {
2814 printf(" [%2d]: %8u", depth, depths[depth]);
2815 if ((depth & 3) == 3)
2816 printf("\n");
2817 if (depth == max_depth)
2818 break;
2819 }
2820 if ((depth & 3) != 3)
2821 printf("\n");
2822 printf("Max depth found was %d\n", max_depth);
2823 }
2824 #endif /* DEBUG */
2825
2826 #if defined(PMAPCHECK) || defined(DEBUG)
2827 void
pmap_pvo_verify(void)2828 pmap_pvo_verify(void)
2829 {
2830 int ptegidx;
2831 int s;
2832
2833 s = splvm();
2834 for (ptegidx = 0; ptegidx < pmap_pteg_cnt; ptegidx++) {
2835 struct pvo_entry *pvo;
2836 TAILQ_FOREACH(pvo, &pmap_pvo_table[ptegidx], pvo_olink) {
2837 if ((uintptr_t) pvo >= PMAP_DIRECT_MAPPED_LEN)
2838 panic("pmap_pvo_verify: invalid pvo %p "
2839 "on list %#x", pvo, ptegidx);
2840 pmap_pvo_check(pvo);
2841 }
2842 }
2843 splx(s);
2844 }
2845 #endif /* PMAPCHECK */
2846
2847 /*
2848 * Queue for unmanaged pages, used before uvm.page_init_done.
2849 * Reuse when pool shortage; see pmap_pool_alloc() below.
2850 */
2851 struct pup {
2852 SIMPLEQ_ENTRY(pup) pup_link;
2853 };
2854 SIMPLEQ_HEAD(pup_head, pup);
2855 static struct pup_head pup_head = SIMPLEQ_HEAD_INITIALIZER(pup_head);
2856
2857 static struct pup *
pmap_alloc_unmanaged(void)2858 pmap_alloc_unmanaged(void)
2859 {
2860 struct pup *pup;
2861 register_t msr;
2862
2863 PMAP_LOCK();
2864 msr = pmap_interrupts_off();
2865 pup = SIMPLEQ_FIRST(&pup_head);
2866 if (pup != NULL)
2867 SIMPLEQ_REMOVE_HEAD(&pup_head, pup_link);
2868 pmap_interrupts_restore(msr);
2869 PMAP_UNLOCK();
2870 return pup;
2871 }
2872
2873 static void
pmap_free_unmanaged(struct pup * pup)2874 pmap_free_unmanaged(struct pup *pup)
2875 {
2876 register_t msr;
2877
2878 PMAP_LOCK();
2879 msr = pmap_interrupts_off();
2880 SIMPLEQ_INSERT_HEAD(&pup_head, pup, pup_link);
2881 pmap_interrupts_restore(msr);
2882 PMAP_UNLOCK();
2883 }
2884
2885 void *
pmap_pool_alloc(struct pool * pp,int flags)2886 pmap_pool_alloc(struct pool *pp, int flags)
2887 {
2888 struct vm_page *pg;
2889 paddr_t pa;
2890
2891 if (__predict_false(!uvm.page_init_done))
2892 return (void *)uvm_pageboot_alloc(PAGE_SIZE);
2893
2894 retry:
2895 pg = uvm_pagealloc_strat(NULL /*obj*/, 0 /*off*/, NULL /*anon*/,
2896 UVM_PGA_USERESERVE /*flags*/, UVM_PGA_STRAT_ONLY /*strat*/,
2897 VM_FREELIST_DIRECT_MAPPED /*free_list*/);
2898 if (__predict_false(pg == NULL)) {
2899 void *va = pmap_alloc_unmanaged();
2900 if (va != NULL)
2901 return va;
2902
2903 if ((flags & PR_WAITOK) == 0)
2904 return NULL;
2905 uvm_wait("plpg");
2906 goto retry;
2907 }
2908 KDASSERT(VM_PAGE_TO_PHYS(pg) == (uintptr_t)VM_PAGE_TO_PHYS(pg));
2909 pa = VM_PAGE_TO_PHYS(pg);
2910 return (void *)(uintptr_t)pa;
2911 }
2912
2913 void
pmap_pool_free(struct pool * pp,void * va)2914 pmap_pool_free(struct pool *pp, void *va)
2915 {
2916 struct vm_page *pg;
2917
2918 pg = PHYS_TO_VM_PAGE((paddr_t)va);
2919 if (__predict_false(pg == NULL)) {
2920 pmap_free_unmanaged(va);
2921 return;
2922 }
2923 uvm_pagefree(pg);
2924 }
2925
2926 /*
2927 * This routine in bootstraping to steal to-be-managed memory (which will
2928 * then be unmanaged). We use it to grab from the first PMAP_DIRECT_MAPPED_LEN
2929 * for our pmap needs and above it for other stuff.
2930 */
2931 vaddr_t
pmap_steal_memory(vsize_t vsize,vaddr_t * vstartp,vaddr_t * vendp)2932 pmap_steal_memory(vsize_t vsize, vaddr_t *vstartp, vaddr_t *vendp)
2933 {
2934 vsize_t size;
2935 vaddr_t va;
2936 paddr_t start, end, pa = 0;
2937 int npgs, freelist;
2938 uvm_physseg_t bank;
2939
2940 if (uvm.page_init_done == true)
2941 panic("pmap_steal_memory: called _after_ bootstrap");
2942
2943 *vstartp = VM_MIN_KERNEL_ADDRESS;
2944 *vendp = VM_MAX_KERNEL_ADDRESS;
2945
2946 size = round_page(vsize);
2947 npgs = atop(size);
2948
2949 /*
2950 * PA 0 will never be among those given to UVM so we can use it
2951 * to indicate we couldn't steal any memory.
2952 */
2953
2954 for (bank = uvm_physseg_get_first();
2955 uvm_physseg_valid_p(bank);
2956 bank = uvm_physseg_get_next(bank)) {
2957
2958 freelist = uvm_physseg_get_free_list(bank);
2959 start = uvm_physseg_get_start(bank);
2960 end = uvm_physseg_get_end(bank);
2961
2962 if (freelist == VM_FREELIST_DIRECT_MAPPED &&
2963 (end - start) >= npgs) {
2964 pa = ptoa(start);
2965 break;
2966 }
2967 }
2968
2969 if (pa == 0)
2970 panic("pmap_steal_memory: no approriate memory to steal!");
2971
2972 uvm_physseg_unplug(start, npgs);
2973
2974 va = (vaddr_t) pa;
2975 memset((void *) va, 0, size);
2976 pmap_pages_stolen += npgs;
2977 #ifdef DEBUG
2978 if (pmapdebug && npgs > 1) {
2979 u_int cnt = 0;
2980 for (bank = uvm_physseg_get_first();
2981 uvm_physseg_valid_p(bank);
2982 bank = uvm_physseg_get_next(bank)) {
2983 cnt += uvm_physseg_get_avail_end(bank) - uvm_physseg_get_avail_start(bank);
2984 }
2985 printf("pmap_steal_memory: stole %u (total %u) pages (%u left)\n",
2986 npgs, pmap_pages_stolen, cnt);
2987 }
2988 #endif
2989
2990 return va;
2991 }
2992
2993 /*
2994 * Find a chuck of memory with right size and alignment.
2995 */
2996 paddr_t
pmap_boot_find_memory(psize_t size,psize_t alignment,int at_end)2997 pmap_boot_find_memory(psize_t size, psize_t alignment, int at_end)
2998 {
2999 struct mem_region *mp;
3000 paddr_t s, e;
3001 int i, j;
3002
3003 size = round_page(size);
3004
3005 DPRINTFN(BOOT,
3006 "pmap_boot_find_memory: size=%#" _PRIxpa ", alignment=%#" _PRIxpa ", at_end=%d",
3007 size, alignment, at_end);
3008
3009 if (alignment < PAGE_SIZE || (alignment & (alignment-1)) != 0)
3010 panic("pmap_boot_find_memory: invalid alignment %#" _PRIxpa,
3011 alignment);
3012
3013 if (at_end) {
3014 if (alignment != PAGE_SIZE)
3015 panic("pmap_boot_find_memory: invalid ending "
3016 "alignment %#" _PRIxpa, alignment);
3017
3018 for (mp = &avail[avail_cnt-1]; mp >= avail; mp--) {
3019 s = mp->start + mp->size - size;
3020 if (s >= mp->start && mp->size >= size) {
3021 DPRINTFN(BOOT, ": %#" _PRIxpa "\n", s);
3022 DPRINTFN(BOOT,
3023 "pmap_boot_find_memory: b-avail[%d] start "
3024 "%#" _PRIxpa " size %#" _PRIxpa "\n", mp - avail,
3025 mp->start, mp->size);
3026 mp->size -= size;
3027 DPRINTFN(BOOT,
3028 "pmap_boot_find_memory: a-avail[%d] start "
3029 "%#" _PRIxpa " size %#" _PRIxpa "\n", mp - avail,
3030 mp->start, mp->size);
3031 return s;
3032 }
3033 }
3034 panic("pmap_boot_find_memory: no available memory");
3035 }
3036
3037 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3038 s = (mp->start + alignment - 1) & ~(alignment-1);
3039 e = s + size;
3040
3041 /*
3042 * Is the calculated region entirely within the region?
3043 */
3044 if (s < mp->start || e > mp->start + mp->size)
3045 continue;
3046
3047 DPRINTFN(BOOT, ": %#" _PRIxpa "\n", s);
3048 if (s == mp->start) {
3049 /*
3050 * If the block starts at the beginning of region,
3051 * adjust the size & start. (the region may now be
3052 * zero in length)
3053 */
3054 DPRINTFN(BOOT,
3055 "pmap_boot_find_memory: b-avail[%d] start "
3056 "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size);
3057 mp->start += size;
3058 mp->size -= size;
3059 DPRINTFN(BOOT,
3060 "pmap_boot_find_memory: a-avail[%d] start "
3061 "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size);
3062 } else if (e == mp->start + mp->size) {
3063 /*
3064 * If the block starts at the beginning of region,
3065 * adjust only the size.
3066 */
3067 DPRINTFN(BOOT,
3068 "pmap_boot_find_memory: b-avail[%d] start "
3069 "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size);
3070 mp->size -= size;
3071 DPRINTFN(BOOT,
3072 "pmap_boot_find_memory: a-avail[%d] start "
3073 "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size);
3074 } else {
3075 /*
3076 * Block is in the middle of the region, so we
3077 * have to split it in two.
3078 */
3079 for (j = avail_cnt; j > i + 1; j--) {
3080 avail[j] = avail[j-1];
3081 }
3082 DPRINTFN(BOOT,
3083 "pmap_boot_find_memory: b-avail[%d] start "
3084 "%#" _PRIxpa " size %#" _PRIxpa "\n", i, mp->start, mp->size);
3085 mp[1].start = e;
3086 mp[1].size = mp[0].start + mp[0].size - e;
3087 mp[0].size = s - mp[0].start;
3088 avail_cnt++;
3089 for (; i < avail_cnt; i++) {
3090 DPRINTFN(BOOT,
3091 "pmap_boot_find_memory: a-avail[%d] "
3092 "start %#" _PRIxpa " size %#" _PRIxpa "\n", i,
3093 avail[i].start, avail[i].size);
3094 }
3095 }
3096 KASSERT(s == (uintptr_t) s);
3097 return s;
3098 }
3099 panic("pmap_boot_find_memory: not enough memory for "
3100 "%#" _PRIxpa "/%#" _PRIxpa " allocation?", size, alignment);
3101 }
3102
3103 /* XXXSL: we dont have any BATs to do this, map in Segment 0 1:1 using page tables */
3104 #if defined (PMAP_OEA64_BRIDGE)
3105 int
pmap_setup_segment0_map(int use_large_pages,...)3106 pmap_setup_segment0_map(int use_large_pages, ...)
3107 {
3108 vaddr_t va, va_end;
3109
3110 register_t pte_lo = 0x0;
3111 int ptegidx = 0;
3112 struct pte pte;
3113 va_list ap;
3114
3115 /* Coherent + Supervisor RW, no user access */
3116 pte_lo = PTE_M;
3117
3118 /* XXXSL
3119 * Map in 1st segment 1:1, we'll be careful not to spill kernel entries later,
3120 * these have to take priority.
3121 */
3122 for (va = 0x0; va < SEGMENT_LENGTH; va += 0x1000) {
3123 ptegidx = va_to_pteg(pmap_kernel(), va);
3124 pmap_pte_create(&pte, pmap_kernel(), va, va | pte_lo);
3125 (void)pmap_pte_insert(ptegidx, &pte);
3126 }
3127
3128 va_start(ap, use_large_pages);
3129 while (1) {
3130 paddr_t pa;
3131 size_t size;
3132
3133 va = va_arg(ap, vaddr_t);
3134
3135 if (va == 0)
3136 break;
3137
3138 pa = va_arg(ap, paddr_t);
3139 size = va_arg(ap, size_t);
3140
3141 for (va_end = va + size; va < va_end; va += 0x1000, pa += 0x1000) {
3142 #if 0
3143 printf("%s: Inserting: va: %#" _PRIxva ", pa: %#" _PRIxpa "\n", __func__, va, pa);
3144 #endif
3145 ptegidx = va_to_pteg(pmap_kernel(), va);
3146 pmap_pte_create(&pte, pmap_kernel(), va, pa | pte_lo);
3147 (void)pmap_pte_insert(ptegidx, &pte);
3148 }
3149 }
3150 va_end(ap);
3151
3152 TLBSYNC();
3153 SYNC();
3154 return (0);
3155 }
3156 #endif /* PMAP_OEA64_BRIDGE */
3157
3158 /*
3159 * Set up the bottom level of the data structures necessary for the kernel
3160 * to manage memory. MMU hardware is programmed in pmap_bootstrap2().
3161 */
3162 void
pmap_bootstrap1(paddr_t kernelstart,paddr_t kernelend)3163 pmap_bootstrap1(paddr_t kernelstart, paddr_t kernelend)
3164 {
3165 struct mem_region *mp, tmp;
3166 paddr_t s, e;
3167 psize_t size;
3168 int i, j;
3169
3170 /*
3171 * Get memory.
3172 */
3173 mem_regions(&mem, &avail);
3174 #if defined(DEBUG)
3175 if (pmapdebug & PMAPDEBUG_BOOT) {
3176 printf("pmap_bootstrap: memory configuration:\n");
3177 for (mp = mem; mp->size; mp++) {
3178 printf("pmap_bootstrap: mem start %#" _PRIxpa " size %#" _PRIxpa "\n",
3179 mp->start, mp->size);
3180 }
3181 for (mp = avail; mp->size; mp++) {
3182 printf("pmap_bootstrap: avail start %#" _PRIxpa " size %#" _PRIxpa "\n",
3183 mp->start, mp->size);
3184 }
3185 }
3186 #endif
3187
3188 /*
3189 * Find out how much physical memory we have and in how many chunks.
3190 */
3191 for (mem_cnt = 0, mp = mem; mp->size; mp++) {
3192 if (mp->start >= pmap_memlimit)
3193 continue;
3194 if (mp->start + mp->size > pmap_memlimit) {
3195 size = pmap_memlimit - mp->start;
3196 physmem += btoc(size);
3197 } else {
3198 physmem += btoc(mp->size);
3199 }
3200 mem_cnt++;
3201 }
3202
3203 /*
3204 * Count the number of available entries.
3205 */
3206 for (avail_cnt = 0, mp = avail; mp->size; mp++)
3207 avail_cnt++;
3208
3209 /*
3210 * Page align all regions.
3211 */
3212 kernelstart = trunc_page(kernelstart);
3213 kernelend = round_page(kernelend);
3214 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3215 s = round_page(mp->start);
3216 mp->size -= (s - mp->start);
3217 mp->size = trunc_page(mp->size);
3218 mp->start = s;
3219 e = mp->start + mp->size;
3220
3221 DPRINTFN(BOOT,
3222 "pmap_bootstrap: b-avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3223 i, mp->start, mp->size);
3224
3225 /*
3226 * Don't allow the end to run beyond our artificial limit
3227 */
3228 if (e > pmap_memlimit)
3229 e = pmap_memlimit;
3230
3231 /*
3232 * Is this region empty or strange? skip it.
3233 */
3234 if (e <= s) {
3235 mp->start = 0;
3236 mp->size = 0;
3237 continue;
3238 }
3239
3240 /*
3241 * Does this overlap the beginning of kernel?
3242 * Does extend past the end of the kernel?
3243 */
3244 else if (s < kernelstart && e > kernelstart) {
3245 if (e > kernelend) {
3246 avail[avail_cnt].start = kernelend;
3247 avail[avail_cnt].size = e - kernelend;
3248 avail_cnt++;
3249 }
3250 mp->size = kernelstart - s;
3251 }
3252 /*
3253 * Check whether this region overlaps the end of the kernel.
3254 */
3255 else if (s < kernelend && e > kernelend) {
3256 mp->start = kernelend;
3257 mp->size = e - kernelend;
3258 }
3259 /*
3260 * Look whether this regions is completely inside the kernel.
3261 * Nuke it if it does.
3262 */
3263 else if (s >= kernelstart && e <= kernelend) {
3264 mp->start = 0;
3265 mp->size = 0;
3266 }
3267 /*
3268 * If the user imposed a memory limit, enforce it.
3269 */
3270 else if (s >= pmap_memlimit) {
3271 mp->start = -PAGE_SIZE; /* let's know why */
3272 mp->size = 0;
3273 }
3274 else {
3275 mp->start = s;
3276 mp->size = e - s;
3277 }
3278 DPRINTFN(BOOT,
3279 "pmap_bootstrap: a-avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3280 i, mp->start, mp->size);
3281 }
3282
3283 /*
3284 * Move (and uncount) all the null return to the end.
3285 */
3286 for (mp = avail, i = 0; i < avail_cnt; i++, mp++) {
3287 if (mp->size == 0) {
3288 tmp = avail[i];
3289 avail[i] = avail[--avail_cnt];
3290 avail[avail_cnt] = avail[i];
3291 }
3292 }
3293
3294 /*
3295 * (Bubble)sort them into ascending order.
3296 */
3297 for (i = 0; i < avail_cnt; i++) {
3298 for (j = i + 1; j < avail_cnt; j++) {
3299 if (avail[i].start > avail[j].start) {
3300 tmp = avail[i];
3301 avail[i] = avail[j];
3302 avail[j] = tmp;
3303 }
3304 }
3305 }
3306
3307 /*
3308 * Make sure they don't overlap.
3309 */
3310 for (mp = avail, i = 0; i < avail_cnt - 1; i++, mp++) {
3311 if (mp[0].start + mp[0].size > mp[1].start) {
3312 mp[0].size = mp[1].start - mp[0].start;
3313 }
3314 DPRINTFN(BOOT,
3315 "pmap_bootstrap: avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3316 i, mp->start, mp->size);
3317 }
3318 DPRINTFN(BOOT,
3319 "pmap_bootstrap: avail[%d] start %#" _PRIxpa " size %#" _PRIxpa "\n",
3320 i, mp->start, mp->size);
3321
3322 #ifdef PTEGCOUNT
3323 pmap_pteg_cnt = PTEGCOUNT;
3324 #else /* PTEGCOUNT */
3325
3326 pmap_pteg_cnt = 0x1000;
3327
3328 while (pmap_pteg_cnt < physmem)
3329 pmap_pteg_cnt <<= 1;
3330
3331 pmap_pteg_cnt >>= 1;
3332 #endif /* PTEGCOUNT */
3333
3334 #ifdef DEBUG
3335 DPRINTFN(BOOT, "pmap_pteg_cnt: 0x%x\n", pmap_pteg_cnt);
3336 #endif
3337
3338 /*
3339 * Find suitably aligned memory for PTEG hash table.
3340 */
3341 size = pmap_pteg_cnt * sizeof(struct pteg);
3342 pmap_pteg_table = (void *)(uintptr_t) pmap_boot_find_memory(size, size, 0);
3343
3344 #ifdef DEBUG
3345 DPRINTFN(BOOT,
3346 "PTEG cnt: 0x%x HTAB size: 0x%08x bytes, address: %p\n", pmap_pteg_cnt, (unsigned int)size, pmap_pteg_table);
3347 #endif
3348
3349
3350 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
3351 if ( (uintptr_t) pmap_pteg_table + size > PMAP_DIRECT_MAPPED_LEN)
3352 panic("pmap_bootstrap: pmap_pteg_table end (%p + %#" _PRIxpa ") > PMAP_DIRECT_MAPPED_LEN",
3353 pmap_pteg_table, size);
3354 #endif
3355
3356 memset(__UNVOLATILE(pmap_pteg_table), 0,
3357 pmap_pteg_cnt * sizeof(struct pteg));
3358 pmap_pteg_mask = pmap_pteg_cnt - 1;
3359
3360 /*
3361 * We cannot do pmap_steal_memory here since UVM hasn't been loaded
3362 * with pages. So we just steal them before giving them to UVM.
3363 */
3364 size = sizeof(pmap_pvo_table[0]) * pmap_pteg_cnt;
3365 pmap_pvo_table = (void *)(uintptr_t) pmap_boot_find_memory(size, PAGE_SIZE, 0);
3366 #if defined(DIAGNOSTIC) || defined(DEBUG) || defined(PMAPCHECK)
3367 if ( (uintptr_t) pmap_pvo_table + size > PMAP_DIRECT_MAPPED_LEN)
3368 panic("pmap_bootstrap: pmap_pvo_table end (%p + %#" _PRIxpa ") > PMAP_DIRECT_MAPPED_LEN",
3369 pmap_pvo_table, size);
3370 #endif
3371
3372 for (i = 0; i < pmap_pteg_cnt; i++)
3373 TAILQ_INIT(&pmap_pvo_table[i]);
3374
3375 #ifndef MSGBUFADDR
3376 /*
3377 * Allocate msgbuf in high memory.
3378 */
3379 msgbuf_paddr = pmap_boot_find_memory(MSGBUFSIZE, PAGE_SIZE, 1);
3380 #endif
3381
3382 for (mp = avail, i = 0; i < avail_cnt; mp++, i++) {
3383 paddr_t pfstart = atop(mp->start);
3384 paddr_t pfend = atop(mp->start + mp->size);
3385 if (mp->size == 0)
3386 continue;
3387 if (mp->start + mp->size <= PMAP_DIRECT_MAPPED_LEN) {
3388 uvm_page_physload(pfstart, pfend, pfstart, pfend,
3389 VM_FREELIST_DIRECT_MAPPED);
3390 } else if (mp->start >= PMAP_DIRECT_MAPPED_LEN) {
3391 uvm_page_physload(pfstart, pfend, pfstart, pfend,
3392 VM_FREELIST_DEFAULT);
3393 } else {
3394 pfend = atop(PMAP_DIRECT_MAPPED_LEN);
3395 uvm_page_physload(pfstart, pfend, pfstart, pfend,
3396 VM_FREELIST_DIRECT_MAPPED);
3397 pfstart = atop(PMAP_DIRECT_MAPPED_LEN);
3398 pfend = atop(mp->start + mp->size);
3399 uvm_page_physload(pfstart, pfend, pfstart, pfend,
3400 VM_FREELIST_DEFAULT);
3401 }
3402 }
3403
3404 /*
3405 * Make sure kernel vsid is allocated as well as VSID 0.
3406 */
3407 pmap_vsid_bitmap[(KERNEL_VSIDBITS & (NPMAPS-1)) / VSID_NBPW]
3408 |= 1 << (KERNEL_VSIDBITS % VSID_NBPW);
3409 pmap_vsid_bitmap[(PHYSMAP_VSIDBITS & (NPMAPS-1)) / VSID_NBPW]
3410 |= 1 << (PHYSMAP_VSIDBITS % VSID_NBPW);
3411 pmap_vsid_bitmap[0] |= 1;
3412
3413 /*
3414 * Initialize kernel pmap.
3415 */
3416 #if defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
3417 for (i = 0; i < 16; i++) {
3418 pmap_kernel()->pm_sr[i] = KERNELN_SEGMENT(i)|SR_PRKEY;
3419 }
3420 pmap_kernel()->pm_vsid = KERNEL_VSIDBITS;
3421
3422 pmap_kernel()->pm_sr[KERNEL_SR] = KERNEL_SEGMENT|SR_SUKEY|SR_PRKEY;
3423 #ifdef KERNEL2_SR
3424 pmap_kernel()->pm_sr[KERNEL2_SR] = KERNEL2_SEGMENT|SR_SUKEY|SR_PRKEY;
3425 #endif
3426 #endif /* PMAP_OEA || PMAP_OEA64_BRIDGE */
3427
3428 #if defined(PMAP_OEA) && defined(PPC_OEA601)
3429 if ((MFPVR() >> 16) == MPC601) {
3430 for (i = 0; i < 16; i++) {
3431 if (iosrtable[i] & SR601_T) {
3432 pmap_kernel()->pm_sr[i] = iosrtable[i];
3433 }
3434 }
3435 }
3436 #endif /* PMAP_OEA && PPC_OEA601 */
3437
3438 #ifdef ALTIVEC
3439 pmap_use_altivec = cpu_altivec;
3440 #endif
3441
3442 #ifdef DEBUG
3443 if (pmapdebug & PMAPDEBUG_BOOT) {
3444 u_int cnt;
3445 uvm_physseg_t bank;
3446 char pbuf[9];
3447 for (cnt = 0, bank = uvm_physseg_get_first();
3448 uvm_physseg_valid_p(bank);
3449 bank = uvm_physseg_get_next(bank)) {
3450 cnt += uvm_physseg_get_avail_end(bank) -
3451 uvm_physseg_get_avail_start(bank);
3452 printf("pmap_bootstrap: vm_physmem[%d]=%#" _PRIxpa "-%#" _PRIxpa "/%#" _PRIxpa "\n",
3453 bank,
3454 ptoa(uvm_physseg_get_avail_start(bank)),
3455 ptoa(uvm_physseg_get_avail_end(bank)),
3456 ptoa(uvm_physseg_get_avail_end(bank) - uvm_physseg_get_avail_start(bank)));
3457 }
3458 format_bytes(pbuf, sizeof(pbuf), ptoa((u_int64_t) cnt));
3459 printf("pmap_bootstrap: UVM memory = %s (%u pages)\n",
3460 pbuf, cnt);
3461 }
3462 #endif
3463
3464 pool_init(&pmap_pvo_pool, sizeof(struct pvo_entry),
3465 PMAP_PVO_ENTRY_ALIGN, 0, 0, "pmap_pvopl",
3466 &pmap_pool_allocator, IPL_VM);
3467
3468 pool_setlowat(&pmap_pvo_pool, 1008);
3469
3470 pool_init(&pmap_pool, sizeof(struct pmap),
3471 __alignof(struct pmap), 0, 0, "pmap_pl",
3472 &pmap_pool_allocator, IPL_NONE);
3473
3474 #if defined(PMAP_OEA64_BRIDGE)
3475 {
3476 struct pmap *pm = pmap_kernel();
3477 uvm_physseg_t bank;
3478 paddr_t pa;
3479 struct pte pt;
3480 unsigned int ptegidx;
3481
3482 for (bank = uvm_physseg_get_first();
3483 uvm_physseg_valid_p(bank);
3484 bank = uvm_physseg_get_next(bank)) {
3485 if (uvm_physseg_get_free_list(bank) !=
3486 VM_FREELIST_DIRECT_MAPPED)
3487 continue;
3488 for (pa = uimax(ptoa(uvm_physseg_get_avail_start(bank)),
3489 SEGMENT_LENGTH);
3490 pa < ptoa(uvm_physseg_get_avail_end(bank));
3491 pa += PAGE_SIZE) {
3492 ptegidx = va_to_pteg(pm, pa);
3493 pmap_pte_create(&pt, pm, pa, pa | PTE_M);
3494 pmap_pte_insert(ptegidx, &pt);
3495 }
3496 }
3497 }
3498 #endif
3499
3500 #if defined(PMAP_NEED_MAPKERNEL)
3501 {
3502 struct pmap *pm = pmap_kernel();
3503 #if defined(PMAP_NEED_FULL_MAPKERNEL)
3504 extern int etext[], kernel_text[];
3505 vaddr_t va, va_etext = (paddr_t) etext;
3506 #endif
3507 paddr_t pa, pa_end;
3508 register_t sr;
3509 struct pte pt;
3510 unsigned int ptegidx;
3511 int bank;
3512
3513 sr = PHYSMAPN_SEGMENT(0) | SR_SUKEY|SR_PRKEY;
3514 pm->pm_sr[0] = sr;
3515
3516 for (bank = 0; bank < vm_nphysseg; bank++) {
3517 pa_end = ptoa(VM_PHYSMEM_PTR(bank)->avail_end);
3518 pa = ptoa(VM_PHYSMEM_PTR(bank)->avail_start);
3519 for (; pa < pa_end; pa += PAGE_SIZE) {
3520 ptegidx = va_to_pteg(pm, pa);
3521 pmap_pte_create(&pt, pm, pa, pa | PTE_M|PTE_BW);
3522 pmap_pte_insert(ptegidx, &pt);
3523 }
3524 }
3525
3526 #if defined(PMAP_NEED_FULL_MAPKERNEL)
3527 va = (vaddr_t) kernel_text;
3528
3529 for (pa = kernelstart; va < va_etext;
3530 pa += PAGE_SIZE, va += PAGE_SIZE) {
3531 ptegidx = va_to_pteg(pm, va);
3532 pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BR);
3533 pmap_pte_insert(ptegidx, &pt);
3534 }
3535
3536 for (; pa < kernelend;
3537 pa += PAGE_SIZE, va += PAGE_SIZE) {
3538 ptegidx = va_to_pteg(pm, va);
3539 pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3540 pmap_pte_insert(ptegidx, &pt);
3541 }
3542
3543 for (va = 0, pa = 0; va < kernelstart;
3544 pa += PAGE_SIZE, va += PAGE_SIZE) {
3545 ptegidx = va_to_pteg(pm, va);
3546 if (va < 0x3000)
3547 pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BR);
3548 else
3549 pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3550 pmap_pte_insert(ptegidx, &pt);
3551 }
3552 for (va = kernelend, pa = kernelend; va < SEGMENT_LENGTH;
3553 pa += PAGE_SIZE, va += PAGE_SIZE) {
3554 ptegidx = va_to_pteg(pm, va);
3555 pmap_pte_create(&pt, pm, va, pa | PTE_M|PTE_BW);
3556 pmap_pte_insert(ptegidx, &pt);
3557 }
3558 #endif /* PMAP_NEED_FULL_MAPKERNEL */
3559 }
3560 #endif /* PMAP_NEED_MAPKERNEL */
3561 }
3562
3563 /*
3564 * Using the data structures prepared in pmap_bootstrap1(), program
3565 * the MMU hardware.
3566 */
3567 void
pmap_bootstrap2(void)3568 pmap_bootstrap2(void)
3569 {
3570 #if defined(PMAP_OEA) || defined(PMAP_OEA64_BRIDGE)
3571 for (int i = 0; i < 16; i++) {
3572 __asm volatile("mtsrin %0,%1"
3573 :: "r"(pmap_kernel()->pm_sr[i]),
3574 "r"(i << ADDR_SR_SHFT));
3575 }
3576 #endif /* PMAP_OEA || PMAP_OEA64_BRIDGE */
3577
3578 #if defined(PMAP_OEA)
3579 __asm volatile("sync; mtsdr1 %0; isync"
3580 :
3581 : "r"((uintptr_t)pmap_pteg_table | (pmap_pteg_mask >> 10))
3582 : "memory");
3583 #elif defined(PMAP_OEA64) || defined(PMAP_OEA64_BRIDGE)
3584 __asm volatile("sync; mtsdr1 %0; isync"
3585 :
3586 : "r"((uintptr_t)pmap_pteg_table |
3587 (32 - __builtin_clz(pmap_pteg_mask >> 11)))
3588 : "memory");
3589 #endif
3590 tlbia();
3591
3592 #if defined(PMAPDEBUG)
3593 if (pmapdebug)
3594 pmap_print_mmuregs();
3595 #endif
3596 }
3597
3598 /*
3599 * This is not part of the defined PMAP interface and is specific to the
3600 * PowerPC architecture. This is called during initppc, before the system
3601 * is really initialized.
3602 */
3603 void
pmap_bootstrap(paddr_t kernelstart,paddr_t kernelend)3604 pmap_bootstrap(paddr_t kernelstart, paddr_t kernelend)
3605 {
3606 pmap_bootstrap1(kernelstart, kernelend);
3607 pmap_bootstrap2();
3608 }
3609