xref: /freebsd/sys/powerpc/booke/pmap.c (revision b83204edf101d2c8ad40005ca2fe9a76796cd4f1)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
5  * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN
20  * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
21  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
22  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
23  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
24  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
25  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
26  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  *
28  * Some hw specific parts of this pmap were derived or influenced
29  * by NetBSD's ibm4xx pmap module. More generic code is shared with
30  * a few other pmap modules from the FreeBSD tree.
31  */
32 
33  /*
34   * VM layout notes:
35   *
36   * Kernel and user threads run within one common virtual address space
37   * defined by AS=0.
38   *
39   * 32-bit pmap:
40   * Virtual address space layout:
41   * -----------------------------
42   * 0x0000_0000 - 0x7fff_ffff	: user process
43   * 0x8000_0000 - 0xbfff_ffff	: pmap_mapdev()-ed area (PCI/PCIE etc.)
44   * 0xc000_0000 - 0xc0ff_ffff	: kernel reserved
45   *   0xc000_0000 - data_end	: kernel code+data, env, metadata etc.
46   * 0xc100_0000 - 0xffff_ffff	: KVA
47   *   0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
48   *   0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
49   *   0xc200_4000 - 0xc200_8fff : guard page + kstack0
50   *   0xc200_9000 - 0xfeef_ffff	: actual free KVA space
51   *
52   * 64-bit pmap:
53   * Virtual address space layout:
54   * -----------------------------
55   * 0x0000_0000_0000_0000 - 0xbfff_ffff_ffff_ffff      : user process
56   *   0x0000_0000_0000_0000 - 0x8fff_ffff_ffff_ffff    : text, data, heap, maps, libraries
57   *   0x9000_0000_0000_0000 - 0xafff_ffff_ffff_ffff    : mmio region
58   *   0xb000_0000_0000_0000 - 0xbfff_ffff_ffff_ffff    : stack
59   * 0xc000_0000_0000_0000 - 0xcfff_ffff_ffff_ffff      : kernel reserved
60   *   0xc000_0000_0000_0000 - endkernel-1              : kernel code & data
61   *               endkernel - msgbufp-1                : flat device tree
62   *                 msgbufp - kernel_pdir-1            : message buffer
63   *             kernel_pdir - kernel_pp2d-1            : kernel page directory
64   *             kernel_pp2d - .                        : kernel pointers to page directory
65   *      pmap_zero_copy_min - crashdumpmap-1           : reserved for page zero/copy
66   *            crashdumpmap - ptbl_buf_pool_vabase-1   : reserved for ptbl bufs
67   *    ptbl_buf_pool_vabase - virtual_avail-1          : user page directories and page tables
68   *           virtual_avail - 0xcfff_ffff_ffff_ffff    : actual free KVA space
69   * 0xd000_0000_0000_0000 - 0xdfff_ffff_ffff_ffff      : coprocessor region
70   * 0xe000_0000_0000_0000 - 0xefff_ffff_ffff_ffff      : mmio region
71   * 0xf000_0000_0000_0000 - 0xffff_ffff_ffff_ffff      : direct map
72   *   0xf000_0000_0000_0000 - +Maxmem                  : physmem map
73   *                         - 0xffff_ffff_ffff_ffff    : device direct map
74   */
75 
76 #include <sys/cdefs.h>
77 #include "opt_ddb.h"
78 #include "opt_kstack_pages.h"
79 
80 #include <sys/param.h>
81 #include <sys/conf.h>
82 #include <sys/malloc.h>
83 #include <sys/ktr.h>
84 #include <sys/proc.h>
85 #include <sys/user.h>
86 #include <sys/queue.h>
87 #include <sys/systm.h>
88 #include <sys/kernel.h>
89 #include <sys/kerneldump.h>
90 #include <sys/limits.h>
91 #include <sys/linker.h>
92 #include <sys/msgbuf.h>
93 #include <sys/lock.h>
94 #include <sys/mutex.h>
95 #include <sys/rwlock.h>
96 #include <sys/sched.h>
97 #include <sys/smp.h>
98 #include <sys/vmmeter.h>
99 
100 #include <vm/vm.h>
101 #include <vm/vm_param.h>
102 #include <vm/vm_page.h>
103 #include <vm/vm_kern.h>
104 #include <vm/vm_pageout.h>
105 #include <vm/vm_extern.h>
106 #include <vm/vm_object.h>
107 #include <vm/vm_map.h>
108 #include <vm/vm_pager.h>
109 #include <vm/vm_phys.h>
110 #include <vm/vm_pagequeue.h>
111 #include <vm/vm_radix.h>
112 #include <vm/vm_dumpset.h>
113 #include <vm/uma.h>
114 
115 #include <machine/_inttypes.h>
116 #include <machine/cpu.h>
117 #include <machine/pcb.h>
118 #include <machine/platform.h>
119 
120 #include <machine/tlb.h>
121 #include <machine/spr.h>
122 #include <machine/md_var.h>
123 #include <machine/mmuvar.h>
124 #include <machine/pmap.h>
125 #include <machine/pte.h>
126 
127 #include <ddb/ddb.h>
128 
129 #define	SPARSE_MAPDEV
130 
131 /* Use power-of-two mappings in mmu_booke_mapdev(), to save entries. */
132 #define	POW2_MAPPINGS
133 
134 #ifdef  DEBUG
135 #define debugf(fmt, args...) printf(fmt, ##args)
136 #define	__debug_used
137 #else
138 #define debugf(fmt, args...)
139 #define	__debug_used	__unused
140 #endif
141 
142 #ifdef __powerpc64__
143 #define	PRI0ptrX	"016lx"
144 #else
145 #define	PRI0ptrX	"08x"
146 #endif
147 
148 #define TODO			panic("%s: not implemented", __func__);
149 
150 extern unsigned char _etext[];
151 extern unsigned char _end[];
152 
153 extern uint32_t *bootinfo;
154 
155 vm_paddr_t kernload;
156 vm_offset_t kernstart;
157 vm_size_t kernsize;
158 
159 /* Message buffer and tables. */
160 static vm_offset_t data_start;
161 static vm_size_t data_end;
162 
163 /* Phys/avail memory regions. */
164 static struct mem_region *availmem_regions;
165 static int availmem_regions_sz;
166 static struct mem_region *physmem_regions;
167 static int physmem_regions_sz;
168 
169 #ifndef __powerpc64__
170 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
171 static vm_offset_t zero_page_va;
172 static struct mtx zero_page_mutex;
173 
174 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
175 static vm_offset_t copy_page_src_va;
176 static vm_offset_t copy_page_dst_va;
177 static struct mtx copy_page_mutex;
178 #endif
179 
180 static struct mtx tlbivax_mutex;
181 static bool mmuv2;
182 
183 /**************************************************************************/
184 /* PMAP */
185 /**************************************************************************/
186 
187 static int mmu_booke_enter_locked(pmap_t, vm_offset_t, vm_page_t,
188     vm_prot_t, u_int flags, int8_t psind);
189 
190 unsigned int kptbl_min;		/* Index of the first kernel ptbl. */
191 static uma_zone_t ptbl_root_zone;
192 
193 /*
194  * If user pmap is processed with mmu_booke_remove and the resident count
195  * drops to 0, there are no more pages to remove, so we need not continue.
196  */
197 #define PMAP_REMOVE_DONE(pmap) \
198 	((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
199 
200 #if defined(COMPAT_FREEBSD32) || !defined(__powerpc64__)
201 extern int elf32_nxstack;
202 #endif
203 
204 /**************************************************************************/
205 /* TLB and TID handling */
206 /**************************************************************************/
207 
208 /* Translation ID busy table (dynamically allocated) */
209 static __inline void tid_set_busy(int cpu, int tid, pmap_t pmap);
210 static volatile pmap_t *tidbusy;
211 uint32_t tid_max;
212 
213 /*
214  * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
215  * core revisions and should be read from h/w registers during early config.
216  */
217 uint32_t tlb0_entries;
218 uint32_t tlb0_ways;
219 uint32_t tlb0_entries_per_way;
220 uint32_t tlb1_entries;
221 
222 #define TLB0_ENTRIES		(tlb0_entries)
223 #define TLB0_WAYS		(tlb0_ways)
224 #define TLB0_ENTRIES_PER_WAY	(tlb0_entries_per_way)
225 
226 #define TLB1_ENTRIES (tlb1_entries)
227 
228 static tlbtid_t tid_alloc(struct pmap *);
229 
230 #ifdef DDB
231 #ifdef __powerpc64__
232 static void tlb_print_entry(int, uint32_t, uint64_t, uint32_t, uint32_t);
233 #else
234 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
235 #endif
236 #endif
237 
238 static void tlb1_read_entry(tlb_entry_t *, unsigned int);
239 static void tlb1_write_entry(tlb_entry_t *, unsigned int);
240 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
241 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t, int);
242 
243 static __inline uint32_t tlb_calc_wimg(vm_paddr_t pa, vm_memattr_t ma);
244 
245 static vm_size_t tsize2size(unsigned int);
246 static unsigned int size2tsize(vm_size_t);
247 
248 static void set_mas4_defaults(void);
249 
250 static inline void tlb0_flush_entry(vm_offset_t);
251 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
252 
253 /**************************************************************************/
254 /* Page table management */
255 /**************************************************************************/
256 
257 static struct rwlock_padalign pvh_global_lock;
258 
259 /* Data for the pv entry allocation mechanism */
260 static uma_zone_t pvzone;
261 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
262 
263 #define PV_ENTRY_ZONE_MIN	2048	/* min pv entries in uma zone */
264 
265 #ifndef PMAP_SHPGPERPROC
266 #define PMAP_SHPGPERPROC	200
267 #endif
268 
269 static vm_paddr_t pte_vatopa(pmap_t, vm_offset_t);
270 static int pte_enter(pmap_t, vm_page_t, vm_offset_t, uint32_t, bool);
271 static int pte_remove(pmap_t, vm_offset_t, uint8_t);
272 static pte_t *pte_find(pmap_t, vm_offset_t);
273 static void kernel_pte_alloc(vm_offset_t, vm_offset_t);
274 
275 static pv_entry_t pv_alloc(void);
276 static void pv_free(pv_entry_t);
277 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
278 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
279 
280 static void booke_pmap_init_qpages(void);
281 
282 static inline void tlb_miss_lock(void);
283 static inline void tlb_miss_unlock(void);
284 
285 #ifdef SMP
286 extern tlb_entry_t __boot_tlb1[];
287 void pmap_bootstrap_ap(volatile uint32_t *);
288 #endif
289 
290 /*
291  * Kernel MMU interface
292  */
293 static void		mmu_booke_clear_modify(vm_page_t);
294 static void		mmu_booke_copy(pmap_t, pmap_t, vm_offset_t,
295     vm_size_t, vm_offset_t);
296 static void		mmu_booke_copy_page(vm_page_t, vm_page_t);
297 static void		mmu_booke_copy_pages(vm_page_t *,
298     vm_offset_t, vm_page_t *, vm_offset_t, int);
299 static int		mmu_booke_enter(pmap_t, vm_offset_t, vm_page_t,
300     vm_prot_t, u_int flags, int8_t psind);
301 static void		mmu_booke_enter_object(pmap_t, vm_offset_t, vm_offset_t,
302     vm_page_t, vm_prot_t);
303 static void		mmu_booke_enter_quick(pmap_t, vm_offset_t, vm_page_t,
304     vm_prot_t);
305 static vm_paddr_t	mmu_booke_extract(pmap_t, vm_offset_t);
306 static vm_page_t	mmu_booke_extract_and_hold(pmap_t, vm_offset_t,
307     vm_prot_t);
308 static void		mmu_booke_init(void);
309 static bool		mmu_booke_is_modified(vm_page_t);
310 static bool		mmu_booke_is_prefaultable(pmap_t, vm_offset_t);
311 static bool		mmu_booke_is_referenced(vm_page_t);
312 static int		mmu_booke_ts_referenced(vm_page_t);
313 static void		*mmu_booke_map(vm_offset_t *, vm_paddr_t, vm_paddr_t,
314     int);
315 static int		mmu_booke_mincore(pmap_t, vm_offset_t,
316     vm_paddr_t *);
317 static void		mmu_booke_object_init_pt(pmap_t, vm_offset_t,
318     vm_object_t, vm_pindex_t, vm_size_t);
319 static bool		mmu_booke_page_exists_quick(pmap_t, vm_page_t);
320 static void		mmu_booke_page_init(vm_page_t);
321 static int		mmu_booke_page_wired_mappings(vm_page_t);
322 static int		mmu_booke_pinit(pmap_t);
323 static void		mmu_booke_pinit0(pmap_t);
324 static void		mmu_booke_protect(pmap_t, vm_offset_t, vm_offset_t,
325     vm_prot_t);
326 static void		mmu_booke_qenter(void *, vm_page_t *, int);
327 static void		mmu_booke_qremove(void *, int);
328 static void		mmu_booke_release(pmap_t);
329 static void		mmu_booke_remove(pmap_t, vm_offset_t, vm_offset_t);
330 static void		mmu_booke_remove_all(vm_page_t);
331 static void		mmu_booke_remove_write(vm_page_t);
332 static void		mmu_booke_unwire(pmap_t, vm_offset_t, vm_offset_t);
333 static void		mmu_booke_zero_page(vm_page_t);
334 static void		mmu_booke_zero_page_area(vm_page_t, int, int);
335 static void		mmu_booke_activate(struct thread *);
336 static void		mmu_booke_deactivate(struct thread *);
337 static void		mmu_booke_bootstrap(vm_offset_t, vm_offset_t);
338 static void		*mmu_booke_mapdev(vm_paddr_t, vm_size_t);
339 static void		*mmu_booke_mapdev_attr(vm_paddr_t, vm_size_t, vm_memattr_t);
340 static void		mmu_booke_unmapdev(void *, vm_size_t);
341 static vm_paddr_t	mmu_booke_kextract(vm_offset_t);
342 static void		mmu_booke_kenter(vm_offset_t, vm_paddr_t);
343 static void		mmu_booke_kenter_attr(vm_offset_t, vm_paddr_t, vm_memattr_t);
344 static void		mmu_booke_kremove(vm_offset_t);
345 static int		mmu_booke_dev_direct_mapped(vm_paddr_t, vm_size_t);
346 static void		mmu_booke_sync_icache(pmap_t, vm_offset_t,
347     vm_size_t);
348 static void		mmu_booke_dumpsys_map(vm_paddr_t pa, size_t,
349     void **);
350 static void		mmu_booke_dumpsys_unmap(vm_paddr_t pa, size_t,
351     void *);
352 static void		mmu_booke_scan_init(void);
353 static void		*mmu_booke_quick_enter_page(vm_page_t m);
354 static void		mmu_booke_quick_remove_page(void *addr);
355 static int		mmu_booke_change_attr(void *addr,
356     vm_size_t sz, vm_memattr_t mode);
357 static int		mmu_booke_decode_kernel_ptr(vm_offset_t addr,
358     int *is_user, vm_offset_t *decoded_addr);
359 static void		mmu_booke_page_array_startup(long);
360 static bool mmu_booke_page_is_mapped(vm_page_t m);
361 static bool mmu_booke_ps_enabled(pmap_t pmap);
362 #ifdef __powerpc64__
363 static int		mmu_booke_growkernel(vm_offset_t);
364 #endif
365 
366 static struct pmap_funcs mmu_booke_methods = {
367 	/* pmap dispatcher interface */
368 	.clear_modify = mmu_booke_clear_modify,
369 	.copy = mmu_booke_copy,
370 	.copy_page = mmu_booke_copy_page,
371 	.copy_pages = mmu_booke_copy_pages,
372 	.enter = mmu_booke_enter,
373 	.enter_object = mmu_booke_enter_object,
374 	.enter_quick = mmu_booke_enter_quick,
375 	.extract = mmu_booke_extract,
376 	.extract_and_hold = mmu_booke_extract_and_hold,
377 	.init = mmu_booke_init,
378 	.is_modified = mmu_booke_is_modified,
379 	.is_prefaultable = mmu_booke_is_prefaultable,
380 	.is_referenced = mmu_booke_is_referenced,
381 	.ts_referenced = mmu_booke_ts_referenced,
382 	.map = mmu_booke_map,
383 	.mincore = mmu_booke_mincore,
384 	.object_init_pt = mmu_booke_object_init_pt,
385 	.page_exists_quick = mmu_booke_page_exists_quick,
386 	.page_init = mmu_booke_page_init,
387 	.page_wired_mappings =  mmu_booke_page_wired_mappings,
388 	.pinit = mmu_booke_pinit,
389 	.pinit0 = mmu_booke_pinit0,
390 	.protect = mmu_booke_protect,
391 	.qenter = mmu_booke_qenter,
392 	.qremove = mmu_booke_qremove,
393 	.release = mmu_booke_release,
394 	.remove = mmu_booke_remove,
395 	.remove_all = mmu_booke_remove_all,
396 	.remove_write = mmu_booke_remove_write,
397 	.sync_icache = mmu_booke_sync_icache,
398 	.unwire = mmu_booke_unwire,
399 	.zero_page = mmu_booke_zero_page,
400 	.zero_page_area = mmu_booke_zero_page_area,
401 	.activate = mmu_booke_activate,
402 	.deactivate = mmu_booke_deactivate,
403 	.quick_enter_page =  mmu_booke_quick_enter_page,
404 	.quick_remove_page =  mmu_booke_quick_remove_page,
405 	.page_array_startup = mmu_booke_page_array_startup,
406 	.page_is_mapped = mmu_booke_page_is_mapped,
407 	.ps_enabled = mmu_booke_ps_enabled,
408 #ifdef __powerpc64__
409 	.growkernel_nopanic = mmu_booke_growkernel,
410 #endif
411 
412 	/* Internal interfaces */
413 	.bootstrap = mmu_booke_bootstrap,
414 	.dev_direct_mapped = mmu_booke_dev_direct_mapped,
415 	.mapdev = mmu_booke_mapdev,
416 	.mapdev_attr = mmu_booke_mapdev_attr,
417 	.kenter = mmu_booke_kenter,
418 	.kenter_attr = mmu_booke_kenter_attr,
419 	.kextract = mmu_booke_kextract,
420 	.kremove = mmu_booke_kremove,
421 	.unmapdev = mmu_booke_unmapdev,
422 	.change_attr = mmu_booke_change_attr,
423 	.decode_kernel_ptr =  mmu_booke_decode_kernel_ptr,
424 
425 	/* dumpsys() support */
426 	.dumpsys_map_chunk = mmu_booke_dumpsys_map,
427 	.dumpsys_unmap_chunk = mmu_booke_dumpsys_unmap,
428 	.dumpsys_pa_init = mmu_booke_scan_init,
429 };
430 
431 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods);
432 
433 #ifdef __powerpc64__
434 #include "pmap_64.c"
435 #else
436 #include "pmap_32.c"
437 #endif
438 
439 static vm_offset_t tlb1_map_base = VM_MAPDEV_BASE;
440 
441 static __inline uint32_t
tlb_calc_wimg(vm_paddr_t pa,vm_memattr_t ma)442 tlb_calc_wimg(vm_paddr_t pa, vm_memattr_t ma)
443 {
444 	uint32_t attrib;
445 	int i;
446 
447 	if (ma != VM_MEMATTR_DEFAULT) {
448 		switch (ma) {
449 		case VM_MEMATTR_UNCACHEABLE:
450 			return (MAS2_I | MAS2_G);
451 		case VM_MEMATTR_WRITE_COMBINING:
452 		case VM_MEMATTR_WRITE_BACK:
453 		case VM_MEMATTR_PREFETCHABLE:
454 			return (MAS2_I);
455 		case VM_MEMATTR_WRITE_THROUGH:
456 			return (MAS2_W | MAS2_M);
457 		case VM_MEMATTR_CACHEABLE:
458 			return (MAS2_M);
459 		}
460 	}
461 
462 	/*
463 	 * Assume the page is cache inhibited and access is guarded unless
464 	 * it's in our available memory array.
465 	 */
466 	attrib = _TLB_ENTRY_IO;
467 	for (i = 0; i < physmem_regions_sz; i++) {
468 		if ((pa >= physmem_regions[i].mr_start) &&
469 		    (pa < (physmem_regions[i].mr_start +
470 		     physmem_regions[i].mr_size))) {
471 			attrib = _TLB_ENTRY_MEM;
472 			break;
473 		}
474 	}
475 
476 	return (attrib);
477 }
478 
479 static inline void
tlb_miss_lock(void)480 tlb_miss_lock(void)
481 {
482 #ifdef SMP
483 	struct pcpu *pc;
484 
485 	if (!smp_started)
486 		return;
487 
488 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
489 		if (pc != pcpup) {
490 			CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
491 			    "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke.tlb_lock);
492 
493 			KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
494 			    ("tlb_miss_lock: tried to lock self"));
495 
496 			tlb_lock(pc->pc_booke.tlb_lock);
497 
498 			CTR1(KTR_PMAP, "%s: locked", __func__);
499 		}
500 	}
501 #endif
502 }
503 
504 static inline void
tlb_miss_unlock(void)505 tlb_miss_unlock(void)
506 {
507 #ifdef SMP
508 	struct pcpu *pc;
509 
510 	if (!smp_started)
511 		return;
512 
513 	STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
514 		if (pc != pcpup) {
515 			CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
516 			    __func__, pc->pc_cpuid);
517 
518 			tlb_unlock(pc->pc_booke.tlb_lock);
519 
520 			CTR1(KTR_PMAP, "%s: unlocked", __func__);
521 		}
522 	}
523 #endif
524 }
525 
526 /* Return number of entries in TLB0. */
527 static __inline void
tlb0_get_tlbconf(void)528 tlb0_get_tlbconf(void)
529 {
530 	uint32_t tlb0_cfg;
531 
532 	tlb0_cfg = mfspr(SPR_TLB0CFG);
533 	tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
534 	tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
535 	tlb0_entries_per_way = tlb0_entries / tlb0_ways;
536 }
537 
538 /* Return number of entries in TLB1. */
539 static __inline void
tlb1_get_tlbconf(void)540 tlb1_get_tlbconf(void)
541 {
542 	uint32_t tlb1_cfg;
543 
544 	tlb1_cfg = mfspr(SPR_TLB1CFG);
545 	tlb1_entries = tlb1_cfg & TLBCFG_NENTRY_MASK;
546 }
547 
548 /**************************************************************************/
549 /* Page table related */
550 /**************************************************************************/
551 
552 /* Allocate pv_entry structure. */
553 pv_entry_t
pv_alloc(void)554 pv_alloc(void)
555 {
556 	pv_entry_t pv;
557 
558 	pv_entry_count++;
559 	if (pv_entry_count > pv_entry_high_water)
560 		pagedaemon_wakeup(0); /* XXX powerpc NUMA */
561 	pv = uma_zalloc(pvzone, M_NOWAIT);
562 
563 	return (pv);
564 }
565 
566 /* Free pv_entry structure. */
567 static __inline void
pv_free(pv_entry_t pve)568 pv_free(pv_entry_t pve)
569 {
570 
571 	pv_entry_count--;
572 	uma_zfree(pvzone, pve);
573 }
574 
575 /* Allocate and initialize pv_entry structure. */
576 static void
pv_insert(pmap_t pmap,vm_offset_t va,vm_page_t m)577 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
578 {
579 	pv_entry_t pve;
580 
581 	//int su = (pmap == kernel_pmap);
582 	//debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
583 	//	(u_int32_t)pmap, va, (u_int32_t)m);
584 
585 	pve = pv_alloc();
586 	if (pve == NULL)
587 		panic("pv_insert: no pv entries!");
588 
589 	pve->pv_pmap = pmap;
590 	pve->pv_va = va;
591 
592 	/* add to pv_list */
593 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
594 	rw_assert(&pvh_global_lock, RA_WLOCKED);
595 
596 	TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
597 
598 	//debugf("pv_insert: e\n");
599 }
600 
601 /* Destroy pv entry. */
602 static void
pv_remove(pmap_t pmap,vm_offset_t va,vm_page_t m)603 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
604 {
605 	pv_entry_t pve;
606 
607 	//int su = (pmap == kernel_pmap);
608 	//debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
609 
610 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
611 	rw_assert(&pvh_global_lock, RA_WLOCKED);
612 
613 	/* find pv entry */
614 	TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
615 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
616 			/* remove from pv_list */
617 			TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
618 			if (TAILQ_EMPTY(&m->md.pv_list))
619 				vm_page_aflag_clear(m, PGA_WRITEABLE);
620 
621 			/* free pv entry struct */
622 			pv_free(pve);
623 			break;
624 		}
625 	}
626 
627 	//debugf("pv_remove: e\n");
628 }
629 
630 /**************************************************************************/
631 /* PMAP related */
632 /**************************************************************************/
633 
634 /*
635  * This is called during booke_init, before the system is really initialized.
636  */
637 static void
mmu_booke_bootstrap(vm_offset_t start,vm_offset_t kernelend)638 mmu_booke_bootstrap(vm_offset_t start, vm_offset_t kernelend)
639 {
640 	vm_paddr_t phys_kernelend;
641 	struct mem_region *mp, *mp1;
642 	int cnt, i, j;
643 	vm_paddr_t s, e, sz;
644 	vm_paddr_t physsz, hwphyssz;
645 	u_int phys_avail_count __debug_used;
646 	vm_size_t kstack0_sz;
647 	vm_paddr_t kstack0_phys;
648 	vm_offset_t kstack0;
649 	uint32_t tid_bits;
650 	void *dpcpu;
651 
652 	debugf("mmu_booke_bootstrap: entered\n");
653 
654 	if ((mfspr(SPR_MMUCFG) & MMUCFG_MAVN_M) > 0)
655 		mmuv2 = true;
656 
657 	/* Set interesting system properties */
658 #ifdef __powerpc64__
659 	hw_direct_map = 1;
660 #else
661 	hw_direct_map = 0;
662 #endif
663 #if defined(COMPAT_FREEBSD32) || !defined(__powerpc64__)
664 	elf32_nxstack = 1;
665 #endif
666 
667 	/* Initialize invalidation mutex */
668 	mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
669 
670 	/* Read TLB0 size and associativity. */
671 	tlb0_get_tlbconf();
672 
673 	/*
674 	 * Calculate the max TID from the hardware.  Allow overriding with a
675 	 * tunable.  The tunable should be a power of 2.
676 	 */
677 	tid_bits = ((mfspr(SPR_MMUCFG) & MMUCFG_PIDSIZE_M) >> MMUCFG_PIDSIZE_S);
678 	TUNABLE_INT_FETCH("machdep.tid_max", &tid_max);
679 	if (tid_max <= 0)
680 		tid_max = INT_MAX;
681 	else
682 		tid_max = 1 << ilog2(tid_max);
683 	tid_max = min((1 << tid_bits), tid_max) - 1;
684 
685 	/*
686 	 * Align kernel start and end address (kernel image).
687 	 * Note that kernel end does not necessarily relate to kernsize.
688 	 * kernsize is the size of the kernel that is actually mapped.
689 	 */
690 	data_start = round_page(kernelend);
691 	data_end = data_start;
692 
693 	tidbusy = (void *)data_end;
694 	printf("tidbusy at %p\n", tidbusy);
695 	printf("tidmax = %d\n", tid_max);
696 	data_end += round_page(sizeof(pmap_t) * MAXCPU * (tid_max + 1));
697 
698 	/* Allocate the dynamic per-cpu area. */
699 	dpcpu = (void *)data_end;
700 	data_end += DPCPU_SIZE;
701 
702 	/* Allocate space for the message buffer. */
703 	msgbufp = (struct msgbuf *)data_end;
704 	data_end += msgbufsize;
705 	debugf(" msgbufp at 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
706 	    (uintptr_t)msgbufp, data_end);
707 
708 	data_end = round_page(data_end);
709 
710 	/* Retrieve phys/avail mem regions */
711 	mem_regions(&physmem_regions, &physmem_regions_sz,
712 	    &availmem_regions, &availmem_regions_sz);
713 
714 	if (PHYS_AVAIL_ENTRIES < availmem_regions_sz)
715 		panic("mmu_booke_bootstrap: phys_avail too small");
716 
717 	vm_page_array = (vm_page_t)data_end;
718 	/*
719 	 * Get a rough idea (upper bound) on the size of the page array.  The
720 	 * vm_page_array will not handle any more pages than we have in the
721 	 * avail_regions array, and most likely much less.
722 	 */
723 	sz = 0;
724 	for (mp = availmem_regions; mp->mr_size; mp++) {
725 		sz += mp->mr_size;
726 	}
727 	sz = (round_page(sz) / (PAGE_SIZE + sizeof(struct vm_page)));
728 	data_end += round_page(sz * sizeof(struct vm_page));
729 
730 	/*
731 	 * Reserve kernel page-table pages last, so their reservation size can
732 	 * be computed from the final bootstrap data_end (on 64-bit, only leaf
733 	 * ptbls covering [VM_MIN_KERNEL_ADDRESS, data_end + slack] are
734 	 * pre-allocated; the rest are added on demand by pmap_growkernel()).
735 	 */
736 	data_end = round_page(mmu_booke_alloc_kernel_pgtables(data_end));
737 
738 	/* Pre-round up to 1MB.  This wastes some space, but saves TLB entries */
739 	data_end = roundup2(data_end, 1 << 20);
740 
741 	debugf(" data_end: 0x%"PRI0ptrX"\n", data_end);
742 	debugf(" kernstart: %#zx\n", kernstart);
743 	debugf(" kernsize: %#zx\n", kernsize);
744 
745 	if (data_end - kernstart > kernsize) {
746 		kernsize += tlb1_mapin_region(kernstart + kernsize,
747 		    kernload + kernsize, (data_end - kernstart) - kernsize,
748 		    _TLB_ENTRY_MEM);
749 	}
750 	data_end = kernstart + kernsize;
751 	debugf(" updated data_end: 0x%"PRI0ptrX"\n", data_end);
752 
753 	/*
754 	 * Clear the structures - note we can only do it safely after the
755 	 * possible additional TLB1 translations are in place (above) so that
756 	 * all range up to the currently calculated 'data_end' is covered.
757 	 */
758 	bzero((void *)data_start, data_end - data_start);
759 	dpcpu_init(dpcpu, 0);
760 
761 	/*******************************************************/
762 	/* Set the start and end of kva. */
763 	/*******************************************************/
764 	virtual_avail = round_page(data_end);
765 	virtual_end = VM_MAX_KERNEL_ADDRESS;
766 
767 #ifndef __powerpc64__
768 	/* Allocate KVA space for page zero/copy operations. */
769 	zero_page_va = virtual_avail;
770 	virtual_avail += PAGE_SIZE;
771 	copy_page_src_va = virtual_avail;
772 	virtual_avail += PAGE_SIZE;
773 	copy_page_dst_va = virtual_avail;
774 	virtual_avail += PAGE_SIZE;
775 	debugf("zero_page_va = 0x%"PRI0ptrX"\n", zero_page_va);
776 	debugf("copy_page_src_va = 0x%"PRI0ptrX"\n", copy_page_src_va);
777 	debugf("copy_page_dst_va = 0x%"PRI0ptrX"\n", copy_page_dst_va);
778 
779 	/* Initialize page zero/copy mutexes. */
780 	mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
781 	mtx_init(&copy_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
782 
783 	/* Allocate KVA space for ptbl bufs. */
784 	ptbl_buf_pool_vabase = virtual_avail;
785 	virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
786 	debugf("ptbl_buf_pool_vabase = 0x%"PRI0ptrX" end = 0x%"PRI0ptrX"\n",
787 	    ptbl_buf_pool_vabase, virtual_avail);
788 #endif
789 #ifdef	__powerpc64__
790 	/* Allocate KVA space for crashdumpmap. */
791 	crashdumpmap = (caddr_t)virtual_avail;
792 	virtual_avail += MAXDUMPPGS * PAGE_SIZE;
793 #endif
794 
795 	/* Calculate corresponding physical addresses for the kernel region. */
796 	phys_kernelend = kernload + kernsize;
797 	debugf("kernel image and allocated data:\n");
798 	debugf(" kernload    = 0x%09jx\n", (uintmax_t)kernload);
799 	debugf(" kernstart   = 0x%"PRI0ptrX"\n", kernstart);
800 	debugf(" kernsize    = 0x%"PRI0ptrX"\n", kernsize);
801 
802 	/*
803 	 * Remove kernel physical address range from avail regions list. Page
804 	 * align all regions.  Non-page aligned memory isn't very interesting
805 	 * to us.  Also, sort the entries for ascending addresses.
806 	 */
807 
808 	sz = 0;
809 	cnt = availmem_regions_sz;
810 	debugf("processing avail regions:\n");
811 	for (mp = availmem_regions; mp->mr_size; mp++) {
812 		s = mp->mr_start;
813 		e = mp->mr_start + mp->mr_size;
814 		debugf(" %09jx-%09jx -> ", (uintmax_t)s, (uintmax_t)e);
815 		/* Check whether this region holds all of the kernel. */
816 		if (s < kernload && e > phys_kernelend) {
817 			availmem_regions[cnt].mr_start = phys_kernelend;
818 			availmem_regions[cnt++].mr_size = e - phys_kernelend;
819 			e = kernload;
820 		}
821 		/* Look whether this regions starts within the kernel. */
822 		if (s >= kernload && s < phys_kernelend) {
823 			if (e <= phys_kernelend)
824 				goto empty;
825 			s = phys_kernelend;
826 		}
827 		/* Now look whether this region ends within the kernel. */
828 		if (e > kernload && e <= phys_kernelend) {
829 			if (s >= kernload)
830 				goto empty;
831 			e = kernload;
832 		}
833 		/* Now page align the start and size of the region. */
834 		s = round_page(s);
835 		e = trunc_page(e);
836 		if (e < s)
837 			e = s;
838 		sz = e - s;
839 		debugf("%09jx-%09jx = %jx\n",
840 		    (uintmax_t)s, (uintmax_t)e, (uintmax_t)sz);
841 
842 		/* Check whether some memory is left here. */
843 		if (sz == 0) {
844 		empty:
845 			memmove(mp, mp + 1,
846 			    (cnt - (mp - availmem_regions)) * sizeof(*mp));
847 			cnt--;
848 			mp--;
849 			continue;
850 		}
851 
852 		/* Do an insertion sort. */
853 		for (mp1 = availmem_regions; mp1 < mp; mp1++)
854 			if (s < mp1->mr_start)
855 				break;
856 		if (mp1 < mp) {
857 			memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
858 			mp1->mr_start = s;
859 			mp1->mr_size = sz;
860 		} else {
861 			mp->mr_start = s;
862 			mp->mr_size = sz;
863 		}
864 	}
865 	availmem_regions_sz = cnt;
866 
867 	/*******************************************************/
868 	/* Steal physical memory for kernel stack from the end */
869 	/* of the first avail region                           */
870 	/*******************************************************/
871 	kstack0_sz = kstack_pages * PAGE_SIZE;
872 	kstack0_phys = availmem_regions[0].mr_start +
873 	    availmem_regions[0].mr_size;
874 	kstack0_phys -= kstack0_sz;
875 	availmem_regions[0].mr_size -= kstack0_sz;
876 
877 	/*******************************************************/
878 	/* Fill in phys_avail table, based on availmem_regions */
879 	/*******************************************************/
880 	phys_avail_count = 0;
881 	physsz = 0;
882 	hwphyssz = 0;
883 	TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
884 
885 	debugf("fill in phys_avail:\n");
886 	for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
887 		debugf(" region: 0x%jx - 0x%jx (0x%jx)\n",
888 		    (uintmax_t)availmem_regions[i].mr_start,
889 		    (uintmax_t)availmem_regions[i].mr_start +
890 		        availmem_regions[i].mr_size,
891 		    (uintmax_t)availmem_regions[i].mr_size);
892 
893 		if (hwphyssz != 0 &&
894 		    (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
895 			debugf(" hw.physmem adjust\n");
896 			if (physsz < hwphyssz) {
897 				phys_avail[j] = availmem_regions[i].mr_start;
898 				phys_avail[j + 1] =
899 				    availmem_regions[i].mr_start +
900 				    hwphyssz - physsz;
901 				physsz = hwphyssz;
902 				phys_avail_count++;
903 				dump_avail[j] = phys_avail[j];
904 				dump_avail[j + 1] = phys_avail[j + 1];
905 			}
906 			break;
907 		}
908 
909 		phys_avail[j] = availmem_regions[i].mr_start;
910 		phys_avail[j + 1] = availmem_regions[i].mr_start +
911 		    availmem_regions[i].mr_size;
912 		phys_avail_count++;
913 		physsz += availmem_regions[i].mr_size;
914 		dump_avail[j] = phys_avail[j];
915 		dump_avail[j + 1] = phys_avail[j + 1];
916 	}
917 	physmem = btoc(physsz);
918 
919 	/* Calculate the last available physical address. */
920 	for (i = 0; phys_avail[i + 2] != 0; i += 2)
921 		;
922 	Maxmem = powerpc_btop(phys_avail[i + 1]);
923 
924 	debugf("Maxmem = 0x%08lx\n", Maxmem);
925 	debugf("phys_avail_count = %d\n", phys_avail_count);
926 	debugf("physsz = 0x%09jx physmem = %jd (0x%09jx)\n",
927 	    (uintmax_t)physsz, (uintmax_t)physmem, (uintmax_t)physmem);
928 
929 #ifdef __powerpc64__
930 	/*
931 	 * Map the physical memory contiguously in TLB1.
932 	 * Round so it fits into a single mapping.
933 	 */
934 	tlb1_mapin_region(DMAP_BASE_ADDRESS, 0,
935 	    phys_avail[i + 1], _TLB_ENTRY_MEM);
936 #endif
937 
938 	/*******************************************************/
939 	/* Initialize (statically allocated) kernel pmap. */
940 	/*******************************************************/
941 	mtx_init(&kernel_pmap->pm_mtx, "kernel pmap", NULL, MTX_DEF);
942 
943 	debugf("kernel_pmap = 0x%"PRI0ptrX"\n", (uintptr_t)kernel_pmap);
944 	kernel_pte_alloc(virtual_avail, kernstart);
945 	for (i = 0; i < MAXCPU; i++) {
946 		kernel_pmap->pm_tid[i] = TID_KERNEL;
947 
948 		/* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
949 		tid_set_busy(i, TID_KERNEL, kernel_pmap);
950 	}
951 
952 	/* Mark kernel_pmap active on all CPUs */
953 	CPU_FILL(&kernel_pmap->pm_active);
954 
955  	/*
956 	 * Initialize the global pv list lock.
957 	 */
958 	rw_init(&pvh_global_lock, "pmap pv global");
959 
960 	/*******************************************************/
961 	/* Final setup */
962 	/*******************************************************/
963 
964 	/* Enter kstack0 into kernel map, provide guard page */
965 	kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
966 	thread0.td_kstack = (char *)kstack0;
967 	thread0.td_kstack_pages = kstack_pages;
968 
969 	debugf("kstack_sz = 0x%08jx\n", (uintmax_t)kstack0_sz);
970 	debugf("kstack0_phys at 0x%09jx - 0x%09jx\n",
971 	    (uintmax_t)kstack0_phys, (uintmax_t)kstack0_phys + kstack0_sz);
972 	debugf("kstack0 at 0x%"PRI0ptrX" - 0x%"PRI0ptrX"\n",
973 	    kstack0, kstack0 + kstack0_sz);
974 
975 	virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
976 	for (i = 0; i < kstack_pages; i++) {
977 		mmu_booke_kenter(kstack0, kstack0_phys);
978 		kstack0 += PAGE_SIZE;
979 		kstack0_phys += PAGE_SIZE;
980 	}
981 
982 	pmap_bootstrapped = 1;
983 
984 	debugf("virtual_avail = %"PRI0ptrX"\n", virtual_avail);
985 	debugf("virtual_end   = %"PRI0ptrX"\n", virtual_end);
986 
987 	debugf("mmu_booke_bootstrap: exit\n");
988 }
989 
990 #ifdef SMP
991 void
tlb1_ap_prep(void)992 tlb1_ap_prep(void)
993 {
994 	tlb_entry_t *e, tmp;
995 	unsigned int i;
996 
997 	/* Prepare TLB1 image for AP processors */
998 	e = __boot_tlb1;
999 	for (i = 0; i < TLB1_ENTRIES; i++) {
1000 		tlb1_read_entry(&tmp, i);
1001 
1002 		if ((tmp.mas1 & MAS1_VALID) && (tmp.mas2 & _TLB_ENTRY_SHARED))
1003 			memcpy(e++, &tmp, sizeof(tmp));
1004 	}
1005 }
1006 
1007 void
pmap_bootstrap_ap(volatile uint32_t * trcp __unused)1008 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1009 {
1010 	int i;
1011 
1012 	/*
1013 	 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1014 	 * have the snapshot of its contents in the s/w __boot_tlb1[] table
1015 	 * created by tlb1_ap_prep(), so use these values directly to
1016 	 * (re)program AP's TLB1 hardware.
1017 	 *
1018 	 * Start at index 1 because index 0 has the kernel map.
1019 	 */
1020 	for (i = 1; i < TLB1_ENTRIES; i++) {
1021 		if (__boot_tlb1[i].mas1 & MAS1_VALID)
1022 			tlb1_write_entry(&__boot_tlb1[i], i);
1023 	}
1024 
1025 	set_mas4_defaults();
1026 }
1027 #endif
1028 
1029 static void
booke_pmap_init_qpages(void)1030 booke_pmap_init_qpages(void)
1031 {
1032 	struct pcpu *pc;
1033 	int i;
1034 
1035 	CPU_FOREACH(i) {
1036 		pc = pcpu_find(i);
1037 		pc->pc_qmap_addr = kva_alloc(PAGE_SIZE);
1038 		if (pc->pc_qmap_addr == NULL)
1039 			panic("pmap_init_qpages: unable to allocate KVA");
1040 	}
1041 }
1042 
1043 SYSINIT(qpages_init, SI_SUB_CPU, SI_ORDER_ANY, booke_pmap_init_qpages, NULL);
1044 
1045 /*
1046  * Get the physical page address for the given pmap/virtual address.
1047  */
1048 static vm_paddr_t
mmu_booke_extract(pmap_t pmap,vm_offset_t va)1049 mmu_booke_extract(pmap_t pmap, vm_offset_t va)
1050 {
1051 	vm_paddr_t pa;
1052 
1053 	PMAP_LOCK(pmap);
1054 	pa = pte_vatopa(pmap, va);
1055 	PMAP_UNLOCK(pmap);
1056 
1057 	return (pa);
1058 }
1059 
1060 /*
1061  * Extract the physical page address associated with the given
1062  * kernel virtual address.
1063  */
1064 static vm_paddr_t
mmu_booke_kextract(vm_offset_t va)1065 mmu_booke_kextract(vm_offset_t va)
1066 {
1067 	tlb_entry_t e;
1068 	vm_paddr_t p = 0;
1069 	int i;
1070 
1071 #ifdef __powerpc64__
1072 	if (va >= DMAP_BASE_ADDRESS && va <= DMAP_MAX_ADDRESS)
1073 		return (DMAP_TO_PHYS(va));
1074 #endif
1075 
1076 	if (va >= VM_MIN_KERNEL_ADDRESS && va <= VM_MAX_KERNEL_ADDRESS)
1077 		p = pte_vatopa(kernel_pmap, va);
1078 
1079 	if (p == 0) {
1080 		/* Check TLB1 mappings */
1081 		for (i = 0; i < TLB1_ENTRIES; i++) {
1082 			tlb1_read_entry(&e, i);
1083 			if (!(e.mas1 & MAS1_VALID))
1084 				continue;
1085 			if (va >= e.virt && va < e.virt + e.size)
1086 				return (e.phys + (va - e.virt));
1087 		}
1088 	}
1089 
1090 	return (p);
1091 }
1092 
1093 /*
1094  * Initialize the pmap module.
1095  *
1096  * Called by vm_mem_init(), to initialize any structures that the pmap system
1097  * needs to map virtual memory.
1098  */
1099 static void
mmu_booke_init(void)1100 mmu_booke_init(void)
1101 {
1102 	int shpgperproc = PMAP_SHPGPERPROC;
1103 
1104 	/*
1105 	 * Initialize the address space (zone) for the pv entries.  Set a
1106 	 * high water mark so that the system can recover from excessive
1107 	 * numbers of pv entries.
1108 	 */
1109 	pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1110 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1111 
1112 	TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1113 	pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
1114 
1115 	TUNABLE_INT_FETCH("vm.pmap.pv_entry_max", &pv_entry_max);
1116 	pv_entry_high_water = 9 * (pv_entry_max / 10);
1117 
1118 	uma_zone_reserve_kva(pvzone, pv_entry_max);
1119 
1120 	/* Pre-fill pvzone with initial number of pv entries. */
1121 	uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1122 
1123 	/* Create a UMA zone for page table roots. */
1124 	ptbl_root_zone = uma_zcreate("pmap root", PMAP_ROOT_SIZE,
1125 	    NULL, NULL, NULL, NULL, UMA_ALIGN_CACHE, UMA_ZONE_VM);
1126 
1127 	/* Initialize ptbl allocation. */
1128 	ptbl_init();
1129 }
1130 
1131 /*
1132  * Map a list of wired pages into kernel virtual address space.  This is
1133  * intended for temporary mappings which do not need page modification or
1134  * references recorded.  Existing mappings in the region are overwritten.
1135  */
1136 static void
mmu_booke_qenter(void * sva,vm_page_t * m,int count)1137 mmu_booke_qenter(void *sva, vm_page_t *m, int count)
1138 {
1139 	vm_offset_t va;
1140 
1141 	va = (vm_offset_t)sva;
1142 	while (count-- > 0) {
1143 		mmu_booke_kenter(va, VM_PAGE_TO_PHYS(*m));
1144 		va += PAGE_SIZE;
1145 		m++;
1146 	}
1147 }
1148 
1149 /*
1150  * Remove page mappings from kernel virtual address space.  Intended for
1151  * temporary mappings entered by mmu_booke_qenter.
1152  */
1153 static void
mmu_booke_qremove(void * sva,int count)1154 mmu_booke_qremove(void *sva, int count)
1155 {
1156 	vm_offset_t va;
1157 
1158 	va = (vm_offset_t)sva;
1159 	while (count-- > 0) {
1160 		mmu_booke_kremove(va);
1161 		va += PAGE_SIZE;
1162 	}
1163 }
1164 
1165 /*
1166  * Map a wired page into kernel virtual address space.
1167  */
1168 static void
mmu_booke_kenter(vm_offset_t va,vm_paddr_t pa)1169 mmu_booke_kenter(vm_offset_t va, vm_paddr_t pa)
1170 {
1171 
1172 	mmu_booke_kenter_attr(va, pa, VM_MEMATTR_DEFAULT);
1173 }
1174 
1175 static void
mmu_booke_kenter_attr(vm_offset_t va,vm_paddr_t pa,vm_memattr_t ma)1176 mmu_booke_kenter_attr(vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
1177 {
1178 	uint32_t flags;
1179 	pte_t *pte;
1180 
1181 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1182 	    (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1183 
1184 	flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
1185 	flags |= tlb_calc_wimg(pa, ma) << PTE_MAS2_SHIFT;
1186 	flags |= PTE_PS_4KB;
1187 
1188 	pte = pte_find(kernel_pmap, va);
1189 	KASSERT((pte != NULL), ("mmu_booke_kenter: invalid va.  NULL PTE"));
1190 
1191 	mtx_lock_spin(&tlbivax_mutex);
1192 	tlb_miss_lock();
1193 
1194 	if (PTE_ISVALID(pte)) {
1195 		CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1196 
1197 		/* Flush entry from TLB0 */
1198 		tlb0_flush_entry(va);
1199 	}
1200 
1201 	*pte = PTE_RPN_FROM_PA(pa) | flags;
1202 
1203 	//debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1204 	//		"pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1205 	//		pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1206 
1207 	/* Flush the real memory from the instruction cache. */
1208 	if ((flags & (PTE_I | PTE_G)) == 0)
1209 		__syncicache((void *)va, PAGE_SIZE);
1210 
1211 	tlb_miss_unlock();
1212 	mtx_unlock_spin(&tlbivax_mutex);
1213 }
1214 
1215 /*
1216  * Remove a page from kernel page table.
1217  */
1218 static void
mmu_booke_kremove(vm_offset_t va)1219 mmu_booke_kremove(vm_offset_t va)
1220 {
1221 	pte_t *pte;
1222 
1223 	CTR2(KTR_PMAP,"%s: s (va = 0x%"PRI0ptrX")\n", __func__, va);
1224 
1225 	KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1226 	    (va <= VM_MAX_KERNEL_ADDRESS)),
1227 	    ("mmu_booke_kremove: invalid va"));
1228 
1229 	pte = pte_find(kernel_pmap, va);
1230 
1231 	if (pte == NULL || !PTE_ISVALID(pte)) {
1232 		CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1233 
1234 		return;
1235 	}
1236 
1237 	mtx_lock_spin(&tlbivax_mutex);
1238 	tlb_miss_lock();
1239 
1240 	/* Invalidate entry in TLB0, update PTE. */
1241 	tlb0_flush_entry(va);
1242 	*pte = 0;
1243 
1244 	tlb_miss_unlock();
1245 	mtx_unlock_spin(&tlbivax_mutex);
1246 }
1247 
1248 /*
1249  * Figure out where a given kernel pointer (usually in a fault) points
1250  * to from the VM's perspective, potentially remapping into userland's
1251  * address space.
1252  */
1253 static int
mmu_booke_decode_kernel_ptr(vm_offset_t addr,int * is_user,vm_offset_t * decoded_addr)1254 mmu_booke_decode_kernel_ptr(vm_offset_t addr, int *is_user,
1255     vm_offset_t *decoded_addr)
1256 {
1257 
1258 	if (trunc_page(addr) <= VM_MAXUSER_ADDRESS)
1259 		*is_user = 1;
1260 	else
1261 		*is_user = 0;
1262 
1263 	*decoded_addr = addr;
1264 	return (0);
1265 }
1266 
1267 static bool
mmu_booke_page_is_mapped(vm_page_t m)1268 mmu_booke_page_is_mapped(vm_page_t m)
1269 {
1270 
1271 	return (!TAILQ_EMPTY(&(m)->md.pv_list));
1272 }
1273 
1274 static bool
mmu_booke_ps_enabled(pmap_t pmap __unused)1275 mmu_booke_ps_enabled(pmap_t pmap __unused)
1276 {
1277 	return (false);
1278 }
1279 
1280 /*
1281  * Initialize pmap associated with process 0.
1282  */
1283 static void
mmu_booke_pinit0(pmap_t pmap)1284 mmu_booke_pinit0(pmap_t pmap)
1285 {
1286 
1287 	PMAP_LOCK_INIT(pmap);
1288 	mmu_booke_pinit(pmap);
1289 	PCPU_SET(curpmap, pmap);
1290 }
1291 
1292 /*
1293  * Insert the given physical page at the specified virtual address in the
1294  * target physical map with the protection requested. If specified the page
1295  * will be wired down.
1296  */
1297 static int
mmu_booke_enter(pmap_t pmap,vm_offset_t va,vm_page_t m,vm_prot_t prot,u_int flags,int8_t psind)1298 mmu_booke_enter(pmap_t pmap, vm_offset_t va, vm_page_t m,
1299     vm_prot_t prot, u_int flags, int8_t psind)
1300 {
1301 	int error;
1302 
1303 	rw_wlock(&pvh_global_lock);
1304 	PMAP_LOCK(pmap);
1305 	error = mmu_booke_enter_locked(pmap, va, m, prot, flags, psind);
1306 	PMAP_UNLOCK(pmap);
1307 	rw_wunlock(&pvh_global_lock);
1308 	return (error);
1309 }
1310 
1311 static int
mmu_booke_enter_locked(pmap_t pmap,vm_offset_t va,vm_page_t m,vm_prot_t prot,u_int pmap_flags,int8_t psind __unused)1312 mmu_booke_enter_locked(pmap_t pmap, vm_offset_t va, vm_page_t m,
1313     vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
1314 {
1315 	pte_t *pte;
1316 	vm_paddr_t pa;
1317 	pte_t flags;
1318 	int error, su, sync;
1319 
1320 	pa = VM_PAGE_TO_PHYS(m);
1321 	su = (pmap == kernel_pmap);
1322 	sync = 0;
1323 
1324 	//debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1325 	//		"pa=0x%08x prot=0x%08x flags=%#x)\n",
1326 	//		(u_int32_t)pmap, su, pmap->pm_tid,
1327 	//		(u_int32_t)m, va, pa, prot, flags);
1328 
1329 	if (su) {
1330 		KASSERT(((va >= virtual_avail) &&
1331 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1332 		    ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1333 	} else {
1334 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1335 		    ("mmu_booke_enter_locked: user pmap, non user va"));
1336 	}
1337 	if ((m->oflags & VPO_UNMANAGED) == 0) {
1338 		if ((pmap_flags & PMAP_ENTER_QUICK_LOCKED) == 0)
1339 			VM_PAGE_OBJECT_BUSY_ASSERT(m);
1340 		else
1341 			VM_OBJECT_ASSERT_LOCKED(m->object);
1342 	}
1343 
1344 	PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1345 
1346 	/*
1347 	 * If there is an existing mapping, and the physical address has not
1348 	 * changed, must be protection or wiring change.
1349 	 */
1350 	if (((pte = pte_find(pmap, va)) != NULL) &&
1351 	    (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1352 
1353 		/*
1354 		 * Before actually updating pte->flags we calculate and
1355 		 * prepare its new value in a helper var.
1356 		 */
1357 		flags = *pte;
1358 		flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1359 
1360 		/* Wiring change, just update stats. */
1361 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
1362 			if (!PTE_ISWIRED(pte)) {
1363 				flags |= PTE_WIRED;
1364 				pmap->pm_stats.wired_count++;
1365 			}
1366 		} else {
1367 			if (PTE_ISWIRED(pte)) {
1368 				flags &= ~PTE_WIRED;
1369 				pmap->pm_stats.wired_count--;
1370 			}
1371 		}
1372 
1373 		if (prot & VM_PROT_WRITE) {
1374 			/* Add write permissions. */
1375 			flags |= PTE_SW;
1376 			if (!su)
1377 				flags |= PTE_UW;
1378 
1379 			if ((flags & PTE_MANAGED) != 0)
1380 				vm_page_aflag_set(m, PGA_WRITEABLE);
1381 		} else {
1382 			/* Handle modified pages, sense modify status. */
1383 
1384 			/*
1385 			 * The PTE_MODIFIED flag could be set by underlying
1386 			 * TLB misses since we last read it (above), possibly
1387 			 * other CPUs could update it so we check in the PTE
1388 			 * directly rather than rely on that saved local flags
1389 			 * copy.
1390 			 */
1391 			if (PTE_ISMODIFIED(pte))
1392 				vm_page_dirty(m);
1393 		}
1394 
1395 		if (prot & VM_PROT_EXECUTE) {
1396 			flags |= PTE_SX;
1397 			if (!su)
1398 				flags |= PTE_UX;
1399 
1400 			/*
1401 			 * Check existing flags for execute permissions: if we
1402 			 * are turning execute permissions on, icache should
1403 			 * be flushed.
1404 			 */
1405 			if ((*pte & (PTE_UX | PTE_SX)) == 0)
1406 				sync++;
1407 		}
1408 
1409 		flags &= ~PTE_REFERENCED;
1410 
1411 		/*
1412 		 * The new flags value is all calculated -- only now actually
1413 		 * update the PTE.
1414 		 */
1415 		mtx_lock_spin(&tlbivax_mutex);
1416 		tlb_miss_lock();
1417 
1418 		tlb0_flush_entry(va);
1419 		*pte &= ~PTE_FLAGS_MASK;
1420 		*pte |= flags;
1421 
1422 		tlb_miss_unlock();
1423 		mtx_unlock_spin(&tlbivax_mutex);
1424 
1425 	} else {
1426 		/*
1427 		 * If there is an existing mapping, but it's for a different
1428 		 * physical address, pte_enter() will delete the old mapping.
1429 		 */
1430 		//if ((pte != NULL) && PTE_ISVALID(pte))
1431 		//	debugf("mmu_booke_enter_locked: replace\n");
1432 		//else
1433 		//	debugf("mmu_booke_enter_locked: new\n");
1434 
1435 		/* Now set up the flags and install the new mapping. */
1436 		flags = (PTE_SR | PTE_VALID);
1437 		flags |= PTE_M;
1438 
1439 		if (!su)
1440 			flags |= PTE_UR;
1441 
1442 		if (prot & VM_PROT_WRITE) {
1443 			flags |= PTE_SW;
1444 			if (!su)
1445 				flags |= PTE_UW;
1446 
1447 			if ((m->oflags & VPO_UNMANAGED) == 0)
1448 				vm_page_aflag_set(m, PGA_WRITEABLE);
1449 		}
1450 
1451 		if (prot & VM_PROT_EXECUTE) {
1452 			flags |= PTE_SX;
1453 			if (!su)
1454 				flags |= PTE_UX;
1455 		}
1456 
1457 		/* If its wired update stats. */
1458 		if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
1459 			flags |= PTE_WIRED;
1460 
1461 		error = pte_enter(pmap, m, va, flags,
1462 		    (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
1463 		if (error != 0)
1464 			return (KERN_RESOURCE_SHORTAGE);
1465 
1466 		if ((flags & PMAP_ENTER_WIRED) != 0)
1467 			pmap->pm_stats.wired_count++;
1468 
1469 		/* Flush the real memory from the instruction cache. */
1470 		if (prot & VM_PROT_EXECUTE)
1471 			sync++;
1472 	}
1473 
1474 	if (sync && (su || pmap == PCPU_GET(curpmap))) {
1475 		__syncicache((void *)va, PAGE_SIZE);
1476 		sync = 0;
1477 	}
1478 
1479 	return (KERN_SUCCESS);
1480 }
1481 
1482 /*
1483  * Maps a sequence of resident pages belonging to the same object.
1484  * The sequence begins with the given page m_start.  This page is
1485  * mapped at the given virtual address start.  Each subsequent page is
1486  * mapped at a virtual address that is offset from start by the same
1487  * amount as the page is offset from m_start within the object.  The
1488  * last page in the sequence is the page with the largest offset from
1489  * m_start that can be mapped at a virtual address less than the given
1490  * virtual address end.  Not every virtual page between start and end
1491  * is mapped; only those for which a resident page exists with the
1492  * corresponding offset from m_start are mapped.
1493  */
1494 static void
mmu_booke_enter_object(pmap_t pmap,vm_offset_t start,vm_offset_t end,vm_page_t m_start,vm_prot_t prot)1495 mmu_booke_enter_object(pmap_t pmap, vm_offset_t start,
1496     vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1497 {
1498 	struct pctrie_iter pages;
1499 	vm_offset_t va;
1500 	vm_page_t m;
1501 
1502 	VM_OBJECT_ASSERT_LOCKED(m_start->object);
1503 
1504 	vm_page_iter_limit_init(&pages, m_start->object,
1505 	    m_start->pindex + atop(end - start));
1506 	m = vm_radix_iter_lookup(&pages, m_start->pindex);
1507 	rw_wlock(&pvh_global_lock);
1508 	PMAP_LOCK(pmap);
1509 	while (m != NULL) {
1510 		va = start + ptoa(m->pindex - m_start->pindex);
1511 		mmu_booke_enter_locked(pmap, va, m,
1512 		    prot & (VM_PROT_READ | VM_PROT_EXECUTE),
1513 		    PMAP_ENTER_NOSLEEP | PMAP_ENTER_QUICK_LOCKED, 0);
1514 		m = vm_radix_iter_step(&pages);
1515 	}
1516 	PMAP_UNLOCK(pmap);
1517 	rw_wunlock(&pvh_global_lock);
1518 }
1519 
1520 static void
mmu_booke_enter_quick(pmap_t pmap,vm_offset_t va,vm_page_t m,vm_prot_t prot)1521 mmu_booke_enter_quick(pmap_t pmap, vm_offset_t va, vm_page_t m,
1522     vm_prot_t prot)
1523 {
1524 
1525 	rw_wlock(&pvh_global_lock);
1526 	PMAP_LOCK(pmap);
1527 	mmu_booke_enter_locked(pmap, va, m,
1528 	    prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP |
1529 	    PMAP_ENTER_QUICK_LOCKED, 0);
1530 	PMAP_UNLOCK(pmap);
1531 	rw_wunlock(&pvh_global_lock);
1532 }
1533 
1534 /*
1535  * Remove the given range of addresses from the specified map.
1536  *
1537  * It is assumed that the start and end are properly rounded to the page size.
1538  */
1539 static void
mmu_booke_remove(pmap_t pmap,vm_offset_t va,vm_offset_t endva)1540 mmu_booke_remove(pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1541 {
1542 	pte_t *pte;
1543 	uint8_t hold_flag;
1544 
1545 	int su = (pmap == kernel_pmap);
1546 
1547 	//debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1548 	//		su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1549 
1550 	if (su) {
1551 		KASSERT(((va >= virtual_avail) &&
1552 		    (va <= VM_MAX_KERNEL_ADDRESS)),
1553 		    ("mmu_booke_remove: kernel pmap, non kernel va"));
1554 	} else {
1555 		KASSERT((va <= VM_MAXUSER_ADDRESS),
1556 		    ("mmu_booke_remove: user pmap, non user va"));
1557 	}
1558 
1559 	if (PMAP_REMOVE_DONE(pmap)) {
1560 		//debugf("mmu_booke_remove: e (empty)\n");
1561 		return;
1562 	}
1563 
1564 	hold_flag = PTBL_HOLD_FLAG(pmap);
1565 	//debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1566 
1567 	rw_wlock(&pvh_global_lock);
1568 	PMAP_LOCK(pmap);
1569 	for (; va < endva; va += PAGE_SIZE) {
1570 		pte = pte_find_next(pmap, &va);
1571 		if ((pte == NULL) || !PTE_ISVALID(pte))
1572 			break;
1573 		if (va >= endva)
1574 			break;
1575 		pte_remove(pmap, va, hold_flag);
1576 	}
1577 	PMAP_UNLOCK(pmap);
1578 	rw_wunlock(&pvh_global_lock);
1579 
1580 	//debugf("mmu_booke_remove: e\n");
1581 }
1582 
1583 /*
1584  * Remove physical page from all pmaps in which it resides.
1585  */
1586 static void
mmu_booke_remove_all(vm_page_t m)1587 mmu_booke_remove_all(vm_page_t m)
1588 {
1589 	pv_entry_t pv, pvn;
1590 	uint8_t hold_flag;
1591 
1592 	rw_wlock(&pvh_global_lock);
1593 	TAILQ_FOREACH_SAFE(pv, &m->md.pv_list, pv_link, pvn) {
1594 		PMAP_LOCK(pv->pv_pmap);
1595 		hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1596 		pte_remove(pv->pv_pmap, pv->pv_va, hold_flag);
1597 		PMAP_UNLOCK(pv->pv_pmap);
1598 	}
1599 	vm_page_aflag_clear(m, PGA_WRITEABLE);
1600 	rw_wunlock(&pvh_global_lock);
1601 }
1602 
1603 /*
1604  * Map a range of physical addresses into kernel virtual address space.
1605  */
1606 static void *
mmu_booke_map(vm_offset_t * virt,vm_paddr_t pa_start,vm_paddr_t pa_end,int prot)1607 mmu_booke_map(vm_offset_t *virt, vm_paddr_t pa_start,
1608     vm_paddr_t pa_end, int prot)
1609 {
1610 	vm_offset_t sva = *virt;
1611 	vm_offset_t va = sva;
1612 
1613 #ifdef __powerpc64__
1614 	/* XXX: Handle memory not starting at 0x0. */
1615 	if (pa_end < ctob(Maxmem))
1616 		return (PHYS_TO_DMAP(pa_start));
1617 #endif
1618 
1619 	while (pa_start < pa_end) {
1620 		mmu_booke_kenter(va, pa_start);
1621 		va += PAGE_SIZE;
1622 		pa_start += PAGE_SIZE;
1623 	}
1624 	*virt = va;
1625 
1626 	return ((void *)sva);
1627 }
1628 
1629 /*
1630  * The pmap must be activated before it's address space can be accessed in any
1631  * way.
1632  */
1633 static void
mmu_booke_activate(struct thread * td)1634 mmu_booke_activate(struct thread *td)
1635 {
1636 	pmap_t pmap;
1637 	u_int cpuid;
1638 
1639 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1640 
1641 	CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%"PRI0ptrX")",
1642 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1643 
1644 	KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1645 
1646 	sched_pin();
1647 
1648 	cpuid = PCPU_GET(cpuid);
1649 	CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
1650 	PCPU_SET(curpmap, pmap);
1651 
1652 	if (pmap->pm_tid[cpuid] == TID_NONE)
1653 		tid_alloc(pmap);
1654 
1655 	/* Load PID0 register with pmap tid value. */
1656 	mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
1657 	__asm __volatile("isync");
1658 
1659 	mtspr(SPR_DBCR0, td->td_pcb->pcb_cpu.booke.dbcr0);
1660 
1661 	sched_unpin();
1662 
1663 	CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
1664 	    pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
1665 }
1666 
1667 /*
1668  * Deactivate the specified process's address space.
1669  */
1670 static void
mmu_booke_deactivate(struct thread * td)1671 mmu_booke_deactivate(struct thread *td)
1672 {
1673 	pmap_t pmap;
1674 
1675 	pmap = &td->td_proc->p_vmspace->vm_pmap;
1676 
1677 	CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%"PRI0ptrX,
1678 	    __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1679 
1680 	td->td_pcb->pcb_cpu.booke.dbcr0 = mfspr(SPR_DBCR0);
1681 
1682 	CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
1683 	PCPU_SET(curpmap, NULL);
1684 }
1685 
1686 /*
1687  * Copy the range specified by src_addr/len
1688  * from the source map to the range dst_addr/len
1689  * in the destination map.
1690  *
1691  * This routine is only advisory and need not do anything.
1692  */
1693 static void
mmu_booke_copy(pmap_t dst_pmap,pmap_t src_pmap,vm_offset_t dst_addr,vm_size_t len,vm_offset_t src_addr)1694 mmu_booke_copy(pmap_t dst_pmap, pmap_t src_pmap,
1695     vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
1696 {
1697 
1698 }
1699 
1700 /*
1701  * Set the physical protection on the specified range of this map as requested.
1702  */
1703 static void
mmu_booke_protect(pmap_t pmap,vm_offset_t sva,vm_offset_t eva,vm_prot_t prot)1704 mmu_booke_protect(pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
1705     vm_prot_t prot)
1706 {
1707 	vm_offset_t va;
1708 	vm_page_t m;
1709 	pte_t *pte;
1710 
1711 	if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
1712 		mmu_booke_remove(pmap, sva, eva);
1713 		return;
1714 	}
1715 
1716 	if (prot & VM_PROT_WRITE)
1717 		return;
1718 
1719 	PMAP_LOCK(pmap);
1720 	for (va = sva; va < eva; va += PAGE_SIZE) {
1721 		if ((pte = pte_find(pmap, va)) != NULL) {
1722 			if (PTE_ISVALID(pte)) {
1723 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1724 
1725 				mtx_lock_spin(&tlbivax_mutex);
1726 				tlb_miss_lock();
1727 
1728 				/* Handle modified pages. */
1729 				if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
1730 					vm_page_dirty(m);
1731 
1732 				tlb0_flush_entry(va);
1733 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
1734 
1735 				tlb_miss_unlock();
1736 				mtx_unlock_spin(&tlbivax_mutex);
1737 			}
1738 		}
1739 	}
1740 	PMAP_UNLOCK(pmap);
1741 }
1742 
1743 /*
1744  * Clear the write and modified bits in each of the given page's mappings.
1745  */
1746 static void
mmu_booke_remove_write(vm_page_t m)1747 mmu_booke_remove_write(vm_page_t m)
1748 {
1749 	pv_entry_t pv;
1750 	pte_t *pte;
1751 
1752 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
1753 	    ("mmu_booke_remove_write: page %p is not managed", m));
1754 	vm_page_assert_busied(m);
1755 
1756 	if (!pmap_page_is_write_mapped(m))
1757 	        return;
1758 	rw_wlock(&pvh_global_lock);
1759 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1760 		PMAP_LOCK(pv->pv_pmap);
1761 		if ((pte = pte_find(pv->pv_pmap, pv->pv_va)) != NULL) {
1762 			if (PTE_ISVALID(pte)) {
1763 				m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1764 
1765 				mtx_lock_spin(&tlbivax_mutex);
1766 				tlb_miss_lock();
1767 
1768 				/* Handle modified pages. */
1769 				if (PTE_ISMODIFIED(pte))
1770 					vm_page_dirty(m);
1771 
1772 				/* Flush mapping from TLB0. */
1773 				*pte &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
1774 
1775 				tlb_miss_unlock();
1776 				mtx_unlock_spin(&tlbivax_mutex);
1777 			}
1778 		}
1779 		PMAP_UNLOCK(pv->pv_pmap);
1780 	}
1781 	vm_page_aflag_clear(m, PGA_WRITEABLE);
1782 	rw_wunlock(&pvh_global_lock);
1783 }
1784 
1785 /*
1786  * Atomically extract and hold the physical page with the given
1787  * pmap and virtual address pair if that mapping permits the given
1788  * protection.
1789  */
1790 static vm_page_t
mmu_booke_extract_and_hold(pmap_t pmap,vm_offset_t va,vm_prot_t prot)1791 mmu_booke_extract_and_hold(pmap_t pmap, vm_offset_t va,
1792     vm_prot_t prot)
1793 {
1794 	pte_t *pte;
1795 	vm_page_t m;
1796 	uint32_t pte_wbit;
1797 
1798 	m = NULL;
1799 	PMAP_LOCK(pmap);
1800 	pte = pte_find(pmap, va);
1801 	if ((pte != NULL) && PTE_ISVALID(pte)) {
1802 		if (pmap == kernel_pmap)
1803 			pte_wbit = PTE_SW;
1804 		else
1805 			pte_wbit = PTE_UW;
1806 
1807 		if ((*pte & pte_wbit) != 0 || (prot & VM_PROT_WRITE) == 0) {
1808 			m = PHYS_TO_VM_PAGE(PTE_PA(pte));
1809 			if (!vm_page_wire_mapped(m))
1810 				m = NULL;
1811 		}
1812 	}
1813 	PMAP_UNLOCK(pmap);
1814 	return (m);
1815 }
1816 
1817 /*
1818  * Initialize a vm_page's machine-dependent fields.
1819  */
1820 static void
mmu_booke_page_init(vm_page_t m)1821 mmu_booke_page_init(vm_page_t m)
1822 {
1823 
1824 	m->md.pv_tracked = 0;
1825 	TAILQ_INIT(&m->md.pv_list);
1826 }
1827 
1828 /*
1829  * Return whether or not the specified physical page was modified
1830  * in any of physical maps.
1831  */
1832 static bool
mmu_booke_is_modified(vm_page_t m)1833 mmu_booke_is_modified(vm_page_t m)
1834 {
1835 	pte_t *pte;
1836 	pv_entry_t pv;
1837 	bool rv;
1838 
1839 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
1840 	    ("mmu_booke_is_modified: page %p is not managed", m));
1841 	rv = false;
1842 
1843 	/*
1844 	 * If the page is not busied then this check is racy.
1845 	 */
1846 	if (!pmap_page_is_write_mapped(m))
1847 		return (false);
1848 
1849 	rw_wlock(&pvh_global_lock);
1850 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1851 		PMAP_LOCK(pv->pv_pmap);
1852 		if ((pte = pte_find(pv->pv_pmap, pv->pv_va)) != NULL &&
1853 		    PTE_ISVALID(pte)) {
1854 			if (PTE_ISMODIFIED(pte))
1855 				rv = true;
1856 		}
1857 		PMAP_UNLOCK(pv->pv_pmap);
1858 		if (rv)
1859 			break;
1860 	}
1861 	rw_wunlock(&pvh_global_lock);
1862 	return (rv);
1863 }
1864 
1865 /*
1866  * Return whether or not the specified virtual address is eligible
1867  * for prefault.
1868  */
1869 static bool
mmu_booke_is_prefaultable(pmap_t pmap,vm_offset_t addr)1870 mmu_booke_is_prefaultable(pmap_t pmap, vm_offset_t addr)
1871 {
1872 
1873 	return (false);
1874 }
1875 
1876 /*
1877  * Return whether or not the specified physical page was referenced
1878  * in any physical maps.
1879  */
1880 static bool
mmu_booke_is_referenced(vm_page_t m)1881 mmu_booke_is_referenced(vm_page_t m)
1882 {
1883 	pte_t *pte;
1884 	pv_entry_t pv;
1885 	bool rv;
1886 
1887 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
1888 	    ("mmu_booke_is_referenced: page %p is not managed", m));
1889 	rv = false;
1890 	rw_wlock(&pvh_global_lock);
1891 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1892 		PMAP_LOCK(pv->pv_pmap);
1893 		if ((pte = pte_find(pv->pv_pmap, pv->pv_va)) != NULL &&
1894 		    PTE_ISVALID(pte)) {
1895 			if (PTE_ISREFERENCED(pte))
1896 				rv = true;
1897 		}
1898 		PMAP_UNLOCK(pv->pv_pmap);
1899 		if (rv)
1900 			break;
1901 	}
1902 	rw_wunlock(&pvh_global_lock);
1903 	return (rv);
1904 }
1905 
1906 /*
1907  * Clear the modify bits on the specified physical page.
1908  */
1909 static void
mmu_booke_clear_modify(vm_page_t m)1910 mmu_booke_clear_modify(vm_page_t m)
1911 {
1912 	pte_t *pte;
1913 	pv_entry_t pv;
1914 
1915 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
1916 	    ("mmu_booke_clear_modify: page %p is not managed", m));
1917 	vm_page_assert_busied(m);
1918 
1919 	if (!pmap_page_is_write_mapped(m))
1920 	        return;
1921 
1922 	rw_wlock(&pvh_global_lock);
1923 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1924 		PMAP_LOCK(pv->pv_pmap);
1925 		if ((pte = pte_find(pv->pv_pmap, pv->pv_va)) != NULL &&
1926 		    PTE_ISVALID(pte)) {
1927 			mtx_lock_spin(&tlbivax_mutex);
1928 			tlb_miss_lock();
1929 
1930 			if (*pte & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
1931 				tlb0_flush_entry(pv->pv_va);
1932 				*pte &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
1933 				    PTE_REFERENCED);
1934 			}
1935 
1936 			tlb_miss_unlock();
1937 			mtx_unlock_spin(&tlbivax_mutex);
1938 		}
1939 		PMAP_UNLOCK(pv->pv_pmap);
1940 	}
1941 	rw_wunlock(&pvh_global_lock);
1942 }
1943 
1944 /*
1945  * Return a count of reference bits for a page, clearing those bits.
1946  * It is not necessary for every reference bit to be cleared, but it
1947  * is necessary that 0 only be returned when there are truly no
1948  * reference bits set.
1949  *
1950  * As an optimization, update the page's dirty field if a modified bit is
1951  * found while counting reference bits.  This opportunistic update can be
1952  * performed at low cost and can eliminate the need for some future calls
1953  * to pmap_is_modified().  However, since this function stops after
1954  * finding PMAP_TS_REFERENCED_MAX reference bits, it may not detect some
1955  * dirty pages.  Those dirty pages will only be detected by a future call
1956  * to pmap_is_modified().
1957  */
1958 static int
mmu_booke_ts_referenced(vm_page_t m)1959 mmu_booke_ts_referenced(vm_page_t m)
1960 {
1961 	pte_t *pte;
1962 	pv_entry_t pv;
1963 	int count;
1964 
1965 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
1966 	    ("mmu_booke_ts_referenced: page %p is not managed", m));
1967 	count = 0;
1968 	rw_wlock(&pvh_global_lock);
1969 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
1970 		PMAP_LOCK(pv->pv_pmap);
1971 		if ((pte = pte_find(pv->pv_pmap, pv->pv_va)) != NULL &&
1972 		    PTE_ISVALID(pte)) {
1973 			if (PTE_ISMODIFIED(pte))
1974 				vm_page_dirty(m);
1975 			if (PTE_ISREFERENCED(pte)) {
1976 				mtx_lock_spin(&tlbivax_mutex);
1977 				tlb_miss_lock();
1978 
1979 				tlb0_flush_entry(pv->pv_va);
1980 				*pte &= ~PTE_REFERENCED;
1981 
1982 				tlb_miss_unlock();
1983 				mtx_unlock_spin(&tlbivax_mutex);
1984 
1985 				if (++count >= PMAP_TS_REFERENCED_MAX) {
1986 					PMAP_UNLOCK(pv->pv_pmap);
1987 					break;
1988 				}
1989 			}
1990 		}
1991 		PMAP_UNLOCK(pv->pv_pmap);
1992 	}
1993 	rw_wunlock(&pvh_global_lock);
1994 	return (count);
1995 }
1996 
1997 /*
1998  * Clear the wired attribute from the mappings for the specified range of
1999  * addresses in the given pmap.  Every valid mapping within that range must
2000  * have the wired attribute set.  In contrast, invalid mappings cannot have
2001  * the wired attribute set, so they are ignored.
2002  *
2003  * The wired attribute of the page table entry is not a hardware feature, so
2004  * there is no need to invalidate any TLB entries.
2005  */
2006 static void
mmu_booke_unwire(pmap_t pmap,vm_offset_t sva,vm_offset_t eva)2007 mmu_booke_unwire(pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
2008 {
2009 	vm_offset_t va;
2010 	pte_t *pte;
2011 
2012 	PMAP_LOCK(pmap);
2013 	for (va = sva; va < eva; va += PAGE_SIZE) {
2014 		if ((pte = pte_find(pmap, va)) != NULL &&
2015 		    PTE_ISVALID(pte)) {
2016 			if (!PTE_ISWIRED(pte))
2017 				panic("mmu_booke_unwire: pte %p isn't wired",
2018 				    pte);
2019 			*pte &= ~PTE_WIRED;
2020 			pmap->pm_stats.wired_count--;
2021 		}
2022 	}
2023 	PMAP_UNLOCK(pmap);
2024 
2025 }
2026 
2027 /*
2028  * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2029  * page.  This count may be changed upwards or downwards in the future; it is
2030  * only necessary that true be returned for a small subset of pmaps for proper
2031  * page aging.
2032  */
2033 static bool
mmu_booke_page_exists_quick(pmap_t pmap,vm_page_t m)2034 mmu_booke_page_exists_quick(pmap_t pmap, vm_page_t m)
2035 {
2036 	pv_entry_t pv;
2037 	int loops;
2038 	bool rv;
2039 
2040 	KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2041 	    ("mmu_booke_page_exists_quick: page %p is not managed", m));
2042 	loops = 0;
2043 	rv = false;
2044 	rw_wlock(&pvh_global_lock);
2045 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2046 		if (pv->pv_pmap == pmap) {
2047 			rv = true;
2048 			break;
2049 		}
2050 		if (++loops >= 16)
2051 			break;
2052 	}
2053 	rw_wunlock(&pvh_global_lock);
2054 	return (rv);
2055 }
2056 
2057 /*
2058  * Return the number of managed mappings to the given physical page that are
2059  * wired.
2060  */
2061 static int
mmu_booke_page_wired_mappings(vm_page_t m)2062 mmu_booke_page_wired_mappings(vm_page_t m)
2063 {
2064 	pv_entry_t pv;
2065 	pte_t *pte;
2066 	int count = 0;
2067 
2068 	if ((m->oflags & VPO_UNMANAGED) != 0)
2069 		return (count);
2070 	rw_wlock(&pvh_global_lock);
2071 	TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2072 		PMAP_LOCK(pv->pv_pmap);
2073 		if ((pte = pte_find(pv->pv_pmap, pv->pv_va)) != NULL)
2074 			if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2075 				count++;
2076 		PMAP_UNLOCK(pv->pv_pmap);
2077 	}
2078 	rw_wunlock(&pvh_global_lock);
2079 	return (count);
2080 }
2081 
2082 static int
mmu_booke_dev_direct_mapped(vm_paddr_t pa,vm_size_t size)2083 mmu_booke_dev_direct_mapped(vm_paddr_t pa, vm_size_t size)
2084 {
2085 	int i;
2086 	vm_offset_t va;
2087 
2088 	/*
2089 	 * This currently does not work for entries that
2090 	 * overlap TLB1 entries.
2091 	 */
2092 	for (i = 0; i < TLB1_ENTRIES; i ++) {
2093 		if (tlb1_iomapped(i, pa, size, &va) == 0)
2094 			return (0);
2095 	}
2096 
2097 	return (EFAULT);
2098 }
2099 
2100 void
mmu_booke_dumpsys_map(vm_paddr_t pa,size_t sz,void ** va)2101 mmu_booke_dumpsys_map(vm_paddr_t pa, size_t sz, void **va)
2102 {
2103 	vm_paddr_t ppa;
2104 	vm_offset_t ofs;
2105 	vm_size_t gran;
2106 
2107 	/* Minidumps are based on virtual memory addresses. */
2108 	if (do_minidump) {
2109 		*va = (void *)(vm_offset_t)pa;
2110 		return;
2111 	}
2112 
2113 	/* Raw physical memory dumps don't have a virtual address. */
2114 	/* We always map a 256MB page at 256M. */
2115 	gran = 256 * 1024 * 1024;
2116 	ppa = rounddown2(pa, gran);
2117 	ofs = pa - ppa;
2118 	*va = (void *)gran;
2119 	tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
2120 
2121 	if (sz > (gran - ofs))
2122 		tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
2123 		    _TLB_ENTRY_IO);
2124 }
2125 
2126 void
mmu_booke_dumpsys_unmap(vm_paddr_t pa,size_t sz,void * va)2127 mmu_booke_dumpsys_unmap(vm_paddr_t pa, size_t sz, void *va)
2128 {
2129 	vm_paddr_t ppa;
2130 	vm_offset_t ofs;
2131 	vm_size_t gran;
2132 	tlb_entry_t e;
2133 	int i;
2134 
2135 	/* Minidumps are based on virtual memory addresses. */
2136 	/* Nothing to do... */
2137 	if (do_minidump)
2138 		return;
2139 
2140 	for (i = 0; i < TLB1_ENTRIES; i++) {
2141 		tlb1_read_entry(&e, i);
2142 		if (!(e.mas1 & MAS1_VALID))
2143 			break;
2144 	}
2145 
2146 	/* Raw physical memory dumps don't have a virtual address. */
2147 	i--;
2148 	e.mas1 = 0;
2149 	e.mas2 = 0;
2150 	e.mas3 = 0;
2151 	tlb1_write_entry(&e, i);
2152 
2153 	gran = 256 * 1024 * 1024;
2154 	ppa = rounddown2(pa, gran);
2155 	ofs = pa - ppa;
2156 	if (sz > (gran - ofs)) {
2157 		i--;
2158 		e.mas1 = 0;
2159 		e.mas2 = 0;
2160 		e.mas3 = 0;
2161 		tlb1_write_entry(&e, i);
2162 	}
2163 }
2164 
2165 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
2166 
2167 void
mmu_booke_scan_init(void)2168 mmu_booke_scan_init(void)
2169 {
2170 	vm_offset_t va;
2171 	pte_t *pte;
2172 	int i;
2173 
2174 	if (!do_minidump) {
2175 		/* Initialize phys. segments for dumpsys(). */
2176 		memset(&dump_map, 0, sizeof(dump_map));
2177 		mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
2178 		    &availmem_regions_sz);
2179 		for (i = 0; i < physmem_regions_sz; i++) {
2180 			dump_map[i].pa_start = physmem_regions[i].mr_start;
2181 			dump_map[i].pa_size = physmem_regions[i].mr_size;
2182 		}
2183 		return;
2184 	}
2185 
2186 	/* Virtual segments for minidumps: */
2187 	memset(&dump_map, 0, sizeof(dump_map));
2188 
2189 	/* 1st: kernel .data and .bss. */
2190 	dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
2191 	dump_map[0].pa_size =
2192 	    round_page((uintptr_t)_end) - dump_map[0].pa_start;
2193 
2194 	/* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2195 	dump_map[1].pa_start = data_start;
2196 	dump_map[1].pa_size = data_end - data_start;
2197 
2198 	/* 3rd: kernel VM. */
2199 	va = dump_map[1].pa_start + dump_map[1].pa_size;
2200 	/* Find start of next chunk (from va). */
2201 	while (va < virtual_end) {
2202 		/* Don't dump the buffer cache. */
2203 		if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
2204 			va = kmi.buffer_eva;
2205 			continue;
2206 		}
2207 		pte = pte_find(kernel_pmap, va);
2208 		if (pte != NULL && PTE_ISVALID(pte))
2209 			break;
2210 		va += PAGE_SIZE;
2211 	}
2212 	if (va < virtual_end) {
2213 		dump_map[2].pa_start = va;
2214 		va += PAGE_SIZE;
2215 		/* Find last page in chunk. */
2216 		while (va < virtual_end) {
2217 			/* Don't run into the buffer cache. */
2218 			if (va == kmi.buffer_sva)
2219 				break;
2220 			pte = pte_find(kernel_pmap, va);
2221 			if (pte == NULL || !PTE_ISVALID(pte))
2222 				break;
2223 			va += PAGE_SIZE;
2224 		}
2225 		dump_map[2].pa_size = va - dump_map[2].pa_start;
2226 	}
2227 }
2228 
2229 /*
2230  * Map a set of physical memory pages into the kernel virtual address space.
2231  * Return a pointer to where it is mapped. This routine is intended to be used
2232  * for mapping device memory, NOT real memory.
2233  */
2234 static void *
mmu_booke_mapdev(vm_paddr_t pa,vm_size_t size)2235 mmu_booke_mapdev(vm_paddr_t pa, vm_size_t size)
2236 {
2237 
2238 	return (mmu_booke_mapdev_attr(pa, size, VM_MEMATTR_DEFAULT));
2239 }
2240 
2241 static int
tlb1_find_pa(vm_paddr_t pa,tlb_entry_t * e)2242 tlb1_find_pa(vm_paddr_t pa, tlb_entry_t *e)
2243 {
2244 	int i;
2245 
2246 	for (i = 0; i < TLB1_ENTRIES; i++) {
2247 		tlb1_read_entry(e, i);
2248 		if ((e->mas1 & MAS1_VALID) == 0)
2249 			continue;
2250 		if (e->phys == pa)
2251 			return (i);
2252 	}
2253 	return (-1);
2254 }
2255 
2256 static void *
mmu_booke_mapdev_attr(vm_paddr_t pa,vm_size_t size,vm_memattr_t ma)2257 mmu_booke_mapdev_attr(vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
2258 {
2259 	tlb_entry_t e;
2260 	vm_paddr_t tmppa;
2261 #ifndef __powerpc64__
2262 	uintptr_t tmpva;
2263 #endif
2264 	uintptr_t va, retva;
2265 	vm_size_t sz;
2266 	int i;
2267 	int wimge;
2268 
2269 	/*
2270 	 * Check if this is premapped in TLB1.
2271 	 */
2272 	sz = size;
2273 	tmppa = pa;
2274 	va = ~0;
2275 	wimge = tlb_calc_wimg(pa, ma);
2276 	for (i = 0; i < TLB1_ENTRIES; i++) {
2277 		tlb1_read_entry(&e, i);
2278 		if (!(e.mas1 & MAS1_VALID))
2279 			continue;
2280 		if (wimge != (e.mas2 & (MAS2_WIMGE_MASK & ~_TLB_ENTRY_SHARED)))
2281 			continue;
2282 		if (tmppa >= e.phys && tmppa < e.phys + e.size) {
2283 			va = e.virt + (pa - e.phys);
2284 			tmppa = e.phys + e.size;
2285 			sz -= MIN(sz, e.size - (pa - e.phys));
2286 			while (sz > 0 && (i = tlb1_find_pa(tmppa, &e)) != -1) {
2287 				if (wimge != (e.mas2 & (MAS2_WIMGE_MASK & ~_TLB_ENTRY_SHARED)))
2288 					break;
2289 				sz -= MIN(sz, e.size);
2290 				tmppa = e.phys + e.size;
2291 			}
2292 			if (sz != 0)
2293 				break;
2294 			return ((void *)va);
2295 		}
2296 	}
2297 
2298 	size = roundup(size, PAGE_SIZE);
2299 
2300 #ifdef __powerpc64__
2301 	KASSERT(pa < VM_MAPDEV_PA_MAX,
2302 	    ("Unsupported physical address! %lx", pa));
2303 	va = VM_MAPDEV_BASE + pa;
2304 	retva = va;
2305 #ifdef POW2_MAPPINGS
2306 	/*
2307 	 * Align the mapping to a power of 2 size, taking into account that we
2308 	 * may need to increase the size multiple times to satisfy the size and
2309 	 * alignment requirements.
2310 	 *
2311 	 * This works in the general case because it's very rare (near never?)
2312 	 * to have different access properties (WIMG) within a single
2313 	 * power-of-two region.  If a design does call for that, POW2_MAPPINGS
2314 	 * can be undefined, and exact mappings will be used instead.
2315 	 */
2316 	sz = size;
2317 	size = roundup2(size, 1 << ilog2(size));
2318 	while (rounddown2(va, size) + size < va + sz)
2319 		size <<= 1;
2320 	va = rounddown2(va, size);
2321 	pa = rounddown2(pa, size);
2322 #endif
2323 #else
2324 	/*
2325 	 * The device mapping area is between VM_MAXUSER_ADDRESS and
2326 	 * VM_MIN_KERNEL_ADDRESS.  This gives 1GB of device addressing.
2327 	 */
2328 #ifdef SPARSE_MAPDEV
2329 	/*
2330 	 * With a sparse mapdev, align to the largest starting region.  This
2331 	 * could feasibly be optimized for a 'best-fit' alignment, but that
2332 	 * calculation could be very costly.
2333 	 * Align to the smaller of:
2334 	 * - first set bit in overlap of (pa & size mask)
2335 	 * - largest size envelope
2336 	 *
2337 	 * It's possible the device mapping may start at a PA that's not larger
2338 	 * than the size mask, so we need to offset in to maximize the TLB entry
2339 	 * range and minimize the number of used TLB entries.
2340 	 */
2341 	do {
2342 	    tmpva = tlb1_map_base;
2343 	    sz = ffsl((~((1 << flsl(size-1)) - 1)) & pa);
2344 	    sz = sz ? min(roundup(sz + 3, 4), flsl(size) - 1) : flsl(size) - 1;
2345 	    va = roundup(tlb1_map_base, 1 << sz) | (((1 << sz) - 1) & pa);
2346 	} while (!atomic_cmpset_int(&tlb1_map_base, tmpva, va + size));
2347 #endif
2348 	va = atomic_fetchadd_int(&tlb1_map_base, size);
2349 	retva = va;
2350 #endif
2351 
2352 	if (tlb1_mapin_region(va, pa, size, tlb_calc_wimg(pa, ma)) != size)
2353 		return (NULL);
2354 
2355 	return ((void *)retva);
2356 }
2357 
2358 /*
2359  * 'Unmap' a range mapped by mmu_booke_mapdev().
2360  */
2361 static void
mmu_booke_unmapdev(void * p,vm_size_t size)2362 mmu_booke_unmapdev(void *p, vm_size_t size)
2363 {
2364 #ifdef SUPPORTS_SHRINKING_TLB1
2365 	void *base;
2366 	vm_offset_t offset, va;
2367 
2368 	/*
2369 	 * Unmap only if this is inside kernel virtual space.
2370 	 */
2371 	va = (vm_offset_t)p;
2372 	if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2373 		base = trunc_page(va);
2374 		offset = va & PAGE_MASK;
2375 		size = roundup(offset + size, PAGE_SIZE);
2376 		mmu_booke_qremove(base, atop(size));
2377 		kva_free((vm_offset_t)base, size);
2378 	}
2379 #endif
2380 }
2381 
2382 /*
2383  * mmu_booke_object_init_pt preloads the ptes for a given object into the
2384  * specified pmap. This eliminates the blast of soft faults on process startup
2385  * and immediately after an mmap.
2386  */
2387 static void
mmu_booke_object_init_pt(pmap_t pmap,vm_offset_t addr,vm_object_t object,vm_pindex_t pindex,vm_size_t size)2388 mmu_booke_object_init_pt(pmap_t pmap, vm_offset_t addr,
2389     vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2390 {
2391 
2392 	VM_OBJECT_ASSERT_WLOCKED(object);
2393 	KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2394 	    ("mmu_booke_object_init_pt: non-device object"));
2395 }
2396 
2397 /*
2398  * Perform the pmap work for mincore.
2399  */
2400 static int
mmu_booke_mincore(pmap_t pmap,vm_offset_t addr,vm_paddr_t * pap)2401 mmu_booke_mincore(pmap_t pmap, vm_offset_t addr, vm_paddr_t *pap)
2402 {
2403 
2404 	/* XXX: this should be implemented at some point */
2405 	return (0);
2406 }
2407 
2408 static int
mmu_booke_change_attr(void * sva,vm_size_t sz,vm_memattr_t mode)2409 mmu_booke_change_attr(void *sva, vm_size_t sz, vm_memattr_t mode)
2410 {
2411 	vm_offset_t addr, va;
2412 	pte_t *pte;
2413 	int i, j;
2414 	tlb_entry_t e;
2415 
2416 	addr = (vm_offset_t)sva;
2417 	addr = trunc_page(addr);
2418 
2419 	/* Only allow changes to mapped kernel addresses.  This includes:
2420 	 * - KVA
2421 	 * - DMAP (powerpc64)
2422 	 * - Device mappings
2423 	 */
2424 	if (addr <= VM_MAXUSER_ADDRESS ||
2425 #ifdef __powerpc64__
2426 	    (addr >= tlb1_map_base && addr < DMAP_BASE_ADDRESS) ||
2427 	    (addr > DMAP_MAX_ADDRESS && addr < VM_MIN_KERNEL_ADDRESS) ||
2428 #else
2429 	    (addr >= tlb1_map_base && addr < VM_MIN_KERNEL_ADDRESS) ||
2430 #endif
2431 	    (addr > VM_MAX_KERNEL_ADDRESS))
2432 		return (EINVAL);
2433 
2434 	/* Check TLB1 mappings */
2435 	for (i = 0; i < TLB1_ENTRIES; i++) {
2436 		tlb1_read_entry(&e, i);
2437 		if (!(e.mas1 & MAS1_VALID))
2438 			continue;
2439 		if (addr >= e.virt && addr < e.virt + e.size)
2440 			break;
2441 	}
2442 	if (i < TLB1_ENTRIES) {
2443 		/* Only allow full mappings to be modified for now. */
2444 		/* Validate the range. */
2445 		for (j = i, va = addr; va < addr + sz; va += e.size, j++) {
2446 			tlb1_read_entry(&e, j);
2447 			if (va != e.virt || (sz - (va - addr) < e.size))
2448 				return (EINVAL);
2449 		}
2450 		for (va = addr; va < addr + sz; va += e.size, i++) {
2451 			tlb1_read_entry(&e, i);
2452 			e.mas2 &= ~MAS2_WIMGE_MASK;
2453 			e.mas2 |= tlb_calc_wimg(e.phys, mode);
2454 
2455 			/*
2456 			 * Write it out to the TLB.  Should really re-sync with other
2457 			 * cores.
2458 			 */
2459 			tlb1_write_entry(&e, i);
2460 		}
2461 		return (0);
2462 	}
2463 
2464 	/* Not in TLB1, try through pmap */
2465 	/* First validate the range. */
2466 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
2467 		pte = pte_find(kernel_pmap, va);
2468 		if (pte == NULL || !PTE_ISVALID(pte))
2469 			return (EINVAL);
2470 	}
2471 
2472 	mtx_lock_spin(&tlbivax_mutex);
2473 	tlb_miss_lock();
2474 	for (va = addr; va < addr + sz; va += PAGE_SIZE) {
2475 		pte = pte_find(kernel_pmap, va);
2476 		*pte &= ~(PTE_MAS2_MASK << PTE_MAS2_SHIFT);
2477 		*pte |= tlb_calc_wimg(PTE_PA(pte), mode) << PTE_MAS2_SHIFT;
2478 		tlb0_flush_entry(va);
2479 	}
2480 	tlb_miss_unlock();
2481 	mtx_unlock_spin(&tlbivax_mutex);
2482 
2483 	return (0);
2484 }
2485 
2486 static void
mmu_booke_page_array_startup(long pages)2487 mmu_booke_page_array_startup(long pages)
2488 {
2489 	vm_page_array_size = pages;
2490 }
2491 
2492 /**************************************************************************/
2493 /* TID handling */
2494 /**************************************************************************/
2495 
2496 static __inline void
tid_set_busy(int cpu,int tid,pmap_t pmap)2497 tid_set_busy(int cpu, int tid, pmap_t pmap)
2498 {
2499 	volatile pmap_t *pm = &tidbusy[cpu * (tid_max + 1) + tid];
2500 
2501 	if (pmap == NULL) {
2502 		if (*pm != NULL)
2503 			(*pm)->pm_tid[cpu] = TID_NONE;
2504 	} else
2505 		pmap->pm_tid[cpu] = tid;
2506 	*pm = pmap;
2507 }
2508 
2509 static __inline pmap_t
tid_get_busy(int cpu,int tid)2510 tid_get_busy(int cpu, int tid)
2511 {
2512 	return (tidbusy[cpu * (tid_max + 1) + tid]);
2513 }
2514 
2515 /*
2516  * Allocate a TID. If necessary, steal one from someone else.
2517  * The new TID is flushed from the TLB before returning.
2518  */
2519 static tlbtid_t
tid_alloc(pmap_t pmap)2520 tid_alloc(pmap_t pmap)
2521 {
2522 	tlbtid_t tid;
2523 	int thiscpu;
2524 
2525 	KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2526 
2527 	CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
2528 
2529 	thiscpu = PCPU_GET(cpuid);
2530 
2531 	tid = PCPU_GET(booke.tid_next);
2532 	/* tid_max is always a power-of-2-minus-1, so check for overflow. */
2533 	if ((tid & ~tid_max) != 0)
2534 		tid = TID_MIN;
2535 	PCPU_SET(booke.tid_next, tid + 1);
2536 
2537 	/* If we are stealing TID then clear the relevant pmap's field */
2538 	if (tid_get_busy(thiscpu, tid) != NULL) {
2539 		CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
2540 
2541 		tid_set_busy(thiscpu, tid, NULL);
2542 
2543 		/* Flush all entries from TLB0 matching this TID. */
2544 		tid_flush(tid);
2545 	}
2546 
2547 	tid_set_busy(thiscpu, tid, pmap);
2548 	pmap->pm_tid[thiscpu] = tid;
2549 	__asm __volatile("msync; isync");
2550 
2551 	CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
2552 	    PCPU_GET(booke.tid_next));
2553 
2554 	return (tid);
2555 }
2556 
2557 /**************************************************************************/
2558 /* TLB0 handling */
2559 /**************************************************************************/
2560 
2561 /* Convert TLB0 va and way number to tlb0[] table index. */
2562 static inline unsigned int
tlb0_tableidx(vm_offset_t va,unsigned int way)2563 tlb0_tableidx(vm_offset_t va, unsigned int way)
2564 {
2565 	unsigned int idx;
2566 
2567 	idx = (way * TLB0_ENTRIES_PER_WAY);
2568 	idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
2569 	return (idx);
2570 }
2571 
2572 /*
2573  * Invalidate TLB0 entry.
2574  */
2575 static inline void
tlb0_flush_entry(vm_offset_t va)2576 tlb0_flush_entry(vm_offset_t va)
2577 {
2578 
2579 	CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
2580 
2581 	mtx_assert(&tlbivax_mutex, MA_OWNED);
2582 
2583 	__asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
2584 	__asm __volatile("isync; msync");
2585 	__asm __volatile("tlbsync; msync");
2586 
2587 	CTR1(KTR_PMAP, "%s: e", __func__);
2588 }
2589 
2590 /**************************************************************************/
2591 /* TLB1 handling */
2592 /**************************************************************************/
2593 
2594 /*
2595  * TLB1 mapping notes:
2596  *
2597  * TLB1[0]	Kernel text and data.
2598  * TLB1[1-15]	Additional kernel text and data mappings (if required), PCI
2599  *		windows, other devices mappings.
2600  */
2601 
2602  /*
2603  * Read an entry from given TLB1 slot.
2604  */
2605 void
tlb1_read_entry(tlb_entry_t * entry,unsigned int slot)2606 tlb1_read_entry(tlb_entry_t *entry, unsigned int slot)
2607 {
2608 	register_t msr;
2609 	uint32_t mas0;
2610 
2611 	KASSERT((entry != NULL), ("%s(): Entry is NULL!", __func__));
2612 
2613 	msr = mfmsr();
2614 	__asm __volatile("wrteei 0");
2615 
2616 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(slot);
2617 	mtspr(SPR_MAS0, mas0);
2618 	__asm __volatile("isync; tlbre");
2619 
2620 	entry->mas1 = mfspr(SPR_MAS1);
2621 	entry->mas2 = mfspr(SPR_MAS2);
2622 	entry->mas3 = mfspr(SPR_MAS3);
2623 
2624 	switch ((mfpvr() >> 16) & 0xFFFF) {
2625 	case FSL_E500v2:
2626 	case FSL_E500mc:
2627 	case FSL_E5500:
2628 	case FSL_E6500:
2629 		entry->mas7 = mfspr(SPR_MAS7);
2630 		break;
2631 	default:
2632 		entry->mas7 = 0;
2633 		break;
2634 	}
2635 	__asm __volatile("wrtee %0" :: "r"(msr));
2636 
2637 	entry->virt = entry->mas2 & MAS2_EPN_MASK;
2638 	entry->phys = ((vm_paddr_t)(entry->mas7 & MAS7_RPN) << 32) |
2639 	    (entry->mas3 & MAS3_RPN);
2640 	entry->size =
2641 	    tsize2size((entry->mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT);
2642 }
2643 
2644 struct tlbwrite_args {
2645 	tlb_entry_t *e;
2646 	unsigned int idx;
2647 };
2648 
2649 static uint32_t
tlb1_find_free(void)2650 tlb1_find_free(void)
2651 {
2652 	tlb_entry_t e;
2653 	int i;
2654 
2655 	for (i = 0; i < TLB1_ENTRIES; i++) {
2656 		tlb1_read_entry(&e, i);
2657 		if ((e.mas1 & MAS1_VALID) == 0)
2658 			return (i);
2659 	}
2660 	return (-1);
2661 }
2662 
2663 static void
tlb1_purge_va_range(vm_offset_t va,vm_size_t size)2664 tlb1_purge_va_range(vm_offset_t va, vm_size_t size)
2665 {
2666 	tlb_entry_t e;
2667 	int i;
2668 
2669 	for (i = 0; i < TLB1_ENTRIES; i++) {
2670 		tlb1_read_entry(&e, i);
2671 		if ((e.mas1 & MAS1_VALID) == 0)
2672 			continue;
2673 		if ((e.mas2 & MAS2_EPN_MASK) >= va &&
2674 		    (e.mas2 & MAS2_EPN_MASK) < va + size) {
2675 			mtspr(SPR_MAS1, e.mas1 & ~MAS1_VALID);
2676 			__asm __volatile("isync; tlbwe; isync; msync");
2677 		}
2678 	}
2679 }
2680 
2681 static void
tlb1_write_entry_int(void * arg)2682 tlb1_write_entry_int(void *arg)
2683 {
2684 	struct tlbwrite_args *args = arg;
2685 	uint32_t idx, mas0;
2686 
2687 	idx = args->idx;
2688 	if (idx == -1) {
2689 		tlb1_purge_va_range(args->e->virt, args->e->size);
2690 		idx = tlb1_find_free();
2691 		if (idx == -1)
2692 			panic("No free TLB1 entries!\n");
2693 	}
2694 	/* Select entry */
2695 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
2696 
2697 	mtspr(SPR_MAS0, mas0);
2698 	mtspr(SPR_MAS1, args->e->mas1);
2699 	mtspr(SPR_MAS2, args->e->mas2);
2700 	mtspr(SPR_MAS3, args->e->mas3);
2701 	switch ((mfpvr() >> 16) & 0xFFFF) {
2702 	case FSL_E500mc:
2703 	case FSL_E5500:
2704 	case FSL_E6500:
2705 		mtspr(SPR_MAS8, 0);
2706 		/* FALLTHROUGH */
2707 	case FSL_E500v2:
2708 		mtspr(SPR_MAS7, args->e->mas7);
2709 		break;
2710 	default:
2711 		break;
2712 	}
2713 
2714 	__asm __volatile("isync; tlbwe; isync; msync");
2715 
2716 }
2717 
2718 static void
tlb1_write_entry_sync(void * arg)2719 tlb1_write_entry_sync(void *arg)
2720 {
2721 	/* Empty synchronization point for smp_rendezvous(). */
2722 }
2723 
2724 /*
2725  * Write given entry to TLB1 hardware.
2726  */
2727 static void
tlb1_write_entry(tlb_entry_t * e,unsigned int idx)2728 tlb1_write_entry(tlb_entry_t *e, unsigned int idx)
2729 {
2730 	struct tlbwrite_args args;
2731 
2732 	args.e = e;
2733 	args.idx = idx;
2734 
2735 #ifdef SMP
2736 	if ((e->mas2 & _TLB_ENTRY_SHARED) && smp_started) {
2737 		mb();
2738 		smp_rendezvous(tlb1_write_entry_sync,
2739 		    tlb1_write_entry_int,
2740 		    tlb1_write_entry_sync, &args);
2741 	} else
2742 #endif
2743 	{
2744 		register_t msr;
2745 
2746 		msr = mfmsr();
2747 		__asm __volatile("wrteei 0");
2748 		tlb1_write_entry_int(&args);
2749 		__asm __volatile("wrtee %0" :: "r"(msr));
2750 	}
2751 }
2752 
2753 /*
2754  * Convert TLB TSIZE value to mapped region size.
2755  */
2756 static vm_size_t
tsize2size(unsigned int tsize)2757 tsize2size(unsigned int tsize)
2758 {
2759 
2760 	/*
2761 	 * size = 4^tsize KB
2762 	 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
2763 	 */
2764 
2765 	return ((1UL << tsize) * 1024);
2766 }
2767 
2768 /*
2769  * Convert region size (must be power of 4) to TLB TSIZE value.
2770  */
2771 static unsigned int
size2tsize(vm_size_t size)2772 size2tsize(vm_size_t size)
2773 {
2774 
2775 	return (ilog2(size) - 10);
2776 }
2777 
2778 /*
2779  * Register permanent kernel mapping in TLB1.
2780  *
2781  * Entries are created starting from index 0 (current free entry is
2782  * kept in tlb1_idx) and are not supposed to be invalidated.
2783  */
2784 int
tlb1_set_entry(vm_offset_t va,vm_paddr_t pa,vm_size_t size,uint32_t flags)2785 tlb1_set_entry(vm_offset_t va, vm_paddr_t pa, vm_size_t size,
2786     uint32_t flags)
2787 {
2788 	tlb_entry_t e;
2789 	uint32_t ts, tid;
2790 	int tsize, index;
2791 
2792 	/* First try to update an existing entry. */
2793 	for (index = 0; index < TLB1_ENTRIES; index++) {
2794 		tlb1_read_entry(&e, index);
2795 		/* Check if we're just updating the flags, and update them. */
2796 		if (e.phys == pa && e.virt == va && e.size == size) {
2797 			e.mas2 = (va & MAS2_EPN_MASK) | flags;
2798 			tlb1_write_entry(&e, index);
2799 			return (0);
2800 		}
2801 	}
2802 
2803 	/* Convert size to TSIZE */
2804 	tsize = size2tsize(size);
2805 
2806 	tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
2807 	/* XXX TS is hard coded to 0 for now as we only use single address space */
2808 	ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
2809 
2810 	e.phys = pa;
2811 	e.virt = va;
2812 	e.size = size;
2813 	e.mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
2814 	e.mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
2815 	e.mas2 = (va & MAS2_EPN_MASK) | flags;
2816 
2817 	/* Set supervisor RWX permission bits */
2818 	e.mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
2819 	e.mas7 = (pa >> 32) & MAS7_RPN;
2820 
2821 	tlb1_write_entry(&e, -1);
2822 
2823 	return (0);
2824 }
2825 
2826 /*
2827  * Map in contiguous RAM region into the TLB1.
2828  */
2829 static vm_size_t
tlb1_mapin_region(vm_offset_t va,vm_paddr_t pa,vm_size_t size,int wimge)2830 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size, int wimge)
2831 {
2832 	vm_offset_t base;
2833 	vm_size_t mapped, sz, ssize;
2834 	int shift;
2835 
2836 	mapped = 0;
2837 	base = va;
2838 	ssize = size;
2839 
2840 	if (mmuv2)
2841 		shift = 1;
2842 	else
2843 		shift = 2;
2844 
2845 	while (size > 0) {
2846 		sz = 1UL << (ilog2(size) & ~(shift - 1));
2847 		/* Align size to PA */
2848 		if (pa % sz != 0) {
2849 			do {
2850 				sz >>= shift;
2851 			} while (pa % sz != 0);
2852 		}
2853 		/* Now align from there to VA */
2854 		if (va % sz != 0) {
2855 			do {
2856 				sz >>= shift;
2857 			} while (va % sz != 0);
2858 		}
2859 #ifdef __powerpc64__
2860 		/*
2861 		 * Clamp TLB1 entries to 4G.
2862 		 *
2863 		 * While the e6500 supports up to 1TB mappings, the e5500
2864 		 * only supports up to 4G mappings. (0b1011)
2865 		 *
2866 		 * If any e6500 machines capable of supporting a very
2867 		 * large amount of memory appear in the future, we can
2868 		 * revisit this.
2869 		 *
2870 		 * For now, though, since we have plenty of space in TLB1,
2871 		 * always avoid creating entries larger than 4GB.
2872 		 */
2873 		if (!mmuv2)
2874 			sz = MIN(sz, 1UL << 32);
2875 #endif
2876 		if (bootverbose)
2877 			printf("Wiring VA=%p to PA=%jx (size=%lx)\n",
2878 			    (void *)va, (uintmax_t)pa, (long)sz);
2879 		if (tlb1_set_entry(va, pa, sz,
2880 		    _TLB_ENTRY_SHARED | wimge) < 0)
2881 			return (mapped);
2882 		size -= sz;
2883 		pa += sz;
2884 		va += sz;
2885 	}
2886 
2887 	mapped = (va - base);
2888 	if (bootverbose)
2889 		printf("mapped size 0x%"PRIxPTR" (wasted space 0x%"PRIxPTR")\n",
2890 		    mapped, mapped - ssize);
2891 
2892 	return (mapped);
2893 }
2894 
2895 /*
2896  * TLB1 initialization routine, to be called after the very first
2897  * assembler level setup done in locore.S.
2898  */
2899 void
tlb1_init(void)2900 tlb1_init(void)
2901 {
2902 	vm_offset_t mas2;
2903 	uint32_t mas0, mas1, mas3, mas7;
2904 	uint32_t tsz;
2905 
2906 	tlb1_get_tlbconf();
2907 
2908 	mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(0);
2909 	mtspr(SPR_MAS0, mas0);
2910 	__asm __volatile("isync; tlbre");
2911 
2912 	mas1 = mfspr(SPR_MAS1);
2913 	mas2 = mfspr(SPR_MAS2);
2914 	mas3 = mfspr(SPR_MAS3);
2915 	mas7 = mfspr(SPR_MAS7);
2916 
2917 	kernload =  ((vm_paddr_t)(mas7 & MAS7_RPN) << 32) |
2918 	    (mas3 & MAS3_RPN);
2919 
2920 	tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
2921 	kernsize += (tsz > 0) ? tsize2size(tsz) : 0;
2922 	kernstart = trunc_page(mas2);
2923 
2924 	/* Setup TLB miss defaults */
2925 	set_mas4_defaults();
2926 }
2927 
2928 /*
2929  * pmap_early_io_unmap() should be used in short conjunction with
2930  * pmap_early_io_map(), as in the following snippet:
2931  *
2932  * x = pmap_early_io_map(...);
2933  * <do something with x>
2934  * pmap_early_io_unmap(x, size);
2935  *
2936  * And avoiding more allocations between.
2937  */
2938 void
pmap_early_io_unmap(vm_offset_t va,vm_size_t size)2939 pmap_early_io_unmap(vm_offset_t va, vm_size_t size)
2940 {
2941 	int i;
2942 	tlb_entry_t e;
2943 	vm_size_t isize;
2944 
2945 	size = roundup(size, PAGE_SIZE);
2946 	isize = size;
2947 	for (i = 0; i < TLB1_ENTRIES && size > 0; i++) {
2948 		tlb1_read_entry(&e, i);
2949 		if (!(e.mas1 & MAS1_VALID))
2950 			continue;
2951 		if (va <= e.virt && (va + isize) >= (e.virt + e.size)) {
2952 			size -= e.size;
2953 			e.mas1 &= ~MAS1_VALID;
2954 			tlb1_write_entry(&e, i);
2955 		}
2956 	}
2957 	if (tlb1_map_base == va + isize)
2958 		tlb1_map_base -= isize;
2959 }
2960 
2961 vm_offset_t
pmap_early_io_map(vm_paddr_t pa,vm_size_t size)2962 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
2963 {
2964 	vm_paddr_t pa_base;
2965 	vm_offset_t va, sz;
2966 	int i;
2967 	tlb_entry_t e;
2968 
2969 	KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
2970 
2971 	for (i = 0; i < TLB1_ENTRIES; i++) {
2972 		tlb1_read_entry(&e, i);
2973 		if (!(e.mas1 & MAS1_VALID))
2974 			continue;
2975 		if (pa >= e.phys && (pa + size) <=
2976 		    (e.phys + e.size))
2977 			return (e.virt + (pa - e.phys));
2978 	}
2979 
2980 	pa_base = rounddown(pa, PAGE_SIZE);
2981 	size = roundup(size + (pa - pa_base), PAGE_SIZE);
2982 	tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
2983 	va = tlb1_map_base + (pa - pa_base);
2984 
2985 	do {
2986 		sz = 1 << (ilog2(size) & ~1);
2987 		tlb1_set_entry(tlb1_map_base, pa_base, sz,
2988 		    _TLB_ENTRY_SHARED | _TLB_ENTRY_IO);
2989 		size -= sz;
2990 		pa_base += sz;
2991 		tlb1_map_base += sz;
2992 	} while (size > 0);
2993 
2994 	return (va);
2995 }
2996 
2997 void
pmap_track_page(pmap_t pmap,vm_offset_t va)2998 pmap_track_page(pmap_t pmap, vm_offset_t va)
2999 {
3000 	vm_paddr_t pa;
3001 	vm_page_t page;
3002 	struct pv_entry *pve;
3003 
3004 	va = trunc_page(va);
3005 	pa = pmap_kextract(va);
3006 	page = PHYS_TO_VM_PAGE(pa);
3007 
3008 	rw_wlock(&pvh_global_lock);
3009 	PMAP_LOCK(pmap);
3010 
3011 	TAILQ_FOREACH(pve, &page->md.pv_list, pv_link) {
3012 		if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
3013 			goto out;
3014 		}
3015 	}
3016 	page->md.pv_tracked = true;
3017 	pv_insert(pmap, va, page);
3018 out:
3019 	PMAP_UNLOCK(pmap);
3020 	rw_wunlock(&pvh_global_lock);
3021 }
3022 
3023 /*
3024  * Setup MAS4 defaults.
3025  * These values are loaded to MAS0-2 on a TLB miss.
3026  */
3027 static void
set_mas4_defaults(void)3028 set_mas4_defaults(void)
3029 {
3030 	uint32_t mas4;
3031 
3032 	/* Defaults: TLB0, PID0, TSIZED=4K */
3033 	mas4 = MAS4_TLBSELD0;
3034 	mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3035 #ifdef SMP
3036 	mas4 |= MAS4_MD;
3037 #endif
3038 	mtspr(SPR_MAS4, mas4);
3039 	__asm __volatile("isync");
3040 }
3041 
3042 /*
3043  * Return 0 if the physical IO range is encompassed by one of the
3044  * the TLB1 entries, otherwise return related error code.
3045  */
3046 static int
tlb1_iomapped(int i,vm_paddr_t pa,vm_size_t size,vm_offset_t * va)3047 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3048 {
3049 	uint32_t prot;
3050 	vm_paddr_t pa_start;
3051 	vm_paddr_t pa_end;
3052 	unsigned int entry_tsize;
3053 	vm_size_t entry_size;
3054 	tlb_entry_t e;
3055 
3056 	*va = (vm_offset_t)NULL;
3057 
3058 	tlb1_read_entry(&e, i);
3059 	/* Skip invalid entries */
3060 	if (!(e.mas1 & MAS1_VALID))
3061 		return (EINVAL);
3062 
3063 	/*
3064 	 * The entry must be cache-inhibited, guarded, and r/w
3065 	 * so it can function as an i/o page
3066 	 */
3067 	prot = e.mas2 & (MAS2_I | MAS2_G);
3068 	if (prot != (MAS2_I | MAS2_G))
3069 		return (EPERM);
3070 
3071 	prot = e.mas3 & (MAS3_SR | MAS3_SW);
3072 	if (prot != (MAS3_SR | MAS3_SW))
3073 		return (EPERM);
3074 
3075 	/* The address should be within the entry range. */
3076 	entry_tsize = (e.mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3077 	KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3078 
3079 	entry_size = tsize2size(entry_tsize);
3080 	pa_start = (((vm_paddr_t)e.mas7 & MAS7_RPN) << 32) |
3081 	    (e.mas3 & MAS3_RPN);
3082 	pa_end = pa_start + entry_size;
3083 
3084 	if ((pa < pa_start) || ((pa + size) > pa_end))
3085 		return (ERANGE);
3086 
3087 	/* Return virtual address of this mapping. */
3088 	*va = (e.mas2 & MAS2_EPN_MASK) + (pa - pa_start);
3089 	return (0);
3090 }
3091 
3092 #ifdef DDB
3093 /* Print out contents of the MAS registers for each TLB0 entry */
3094 static void
3095 #ifdef __powerpc64__
tlb_print_entry(int i,uint32_t mas1,uint64_t mas2,uint32_t mas3,uint32_t mas7)3096 tlb_print_entry(int i, uint32_t mas1, uint64_t mas2, uint32_t mas3,
3097 #else
3098 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
3099 #endif
3100     uint32_t mas7)
3101 {
3102 	int as;
3103 	char desc[3];
3104 	tlbtid_t tid;
3105 	vm_size_t size;
3106 	unsigned int tsize;
3107 
3108 	desc[2] = '\0';
3109 	if (mas1 & MAS1_VALID)
3110 		desc[0] = 'V';
3111 	else
3112 		desc[0] = ' ';
3113 
3114 	if (mas1 & MAS1_IPROT)
3115 		desc[1] = 'P';
3116 	else
3117 		desc[1] = ' ';
3118 
3119 	as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
3120 	tid = MAS1_GETTID(mas1);
3121 
3122 	tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3123 	size = 0;
3124 	if (tsize)
3125 		size = tsize2size(tsize);
3126 
3127 	printf("%3d: (%s) [AS=%d] "
3128 	    "sz = 0x%jx tsz = %d tid = %d mas1 = 0x%08x "
3129 	    "mas2(va) = 0x%"PRI0ptrX" mas3(pa) = 0x%08x mas7 = 0x%08x\n",
3130 	    i, desc, as, (uintmax_t)size, tsize, tid, mas1, mas2, mas3, mas7);
3131 }
3132 
DB_SHOW_COMMAND(tlb0,tlb0_print_tlbentries)3133 DB_SHOW_COMMAND(tlb0, tlb0_print_tlbentries)
3134 {
3135 	uint32_t mas0, mas1, mas3, mas7;
3136 #ifdef __powerpc64__
3137 	uint64_t mas2;
3138 #else
3139 	uint32_t mas2;
3140 #endif
3141 	int entryidx, way, idx;
3142 
3143 	printf("TLB0 entries:\n");
3144 	for (way = 0; way < TLB0_WAYS; way ++)
3145 		for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
3146 			mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
3147 			mtspr(SPR_MAS0, mas0);
3148 
3149 			mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
3150 			mtspr(SPR_MAS2, mas2);
3151 
3152 			__asm __volatile("isync; tlbre");
3153 
3154 			mas1 = mfspr(SPR_MAS1);
3155 			mas2 = mfspr(SPR_MAS2);
3156 			mas3 = mfspr(SPR_MAS3);
3157 			mas7 = mfspr(SPR_MAS7);
3158 
3159 			idx = tlb0_tableidx(mas2, way);
3160 			tlb_print_entry(idx, mas1, mas2, mas3, mas7);
3161 		}
3162 }
3163 
3164 /*
3165  * Print out contents of the MAS registers for each TLB1 entry
3166  */
DB_SHOW_COMMAND(tlb1,tlb1_print_tlbentries)3167 DB_SHOW_COMMAND(tlb1, tlb1_print_tlbentries)
3168 {
3169 	uint32_t mas0, mas1, mas3, mas7;
3170 #ifdef __powerpc64__
3171 	uint64_t mas2;
3172 #else
3173 	uint32_t mas2;
3174 #endif
3175 	int i;
3176 
3177 	printf("TLB1 entries:\n");
3178 	for (i = 0; i < TLB1_ENTRIES; i++) {
3179 		mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3180 		mtspr(SPR_MAS0, mas0);
3181 
3182 		__asm __volatile("isync; tlbre");
3183 
3184 		mas1 = mfspr(SPR_MAS1);
3185 		mas2 = mfspr(SPR_MAS2);
3186 		mas3 = mfspr(SPR_MAS3);
3187 		mas7 = mfspr(SPR_MAS7);
3188 
3189 		tlb_print_entry(i, mas1, mas2, mas3, mas7);
3190 	}
3191 }
3192 #endif
3193