2 * Copyright (C) 2007-2009 Semihalf, Rafal Jaworowski <raj@semihalf.com>
3 * Copyright (C) 2006 Semihalf, Marian Balakowicz <m8@semihalf.com>
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN
18 * NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
19 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
20 * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
21 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
22 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
23 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
24 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26 * Some hw specific parts of this pmap were derived or influenced
27 * by NetBSD's ibm4xx pmap module. More generic code is shared with
28 * a few other pmap modules from the FreeBSD tree.
34 * Kernel and user threads run within one common virtual address space
37 * Virtual address space layout:
38 * -----------------------------
39 * 0x0000_0000 - 0xafff_ffff : user process
40 * 0xb000_0000 - 0xbfff_ffff : pmap_mapdev()-ed area (PCI/PCIE etc.)
41 * 0xc000_0000 - 0xc0ff_ffff : kernel reserved
42 * 0xc000_0000 - data_end : kernel code+data, env, metadata etc.
43 * 0xc100_0000 - 0xfeef_ffff : KVA
44 * 0xc100_0000 - 0xc100_3fff : reserved for page zero/copy
45 * 0xc100_4000 - 0xc200_3fff : reserved for ptbl bufs
46 * 0xc200_4000 - 0xc200_8fff : guard page + kstack0
47 * 0xc200_9000 - 0xfeef_ffff : actual free KVA space
48 * 0xfef0_0000 - 0xffff_ffff : I/O devices region
51 #include <sys/cdefs.h>
52 __FBSDID("$FreeBSD$");
54 #include <sys/param.h>
56 #include <sys/malloc.h>
60 #include <sys/queue.h>
61 #include <sys/systm.h>
62 #include <sys/kernel.h>
63 #include <sys/kerneldump.h>
64 #include <sys/linker.h>
65 #include <sys/msgbuf.h>
67 #include <sys/mutex.h>
68 #include <sys/rwlock.h>
69 #include <sys/sched.h>
71 #include <sys/vmmeter.h>
74 #include <vm/vm_page.h>
75 #include <vm/vm_kern.h>
76 #include <vm/vm_pageout.h>
77 #include <vm/vm_extern.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_param.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_pager.h>
84 #include <machine/cpu.h>
85 #include <machine/pcb.h>
86 #include <machine/platform.h>
88 #include <machine/tlb.h>
89 #include <machine/spr.h>
90 #include <machine/md_var.h>
91 #include <machine/mmuvar.h>
92 #include <machine/pmap.h>
93 #include <machine/pte.h>
98 #define debugf(fmt, args...) printf(fmt, ##args)
100 #define debugf(fmt, args...)
103 #define TODO panic("%s: not implemented", __func__);
105 extern unsigned char _etext[];
106 extern unsigned char _end[];
108 extern uint32_t *bootinfo;
111 extern uint32_t bp_ntlb1s;
115 vm_offset_t kernstart;
118 /* Message buffer and tables. */
119 static vm_offset_t data_start;
120 static vm_size_t data_end;
122 /* Phys/avail memory regions. */
123 static struct mem_region *availmem_regions;
124 static int availmem_regions_sz;
125 static struct mem_region *physmem_regions;
126 static int physmem_regions_sz;
128 /* Reserved KVA space and mutex for mmu_booke_zero_page. */
129 static vm_offset_t zero_page_va;
130 static struct mtx zero_page_mutex;
132 static struct mtx tlbivax_mutex;
135 * Reserved KVA space for mmu_booke_zero_page_idle. This is used
136 * by idle thred only, no lock required.
138 static vm_offset_t zero_page_idle_va;
140 /* Reserved KVA space and mutex for mmu_booke_copy_page. */
141 static vm_offset_t copy_page_src_va;
142 static vm_offset_t copy_page_dst_va;
143 static struct mtx copy_page_mutex;
145 /**************************************************************************/
147 /**************************************************************************/
149 static int mmu_booke_enter_locked(mmu_t, pmap_t, vm_offset_t, vm_page_t,
150 vm_prot_t, u_int flags, int8_t psind);
152 unsigned int kptbl_min; /* Index of the first kernel ptbl. */
153 unsigned int kernel_ptbls; /* Number of KVA ptbls. */
156 * If user pmap is processed with mmu_booke_remove and the resident count
157 * drops to 0, there are no more pages to remove, so we need not continue.
159 #define PMAP_REMOVE_DONE(pmap) \
160 ((pmap) != kernel_pmap && (pmap)->pm_stats.resident_count == 0)
162 extern void tid_flush(tlbtid_t tid, int tlb0_ways, int tlb0_entries_per_way);
163 extern int elf32_nxstack;
165 /**************************************************************************/
166 /* TLB and TID handling */
167 /**************************************************************************/
169 /* Translation ID busy table */
170 static volatile pmap_t tidbusy[MAXCPU][TID_MAX + 1];
173 * TLB0 capabilities (entry, way numbers etc.). These can vary between e500
174 * core revisions and should be read from h/w registers during early config.
176 uint32_t tlb0_entries;
178 uint32_t tlb0_entries_per_way;
180 #define TLB0_ENTRIES (tlb0_entries)
181 #define TLB0_WAYS (tlb0_ways)
182 #define TLB0_ENTRIES_PER_WAY (tlb0_entries_per_way)
184 #define TLB1_ENTRIES 16
186 /* In-ram copy of the TLB1 */
187 static tlb_entry_t tlb1[TLB1_ENTRIES];
189 /* Next free entry in the TLB1 */
190 static unsigned int tlb1_idx;
191 static vm_offset_t tlb1_map_base = VM_MAX_KERNEL_ADDRESS;
193 static tlbtid_t tid_alloc(struct pmap *);
195 static void tlb_print_entry(int, uint32_t, uint32_t, uint32_t, uint32_t);
197 static int tlb1_set_entry(vm_offset_t, vm_offset_t, vm_size_t, uint32_t);
198 static void tlb1_write_entry(unsigned int);
199 static int tlb1_iomapped(int, vm_paddr_t, vm_size_t, vm_offset_t *);
200 static vm_size_t tlb1_mapin_region(vm_offset_t, vm_paddr_t, vm_size_t);
202 static vm_size_t tsize2size(unsigned int);
203 static unsigned int size2tsize(vm_size_t);
204 static unsigned int ilog2(unsigned int);
206 static void set_mas4_defaults(void);
208 static inline void tlb0_flush_entry(vm_offset_t);
209 static inline unsigned int tlb0_tableidx(vm_offset_t, unsigned int);
211 /**************************************************************************/
212 /* Page table management */
213 /**************************************************************************/
215 static struct rwlock_padalign pvh_global_lock;
217 /* Data for the pv entry allocation mechanism */
218 static uma_zone_t pvzone;
219 static int pv_entry_count = 0, pv_entry_max = 0, pv_entry_high_water = 0;
221 #define PV_ENTRY_ZONE_MIN 2048 /* min pv entries in uma zone */
223 #ifndef PMAP_SHPGPERPROC
224 #define PMAP_SHPGPERPROC 200
227 static void ptbl_init(void);
228 static struct ptbl_buf *ptbl_buf_alloc(void);
229 static void ptbl_buf_free(struct ptbl_buf *);
230 static void ptbl_free_pmap_ptbl(pmap_t, pte_t *);
232 static pte_t *ptbl_alloc(mmu_t, pmap_t, unsigned int, boolean_t);
233 static void ptbl_free(mmu_t, pmap_t, unsigned int);
234 static void ptbl_hold(mmu_t, pmap_t, unsigned int);
235 static int ptbl_unhold(mmu_t, pmap_t, unsigned int);
237 static vm_paddr_t pte_vatopa(mmu_t, pmap_t, vm_offset_t);
238 static pte_t *pte_find(mmu_t, pmap_t, vm_offset_t);
239 static int pte_enter(mmu_t, pmap_t, vm_page_t, vm_offset_t, uint32_t, boolean_t);
240 static int pte_remove(mmu_t, pmap_t, vm_offset_t, uint8_t);
242 static pv_entry_t pv_alloc(void);
243 static void pv_free(pv_entry_t);
244 static void pv_insert(pmap_t, vm_offset_t, vm_page_t);
245 static void pv_remove(pmap_t, vm_offset_t, vm_page_t);
247 /* Number of kva ptbl buffers, each covering one ptbl (PTBL_PAGES). */
248 #define PTBL_BUFS (128 * 16)
251 TAILQ_ENTRY(ptbl_buf) link; /* list link */
252 vm_offset_t kva; /* va of mapping */
255 /* ptbl free list and a lock used for access synchronization. */
256 static TAILQ_HEAD(, ptbl_buf) ptbl_buf_freelist;
257 static struct mtx ptbl_buf_freelist_lock;
259 /* Base address of kva space allocated fot ptbl bufs. */
260 static vm_offset_t ptbl_buf_pool_vabase;
262 /* Pointer to ptbl_buf structures. */
263 static struct ptbl_buf *ptbl_bufs;
266 void pmap_bootstrap_ap(volatile uint32_t *);
270 * Kernel MMU interface
272 static void mmu_booke_clear_modify(mmu_t, vm_page_t);
273 static void mmu_booke_copy(mmu_t, pmap_t, pmap_t, vm_offset_t,
274 vm_size_t, vm_offset_t);
275 static void mmu_booke_copy_page(mmu_t, vm_page_t, vm_page_t);
276 static void mmu_booke_copy_pages(mmu_t, vm_page_t *,
277 vm_offset_t, vm_page_t *, vm_offset_t, int);
278 static int mmu_booke_enter(mmu_t, pmap_t, vm_offset_t, vm_page_t,
279 vm_prot_t, u_int flags, int8_t psind);
280 static void mmu_booke_enter_object(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
281 vm_page_t, vm_prot_t);
282 static void mmu_booke_enter_quick(mmu_t, pmap_t, vm_offset_t, vm_page_t,
284 static vm_paddr_t mmu_booke_extract(mmu_t, pmap_t, vm_offset_t);
285 static vm_page_t mmu_booke_extract_and_hold(mmu_t, pmap_t, vm_offset_t,
287 static void mmu_booke_init(mmu_t);
288 static boolean_t mmu_booke_is_modified(mmu_t, vm_page_t);
289 static boolean_t mmu_booke_is_prefaultable(mmu_t, pmap_t, vm_offset_t);
290 static boolean_t mmu_booke_is_referenced(mmu_t, vm_page_t);
291 static int mmu_booke_ts_referenced(mmu_t, vm_page_t);
292 static vm_offset_t mmu_booke_map(mmu_t, vm_offset_t *, vm_paddr_t, vm_paddr_t,
294 static int mmu_booke_mincore(mmu_t, pmap_t, vm_offset_t,
296 static void mmu_booke_object_init_pt(mmu_t, pmap_t, vm_offset_t,
297 vm_object_t, vm_pindex_t, vm_size_t);
298 static boolean_t mmu_booke_page_exists_quick(mmu_t, pmap_t, vm_page_t);
299 static void mmu_booke_page_init(mmu_t, vm_page_t);
300 static int mmu_booke_page_wired_mappings(mmu_t, vm_page_t);
301 static void mmu_booke_pinit(mmu_t, pmap_t);
302 static void mmu_booke_pinit0(mmu_t, pmap_t);
303 static void mmu_booke_protect(mmu_t, pmap_t, vm_offset_t, vm_offset_t,
305 static void mmu_booke_qenter(mmu_t, vm_offset_t, vm_page_t *, int);
306 static void mmu_booke_qremove(mmu_t, vm_offset_t, int);
307 static void mmu_booke_release(mmu_t, pmap_t);
308 static void mmu_booke_remove(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
309 static void mmu_booke_remove_all(mmu_t, vm_page_t);
310 static void mmu_booke_remove_write(mmu_t, vm_page_t);
311 static void mmu_booke_unwire(mmu_t, pmap_t, vm_offset_t, vm_offset_t);
312 static void mmu_booke_zero_page(mmu_t, vm_page_t);
313 static void mmu_booke_zero_page_area(mmu_t, vm_page_t, int, int);
314 static void mmu_booke_zero_page_idle(mmu_t, vm_page_t);
315 static void mmu_booke_activate(mmu_t, struct thread *);
316 static void mmu_booke_deactivate(mmu_t, struct thread *);
317 static void mmu_booke_bootstrap(mmu_t, vm_offset_t, vm_offset_t);
318 static void *mmu_booke_mapdev(mmu_t, vm_paddr_t, vm_size_t);
319 static void *mmu_booke_mapdev_attr(mmu_t, vm_paddr_t, vm_size_t, vm_memattr_t);
320 static void mmu_booke_unmapdev(mmu_t, vm_offset_t, vm_size_t);
321 static vm_paddr_t mmu_booke_kextract(mmu_t, vm_offset_t);
322 static void mmu_booke_kenter(mmu_t, vm_offset_t, vm_paddr_t);
323 static void mmu_booke_kenter_attr(mmu_t, vm_offset_t, vm_paddr_t, vm_memattr_t);
324 static void mmu_booke_kremove(mmu_t, vm_offset_t);
325 static boolean_t mmu_booke_dev_direct_mapped(mmu_t, vm_paddr_t, vm_size_t);
326 static void mmu_booke_sync_icache(mmu_t, pmap_t, vm_offset_t,
328 static void mmu_booke_dumpsys_map(mmu_t, vm_paddr_t pa, size_t,
330 static void mmu_booke_dumpsys_unmap(mmu_t, vm_paddr_t pa, size_t,
332 static void mmu_booke_scan_init(mmu_t);
334 static mmu_method_t mmu_booke_methods[] = {
335 /* pmap dispatcher interface */
336 MMUMETHOD(mmu_clear_modify, mmu_booke_clear_modify),
337 MMUMETHOD(mmu_copy, mmu_booke_copy),
338 MMUMETHOD(mmu_copy_page, mmu_booke_copy_page),
339 MMUMETHOD(mmu_copy_pages, mmu_booke_copy_pages),
340 MMUMETHOD(mmu_enter, mmu_booke_enter),
341 MMUMETHOD(mmu_enter_object, mmu_booke_enter_object),
342 MMUMETHOD(mmu_enter_quick, mmu_booke_enter_quick),
343 MMUMETHOD(mmu_extract, mmu_booke_extract),
344 MMUMETHOD(mmu_extract_and_hold, mmu_booke_extract_and_hold),
345 MMUMETHOD(mmu_init, mmu_booke_init),
346 MMUMETHOD(mmu_is_modified, mmu_booke_is_modified),
347 MMUMETHOD(mmu_is_prefaultable, mmu_booke_is_prefaultable),
348 MMUMETHOD(mmu_is_referenced, mmu_booke_is_referenced),
349 MMUMETHOD(mmu_ts_referenced, mmu_booke_ts_referenced),
350 MMUMETHOD(mmu_map, mmu_booke_map),
351 MMUMETHOD(mmu_mincore, mmu_booke_mincore),
352 MMUMETHOD(mmu_object_init_pt, mmu_booke_object_init_pt),
353 MMUMETHOD(mmu_page_exists_quick,mmu_booke_page_exists_quick),
354 MMUMETHOD(mmu_page_init, mmu_booke_page_init),
355 MMUMETHOD(mmu_page_wired_mappings, mmu_booke_page_wired_mappings),
356 MMUMETHOD(mmu_pinit, mmu_booke_pinit),
357 MMUMETHOD(mmu_pinit0, mmu_booke_pinit0),
358 MMUMETHOD(mmu_protect, mmu_booke_protect),
359 MMUMETHOD(mmu_qenter, mmu_booke_qenter),
360 MMUMETHOD(mmu_qremove, mmu_booke_qremove),
361 MMUMETHOD(mmu_release, mmu_booke_release),
362 MMUMETHOD(mmu_remove, mmu_booke_remove),
363 MMUMETHOD(mmu_remove_all, mmu_booke_remove_all),
364 MMUMETHOD(mmu_remove_write, mmu_booke_remove_write),
365 MMUMETHOD(mmu_sync_icache, mmu_booke_sync_icache),
366 MMUMETHOD(mmu_unwire, mmu_booke_unwire),
367 MMUMETHOD(mmu_zero_page, mmu_booke_zero_page),
368 MMUMETHOD(mmu_zero_page_area, mmu_booke_zero_page_area),
369 MMUMETHOD(mmu_zero_page_idle, mmu_booke_zero_page_idle),
370 MMUMETHOD(mmu_activate, mmu_booke_activate),
371 MMUMETHOD(mmu_deactivate, mmu_booke_deactivate),
373 /* Internal interfaces */
374 MMUMETHOD(mmu_bootstrap, mmu_booke_bootstrap),
375 MMUMETHOD(mmu_dev_direct_mapped,mmu_booke_dev_direct_mapped),
376 MMUMETHOD(mmu_mapdev, mmu_booke_mapdev),
377 MMUMETHOD(mmu_mapdev_attr, mmu_booke_mapdev_attr),
378 MMUMETHOD(mmu_kenter, mmu_booke_kenter),
379 MMUMETHOD(mmu_kenter_attr, mmu_booke_kenter_attr),
380 MMUMETHOD(mmu_kextract, mmu_booke_kextract),
381 /* MMUMETHOD(mmu_kremove, mmu_booke_kremove), */
382 MMUMETHOD(mmu_unmapdev, mmu_booke_unmapdev),
384 /* dumpsys() support */
385 MMUMETHOD(mmu_dumpsys_map, mmu_booke_dumpsys_map),
386 MMUMETHOD(mmu_dumpsys_unmap, mmu_booke_dumpsys_unmap),
387 MMUMETHOD(mmu_scan_init, mmu_booke_scan_init),
392 MMU_DEF(booke_mmu, MMU_TYPE_BOOKE, mmu_booke_methods, 0);
394 static __inline uint32_t
395 tlb_calc_wimg(vm_offset_t pa, vm_memattr_t ma)
400 if (ma != VM_MEMATTR_DEFAULT) {
402 case VM_MEMATTR_UNCACHEABLE:
403 return (PTE_I | PTE_G);
404 case VM_MEMATTR_WRITE_COMBINING:
405 case VM_MEMATTR_WRITE_BACK:
406 case VM_MEMATTR_PREFETCHABLE:
408 case VM_MEMATTR_WRITE_THROUGH:
409 return (PTE_W | PTE_M);
414 * Assume the page is cache inhibited and access is guarded unless
415 * it's in our available memory array.
417 attrib = _TLB_ENTRY_IO;
418 for (i = 0; i < physmem_regions_sz; i++) {
419 if ((pa >= physmem_regions[i].mr_start) &&
420 (pa < (physmem_regions[i].mr_start +
421 physmem_regions[i].mr_size))) {
422 attrib = _TLB_ENTRY_MEM;
439 STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
442 CTR3(KTR_PMAP, "%s: tlb miss LOCK of CPU=%d, "
443 "tlb_lock=%p", __func__, pc->pc_cpuid, pc->pc_booke_tlb_lock);
445 KASSERT((pc->pc_cpuid != PCPU_GET(cpuid)),
446 ("tlb_miss_lock: tried to lock self"));
448 tlb_lock(pc->pc_booke_tlb_lock);
450 CTR1(KTR_PMAP, "%s: locked", __func__);
457 tlb_miss_unlock(void)
465 STAILQ_FOREACH(pc, &cpuhead, pc_allcpu) {
467 CTR2(KTR_PMAP, "%s: tlb miss UNLOCK of CPU=%d",
468 __func__, pc->pc_cpuid);
470 tlb_unlock(pc->pc_booke_tlb_lock);
472 CTR1(KTR_PMAP, "%s: unlocked", __func__);
478 /* Return number of entries in TLB0. */
480 tlb0_get_tlbconf(void)
484 tlb0_cfg = mfspr(SPR_TLB0CFG);
485 tlb0_entries = tlb0_cfg & TLBCFG_NENTRY_MASK;
486 tlb0_ways = (tlb0_cfg & TLBCFG_ASSOC_MASK) >> TLBCFG_ASSOC_SHIFT;
487 tlb0_entries_per_way = tlb0_entries / tlb0_ways;
490 /* Initialize pool of kva ptbl buffers. */
496 CTR3(KTR_PMAP, "%s: s (ptbl_bufs = 0x%08x size 0x%08x)", __func__,
497 (uint32_t)ptbl_bufs, sizeof(struct ptbl_buf) * PTBL_BUFS);
498 CTR3(KTR_PMAP, "%s: s (ptbl_buf_pool_vabase = 0x%08x size = 0x%08x)",
499 __func__, ptbl_buf_pool_vabase, PTBL_BUFS * PTBL_PAGES * PAGE_SIZE);
501 mtx_init(&ptbl_buf_freelist_lock, "ptbl bufs lock", NULL, MTX_DEF);
502 TAILQ_INIT(&ptbl_buf_freelist);
504 for (i = 0; i < PTBL_BUFS; i++) {
505 ptbl_bufs[i].kva = ptbl_buf_pool_vabase + i * PTBL_PAGES * PAGE_SIZE;
506 TAILQ_INSERT_TAIL(&ptbl_buf_freelist, &ptbl_bufs[i], link);
510 /* Get a ptbl_buf from the freelist. */
511 static struct ptbl_buf *
514 struct ptbl_buf *buf;
516 mtx_lock(&ptbl_buf_freelist_lock);
517 buf = TAILQ_FIRST(&ptbl_buf_freelist);
519 TAILQ_REMOVE(&ptbl_buf_freelist, buf, link);
520 mtx_unlock(&ptbl_buf_freelist_lock);
522 CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
527 /* Return ptbl buff to free pool. */
529 ptbl_buf_free(struct ptbl_buf *buf)
532 CTR2(KTR_PMAP, "%s: buf = %p", __func__, buf);
534 mtx_lock(&ptbl_buf_freelist_lock);
535 TAILQ_INSERT_TAIL(&ptbl_buf_freelist, buf, link);
536 mtx_unlock(&ptbl_buf_freelist_lock);
540 * Search the list of allocated ptbl bufs and find on list of allocated ptbls
543 ptbl_free_pmap_ptbl(pmap_t pmap, pte_t *ptbl)
545 struct ptbl_buf *pbuf;
547 CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
549 PMAP_LOCK_ASSERT(pmap, MA_OWNED);
551 TAILQ_FOREACH(pbuf, &pmap->pm_ptbl_list, link)
552 if (pbuf->kva == (vm_offset_t)ptbl) {
553 /* Remove from pmap ptbl buf list. */
554 TAILQ_REMOVE(&pmap->pm_ptbl_list, pbuf, link);
556 /* Free corresponding ptbl buf. */
562 /* Allocate page table. */
564 ptbl_alloc(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx, boolean_t nosleep)
566 vm_page_t mtbl[PTBL_PAGES];
568 struct ptbl_buf *pbuf;
573 CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
574 (pmap == kernel_pmap), pdir_idx);
576 KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
577 ("ptbl_alloc: invalid pdir_idx"));
578 KASSERT((pmap->pm_pdir[pdir_idx] == NULL),
579 ("pte_alloc: valid ptbl entry exists!"));
581 pbuf = ptbl_buf_alloc();
583 panic("pte_alloc: couldn't alloc kernel virtual memory");
585 ptbl = (pte_t *)pbuf->kva;
587 CTR2(KTR_PMAP, "%s: ptbl kva = %p", __func__, ptbl);
589 /* Allocate ptbl pages, this will sleep! */
590 for (i = 0; i < PTBL_PAGES; i++) {
591 pidx = (PTBL_PAGES * pdir_idx) + i;
592 while ((m = vm_page_alloc(NULL, pidx,
593 VM_ALLOC_NOOBJ | VM_ALLOC_WIRED)) == NULL) {
595 rw_wunlock(&pvh_global_lock);
597 ptbl_free_pmap_ptbl(pmap, ptbl);
598 for (j = 0; j < i; j++)
599 vm_page_free(mtbl[j]);
600 atomic_subtract_int(&vm_cnt.v_wire_count, i);
604 rw_wlock(&pvh_global_lock);
610 /* Map allocated pages into kernel_pmap. */
611 mmu_booke_qenter(mmu, (vm_offset_t)ptbl, mtbl, PTBL_PAGES);
613 /* Zero whole ptbl. */
614 bzero((caddr_t)ptbl, PTBL_PAGES * PAGE_SIZE);
616 /* Add pbuf to the pmap ptbl bufs list. */
617 TAILQ_INSERT_TAIL(&pmap->pm_ptbl_list, pbuf, link);
622 /* Free ptbl pages and invalidate pdir entry. */
624 ptbl_free(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
632 CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
633 (pmap == kernel_pmap), pdir_idx);
635 KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
636 ("ptbl_free: invalid pdir_idx"));
638 ptbl = pmap->pm_pdir[pdir_idx];
640 CTR2(KTR_PMAP, "%s: ptbl = %p", __func__, ptbl);
642 KASSERT((ptbl != NULL), ("ptbl_free: null ptbl"));
645 * Invalidate the pdir entry as soon as possible, so that other CPUs
646 * don't attempt to look up the page tables we are releasing.
648 mtx_lock_spin(&tlbivax_mutex);
651 pmap->pm_pdir[pdir_idx] = NULL;
654 mtx_unlock_spin(&tlbivax_mutex);
656 for (i = 0; i < PTBL_PAGES; i++) {
657 va = ((vm_offset_t)ptbl + (i * PAGE_SIZE));
658 pa = pte_vatopa(mmu, kernel_pmap, va);
659 m = PHYS_TO_VM_PAGE(pa);
660 vm_page_free_zero(m);
661 atomic_subtract_int(&vm_cnt.v_wire_count, 1);
662 mmu_booke_kremove(mmu, va);
665 ptbl_free_pmap_ptbl(pmap, ptbl);
669 * Decrement ptbl pages hold count and attempt to free ptbl pages.
670 * Called when removing pte entry from ptbl.
672 * Return 1 if ptbl pages were freed.
675 ptbl_unhold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
682 CTR4(KTR_PMAP, "%s: pmap = %p su = %d pdir_idx = %d", __func__, pmap,
683 (pmap == kernel_pmap), pdir_idx);
685 KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
686 ("ptbl_unhold: invalid pdir_idx"));
687 KASSERT((pmap != kernel_pmap),
688 ("ptbl_unhold: unholding kernel ptbl!"));
690 ptbl = pmap->pm_pdir[pdir_idx];
692 //debugf("ptbl_unhold: ptbl = 0x%08x\n", (u_int32_t)ptbl);
693 KASSERT(((vm_offset_t)ptbl >= VM_MIN_KERNEL_ADDRESS),
694 ("ptbl_unhold: non kva ptbl"));
696 /* decrement hold count */
697 for (i = 0; i < PTBL_PAGES; i++) {
698 pa = pte_vatopa(mmu, kernel_pmap,
699 (vm_offset_t)ptbl + (i * PAGE_SIZE));
700 m = PHYS_TO_VM_PAGE(pa);
705 * Free ptbl pages if there are no pte etries in this ptbl.
706 * wire_count has the same value for all ptbl pages, so check the last
709 if (m->wire_count == 0) {
710 ptbl_free(mmu, pmap, pdir_idx);
712 //debugf("ptbl_unhold: e (freed ptbl)\n");
720 * Increment hold count for ptbl pages. This routine is used when a new pte
721 * entry is being inserted into the ptbl.
724 ptbl_hold(mmu_t mmu, pmap_t pmap, unsigned int pdir_idx)
731 CTR3(KTR_PMAP, "%s: pmap = %p pdir_idx = %d", __func__, pmap,
734 KASSERT((pdir_idx <= (VM_MAXUSER_ADDRESS / PDIR_SIZE)),
735 ("ptbl_hold: invalid pdir_idx"));
736 KASSERT((pmap != kernel_pmap),
737 ("ptbl_hold: holding kernel ptbl!"));
739 ptbl = pmap->pm_pdir[pdir_idx];
741 KASSERT((ptbl != NULL), ("ptbl_hold: null ptbl"));
743 for (i = 0; i < PTBL_PAGES; i++) {
744 pa = pte_vatopa(mmu, kernel_pmap,
745 (vm_offset_t)ptbl + (i * PAGE_SIZE));
746 m = PHYS_TO_VM_PAGE(pa);
751 /* Allocate pv_entry structure. */
758 if (pv_entry_count > pv_entry_high_water)
760 pv = uma_zalloc(pvzone, M_NOWAIT);
765 /* Free pv_entry structure. */
767 pv_free(pv_entry_t pve)
771 uma_zfree(pvzone, pve);
775 /* Allocate and initialize pv_entry structure. */
777 pv_insert(pmap_t pmap, vm_offset_t va, vm_page_t m)
781 //int su = (pmap == kernel_pmap);
782 //debugf("pv_insert: s (su = %d pmap = 0x%08x va = 0x%08x m = 0x%08x)\n", su,
783 // (u_int32_t)pmap, va, (u_int32_t)m);
787 panic("pv_insert: no pv entries!");
793 PMAP_LOCK_ASSERT(pmap, MA_OWNED);
794 rw_assert(&pvh_global_lock, RA_WLOCKED);
796 TAILQ_INSERT_TAIL(&m->md.pv_list, pve, pv_link);
798 //debugf("pv_insert: e\n");
801 /* Destroy pv entry. */
803 pv_remove(pmap_t pmap, vm_offset_t va, vm_page_t m)
807 //int su = (pmap == kernel_pmap);
808 //debugf("pv_remove: s (su = %d pmap = 0x%08x va = 0x%08x)\n", su, (u_int32_t)pmap, va);
810 PMAP_LOCK_ASSERT(pmap, MA_OWNED);
811 rw_assert(&pvh_global_lock, RA_WLOCKED);
814 TAILQ_FOREACH(pve, &m->md.pv_list, pv_link) {
815 if ((pmap == pve->pv_pmap) && (va == pve->pv_va)) {
816 /* remove from pv_list */
817 TAILQ_REMOVE(&m->md.pv_list, pve, pv_link);
818 if (TAILQ_EMPTY(&m->md.pv_list))
819 vm_page_aflag_clear(m, PGA_WRITEABLE);
821 /* free pv entry struct */
827 //debugf("pv_remove: e\n");
831 * Clean pte entry, try to free page table page if requested.
833 * Return 1 if ptbl pages were freed, otherwise return 0.
836 pte_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, uint8_t flags)
838 unsigned int pdir_idx = PDIR_IDX(va);
839 unsigned int ptbl_idx = PTBL_IDX(va);
844 //int su = (pmap == kernel_pmap);
845 //debugf("pte_remove: s (su = %d pmap = 0x%08x va = 0x%08x flags = %d)\n",
846 // su, (u_int32_t)pmap, va, flags);
848 ptbl = pmap->pm_pdir[pdir_idx];
849 KASSERT(ptbl, ("pte_remove: null ptbl"));
851 pte = &ptbl[ptbl_idx];
853 if (pte == NULL || !PTE_ISVALID(pte))
856 if (PTE_ISWIRED(pte))
857 pmap->pm_stats.wired_count--;
859 /* Handle managed entry. */
860 if (PTE_ISMANAGED(pte)) {
861 /* Get vm_page_t for mapped pte. */
862 m = PHYS_TO_VM_PAGE(PTE_PA(pte));
864 if (PTE_ISMODIFIED(pte))
867 if (PTE_ISREFERENCED(pte))
868 vm_page_aflag_set(m, PGA_REFERENCED);
870 pv_remove(pmap, va, m);
873 mtx_lock_spin(&tlbivax_mutex);
876 tlb0_flush_entry(va);
881 mtx_unlock_spin(&tlbivax_mutex);
883 pmap->pm_stats.resident_count--;
885 if (flags & PTBL_UNHOLD) {
886 //debugf("pte_remove: e (unhold)\n");
887 return (ptbl_unhold(mmu, pmap, pdir_idx));
890 //debugf("pte_remove: e\n");
895 * Insert PTE for a given page and virtual address.
898 pte_enter(mmu_t mmu, pmap_t pmap, vm_page_t m, vm_offset_t va, uint32_t flags,
901 unsigned int pdir_idx = PDIR_IDX(va);
902 unsigned int ptbl_idx = PTBL_IDX(va);
905 CTR4(KTR_PMAP, "%s: su = %d pmap = %p va = %p", __func__,
906 pmap == kernel_pmap, pmap, va);
908 /* Get the page table pointer. */
909 ptbl = pmap->pm_pdir[pdir_idx];
912 /* Allocate page table pages. */
913 ptbl = ptbl_alloc(mmu, pmap, pdir_idx, nosleep);
915 KASSERT(nosleep, ("nosleep and NULL ptbl"));
920 * Check if there is valid mapping for requested
921 * va, if there is, remove it.
923 pte = &pmap->pm_pdir[pdir_idx][ptbl_idx];
924 if (PTE_ISVALID(pte)) {
925 pte_remove(mmu, pmap, va, PTBL_HOLD);
928 * pte is not used, increment hold count
931 if (pmap != kernel_pmap)
932 ptbl_hold(mmu, pmap, pdir_idx);
937 * Insert pv_entry into pv_list for mapped page if part of managed
940 if ((m->oflags & VPO_UNMANAGED) == 0) {
941 flags |= PTE_MANAGED;
943 /* Create and insert pv entry. */
944 pv_insert(pmap, va, m);
947 pmap->pm_stats.resident_count++;
949 mtx_lock_spin(&tlbivax_mutex);
952 tlb0_flush_entry(va);
953 if (pmap->pm_pdir[pdir_idx] == NULL) {
955 * If we just allocated a new page table, hook it in
958 pmap->pm_pdir[pdir_idx] = ptbl;
960 pte = &(pmap->pm_pdir[pdir_idx][ptbl_idx]);
961 pte->rpn = VM_PAGE_TO_PHYS(m) & ~PTE_PA_MASK;
962 pte->flags |= (PTE_VALID | flags);
965 mtx_unlock_spin(&tlbivax_mutex);
969 /* Return the pa for the given pmap/va. */
971 pte_vatopa(mmu_t mmu, pmap_t pmap, vm_offset_t va)
976 pte = pte_find(mmu, pmap, va);
977 if ((pte != NULL) && PTE_ISVALID(pte))
978 pa = (PTE_PA(pte) | (va & PTE_PA_MASK));
982 /* Get a pointer to a PTE in a page table. */
984 pte_find(mmu_t mmu, pmap_t pmap, vm_offset_t va)
986 unsigned int pdir_idx = PDIR_IDX(va);
987 unsigned int ptbl_idx = PTBL_IDX(va);
989 KASSERT((pmap != NULL), ("pte_find: invalid pmap"));
991 if (pmap->pm_pdir[pdir_idx])
992 return (&(pmap->pm_pdir[pdir_idx][ptbl_idx]));
997 /**************************************************************************/
999 /**************************************************************************/
1002 * This is called during booke_init, before the system is really initialized.
1005 mmu_booke_bootstrap(mmu_t mmu, vm_offset_t start, vm_offset_t kernelend)
1007 vm_offset_t phys_kernelend;
1008 struct mem_region *mp, *mp1;
1011 u_int phys_avail_count;
1012 vm_size_t physsz, hwphyssz, kstack0_sz;
1013 vm_offset_t kernel_pdir, kstack0, va;
1014 vm_paddr_t kstack0_phys;
1018 debugf("mmu_booke_bootstrap: entered\n");
1020 /* Set interesting system properties */
1024 /* Initialize invalidation mutex */
1025 mtx_init(&tlbivax_mutex, "tlbivax", NULL, MTX_SPIN);
1027 /* Read TLB0 size and associativity. */
1031 * Align kernel start and end address (kernel image).
1032 * Note that kernel end does not necessarily relate to kernsize.
1033 * kernsize is the size of the kernel that is actually mapped.
1034 * Also note that "start - 1" is deliberate. With SMP, the
1035 * entry point is exactly a page from the actual load address.
1036 * As such, trunc_page() has no effect and we're off by a page.
1037 * Since we always have the ELF header between the load address
1038 * and the entry point, we can safely subtract 1 to compensate.
1040 kernstart = trunc_page(start - 1);
1041 data_start = round_page(kernelend);
1042 data_end = data_start;
1045 * Addresses of preloaded modules (like file systems) use
1046 * physical addresses. Make sure we relocate those into
1047 * virtual addresses.
1049 preload_addr_relocate = kernstart - kernload;
1051 /* Allocate the dynamic per-cpu area. */
1052 dpcpu = (void *)data_end;
1053 data_end += DPCPU_SIZE;
1055 /* Allocate space for the message buffer. */
1056 msgbufp = (struct msgbuf *)data_end;
1057 data_end += msgbufsize;
1058 debugf(" msgbufp at 0x%08x end = 0x%08x\n", (uint32_t)msgbufp,
1061 data_end = round_page(data_end);
1063 /* Allocate space for ptbl_bufs. */
1064 ptbl_bufs = (struct ptbl_buf *)data_end;
1065 data_end += sizeof(struct ptbl_buf) * PTBL_BUFS;
1066 debugf(" ptbl_bufs at 0x%08x end = 0x%08x\n", (uint32_t)ptbl_bufs,
1069 data_end = round_page(data_end);
1071 /* Allocate PTE tables for kernel KVA. */
1072 kernel_pdir = data_end;
1073 kernel_ptbls = (VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS +
1074 PDIR_SIZE - 1) / PDIR_SIZE;
1075 data_end += kernel_ptbls * PTBL_PAGES * PAGE_SIZE;
1076 debugf(" kernel ptbls: %d\n", kernel_ptbls);
1077 debugf(" kernel pdir at 0x%08x end = 0x%08x\n", kernel_pdir, data_end);
1079 debugf(" data_end: 0x%08x\n", data_end);
1080 if (data_end - kernstart > kernsize) {
1081 kernsize += tlb1_mapin_region(kernstart + kernsize,
1082 kernload + kernsize, (data_end - kernstart) - kernsize);
1084 data_end = kernstart + kernsize;
1085 debugf(" updated data_end: 0x%08x\n", data_end);
1088 * Clear the structures - note we can only do it safely after the
1089 * possible additional TLB1 translations are in place (above) so that
1090 * all range up to the currently calculated 'data_end' is covered.
1092 dpcpu_init(dpcpu, 0);
1093 memset((void *)ptbl_bufs, 0, sizeof(struct ptbl_buf) * PTBL_SIZE);
1094 memset((void *)kernel_pdir, 0, kernel_ptbls * PTBL_PAGES * PAGE_SIZE);
1096 /*******************************************************/
1097 /* Set the start and end of kva. */
1098 /*******************************************************/
1099 virtual_avail = round_page(data_end);
1100 virtual_end = VM_MAX_KERNEL_ADDRESS;
1102 /* Allocate KVA space for page zero/copy operations. */
1103 zero_page_va = virtual_avail;
1104 virtual_avail += PAGE_SIZE;
1105 zero_page_idle_va = virtual_avail;
1106 virtual_avail += PAGE_SIZE;
1107 copy_page_src_va = virtual_avail;
1108 virtual_avail += PAGE_SIZE;
1109 copy_page_dst_va = virtual_avail;
1110 virtual_avail += PAGE_SIZE;
1111 debugf("zero_page_va = 0x%08x\n", zero_page_va);
1112 debugf("zero_page_idle_va = 0x%08x\n", zero_page_idle_va);
1113 debugf("copy_page_src_va = 0x%08x\n", copy_page_src_va);
1114 debugf("copy_page_dst_va = 0x%08x\n", copy_page_dst_va);
1116 /* Initialize page zero/copy mutexes. */
1117 mtx_init(&zero_page_mutex, "mmu_booke_zero_page", NULL, MTX_DEF);
1118 mtx_init(©_page_mutex, "mmu_booke_copy_page", NULL, MTX_DEF);
1120 /* Allocate KVA space for ptbl bufs. */
1121 ptbl_buf_pool_vabase = virtual_avail;
1122 virtual_avail += PTBL_BUFS * PTBL_PAGES * PAGE_SIZE;
1123 debugf("ptbl_buf_pool_vabase = 0x%08x end = 0x%08x\n",
1124 ptbl_buf_pool_vabase, virtual_avail);
1126 /* Calculate corresponding physical addresses for the kernel region. */
1127 phys_kernelend = kernload + kernsize;
1128 debugf("kernel image and allocated data:\n");
1129 debugf(" kernload = 0x%08x\n", kernload);
1130 debugf(" kernstart = 0x%08x\n", kernstart);
1131 debugf(" kernsize = 0x%08x\n", kernsize);
1133 if (sizeof(phys_avail) / sizeof(phys_avail[0]) < availmem_regions_sz)
1134 panic("mmu_booke_bootstrap: phys_avail too small");
1137 * Remove kernel physical address range from avail regions list. Page
1138 * align all regions. Non-page aligned memory isn't very interesting
1139 * to us. Also, sort the entries for ascending addresses.
1142 /* Retrieve phys/avail mem regions */
1143 mem_regions(&physmem_regions, &physmem_regions_sz,
1144 &availmem_regions, &availmem_regions_sz);
1146 cnt = availmem_regions_sz;
1147 debugf("processing avail regions:\n");
1148 for (mp = availmem_regions; mp->mr_size; mp++) {
1150 e = mp->mr_start + mp->mr_size;
1151 debugf(" %08x-%08x -> ", s, e);
1152 /* Check whether this region holds all of the kernel. */
1153 if (s < kernload && e > phys_kernelend) {
1154 availmem_regions[cnt].mr_start = phys_kernelend;
1155 availmem_regions[cnt++].mr_size = e - phys_kernelend;
1158 /* Look whether this regions starts within the kernel. */
1159 if (s >= kernload && s < phys_kernelend) {
1160 if (e <= phys_kernelend)
1164 /* Now look whether this region ends within the kernel. */
1165 if (e > kernload && e <= phys_kernelend) {
1170 /* Now page align the start and size of the region. */
1176 debugf("%08x-%08x = %x\n", s, e, sz);
1178 /* Check whether some memory is left here. */
1182 (cnt - (mp - availmem_regions)) * sizeof(*mp));
1188 /* Do an insertion sort. */
1189 for (mp1 = availmem_regions; mp1 < mp; mp1++)
1190 if (s < mp1->mr_start)
1193 memmove(mp1 + 1, mp1, (char *)mp - (char *)mp1);
1201 availmem_regions_sz = cnt;
1203 /*******************************************************/
1204 /* Steal physical memory for kernel stack from the end */
1205 /* of the first avail region */
1206 /*******************************************************/
1207 kstack0_sz = KSTACK_PAGES * PAGE_SIZE;
1208 kstack0_phys = availmem_regions[0].mr_start +
1209 availmem_regions[0].mr_size;
1210 kstack0_phys -= kstack0_sz;
1211 availmem_regions[0].mr_size -= kstack0_sz;
1213 /*******************************************************/
1214 /* Fill in phys_avail table, based on availmem_regions */
1215 /*******************************************************/
1216 phys_avail_count = 0;
1219 TUNABLE_ULONG_FETCH("hw.physmem", (u_long *) &hwphyssz);
1221 debugf("fill in phys_avail:\n");
1222 for (i = 0, j = 0; i < availmem_regions_sz; i++, j += 2) {
1224 debugf(" region: 0x%08x - 0x%08x (0x%08x)\n",
1225 availmem_regions[i].mr_start,
1226 availmem_regions[i].mr_start +
1227 availmem_regions[i].mr_size,
1228 availmem_regions[i].mr_size);
1230 if (hwphyssz != 0 &&
1231 (physsz + availmem_regions[i].mr_size) >= hwphyssz) {
1232 debugf(" hw.physmem adjust\n");
1233 if (physsz < hwphyssz) {
1234 phys_avail[j] = availmem_regions[i].mr_start;
1236 availmem_regions[i].mr_start +
1244 phys_avail[j] = availmem_regions[i].mr_start;
1245 phys_avail[j + 1] = availmem_regions[i].mr_start +
1246 availmem_regions[i].mr_size;
1248 physsz += availmem_regions[i].mr_size;
1250 physmem = btoc(physsz);
1252 /* Calculate the last available physical address. */
1253 for (i = 0; phys_avail[i + 2] != 0; i += 2)
1255 Maxmem = powerpc_btop(phys_avail[i + 1]);
1257 debugf("Maxmem = 0x%08lx\n", Maxmem);
1258 debugf("phys_avail_count = %d\n", phys_avail_count);
1259 debugf("physsz = 0x%08x physmem = %ld (0x%08lx)\n", physsz, physmem,
1262 /*******************************************************/
1263 /* Initialize (statically allocated) kernel pmap. */
1264 /*******************************************************/
1265 PMAP_LOCK_INIT(kernel_pmap);
1266 kptbl_min = VM_MIN_KERNEL_ADDRESS / PDIR_SIZE;
1268 debugf("kernel_pmap = 0x%08x\n", (uint32_t)kernel_pmap);
1269 debugf("kptbl_min = %d, kernel_ptbls = %d\n", kptbl_min, kernel_ptbls);
1270 debugf("kernel pdir range: 0x%08x - 0x%08x\n",
1271 kptbl_min * PDIR_SIZE, (kptbl_min + kernel_ptbls) * PDIR_SIZE - 1);
1273 /* Initialize kernel pdir */
1274 for (i = 0; i < kernel_ptbls; i++)
1275 kernel_pmap->pm_pdir[kptbl_min + i] =
1276 (pte_t *)(kernel_pdir + (i * PAGE_SIZE * PTBL_PAGES));
1278 for (i = 0; i < MAXCPU; i++) {
1279 kernel_pmap->pm_tid[i] = TID_KERNEL;
1281 /* Initialize each CPU's tidbusy entry 0 with kernel_pmap */
1282 tidbusy[i][0] = kernel_pmap;
1286 * Fill in PTEs covering kernel code and data. They are not required
1287 * for address translation, as this area is covered by static TLB1
1288 * entries, but for pte_vatopa() to work correctly with kernel area
1291 for (va = kernstart; va < data_end; va += PAGE_SIZE) {
1292 pte = &(kernel_pmap->pm_pdir[PDIR_IDX(va)][PTBL_IDX(va)]);
1293 pte->rpn = kernload + (va - kernstart);
1294 pte->flags = PTE_M | PTE_SR | PTE_SW | PTE_SX | PTE_WIRED |
1297 /* Mark kernel_pmap active on all CPUs */
1298 CPU_FILL(&kernel_pmap->pm_active);
1301 * Initialize the global pv list lock.
1303 rw_init(&pvh_global_lock, "pmap pv global");
1305 /*******************************************************/
1307 /*******************************************************/
1309 /* Enter kstack0 into kernel map, provide guard page */
1310 kstack0 = virtual_avail + KSTACK_GUARD_PAGES * PAGE_SIZE;
1311 thread0.td_kstack = kstack0;
1312 thread0.td_kstack_pages = KSTACK_PAGES;
1314 debugf("kstack_sz = 0x%08x\n", kstack0_sz);
1315 debugf("kstack0_phys at 0x%08x - 0x%08x\n",
1316 kstack0_phys, kstack0_phys + kstack0_sz);
1317 debugf("kstack0 at 0x%08x - 0x%08x\n", kstack0, kstack0 + kstack0_sz);
1319 virtual_avail += KSTACK_GUARD_PAGES * PAGE_SIZE + kstack0_sz;
1320 for (i = 0; i < KSTACK_PAGES; i++) {
1321 mmu_booke_kenter(mmu, kstack0, kstack0_phys);
1322 kstack0 += PAGE_SIZE;
1323 kstack0_phys += PAGE_SIZE;
1326 pmap_bootstrapped = 1;
1328 debugf("virtual_avail = %08x\n", virtual_avail);
1329 debugf("virtual_end = %08x\n", virtual_end);
1331 debugf("mmu_booke_bootstrap: exit\n");
1336 pmap_bootstrap_ap(volatile uint32_t *trcp __unused)
1341 * Finish TLB1 configuration: the BSP already set up its TLB1 and we
1342 * have the snapshot of its contents in the s/w tlb1[] table, so use
1343 * these values directly to (re)program AP's TLB1 hardware.
1345 for (i = bp_ntlb1s; i < tlb1_idx; i++) {
1346 /* Skip invalid entries */
1347 if (!(tlb1[i].mas1 & MAS1_VALID))
1350 tlb1_write_entry(i);
1353 set_mas4_defaults();
1358 * Get the physical page address for the given pmap/virtual address.
1361 mmu_booke_extract(mmu_t mmu, pmap_t pmap, vm_offset_t va)
1366 pa = pte_vatopa(mmu, pmap, va);
1373 * Extract the physical page address associated with the given
1374 * kernel virtual address.
1377 mmu_booke_kextract(mmu_t mmu, vm_offset_t va)
1381 /* Check TLB1 mappings */
1382 for (i = 0; i < tlb1_idx; i++) {
1383 if (!(tlb1[i].mas1 & MAS1_VALID))
1385 if (va >= tlb1[i].virt && va < tlb1[i].virt + tlb1[i].size)
1386 return (tlb1[i].phys + (va - tlb1[i].virt));
1389 return (pte_vatopa(mmu, kernel_pmap, va));
1393 * Initialize the pmap module.
1394 * Called by vm_init, to initialize any structures that the pmap
1395 * system needs to map virtual memory.
1398 mmu_booke_init(mmu_t mmu)
1400 int shpgperproc = PMAP_SHPGPERPROC;
1403 * Initialize the address space (zone) for the pv entries. Set a
1404 * high water mark so that the system can recover from excessive
1405 * numbers of pv entries.
1407 pvzone = uma_zcreate("PV ENTRY", sizeof(struct pv_entry), NULL, NULL,
1408 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_VM | UMA_ZONE_NOFREE);
1410 TUNABLE_INT_FETCH("vm.pmap.shpgperproc", &shpgperproc);
1411 pv_entry_max = shpgperproc * maxproc + vm_cnt.v_page_count;
1413 TUNABLE_INT_FETCH("vm.pmap.pv_entries", &pv_entry_max);
1414 pv_entry_high_water = 9 * (pv_entry_max / 10);
1416 uma_zone_reserve_kva(pvzone, pv_entry_max);
1418 /* Pre-fill pvzone with initial number of pv entries. */
1419 uma_prealloc(pvzone, PV_ENTRY_ZONE_MIN);
1421 /* Initialize ptbl allocation. */
1426 * Map a list of wired pages into kernel virtual address space. This is
1427 * intended for temporary mappings which do not need page modification or
1428 * references recorded. Existing mappings in the region are overwritten.
1431 mmu_booke_qenter(mmu_t mmu, vm_offset_t sva, vm_page_t *m, int count)
1436 while (count-- > 0) {
1437 mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(*m));
1444 * Remove page mappings from kernel virtual address space. Intended for
1445 * temporary mappings entered by mmu_booke_qenter.
1448 mmu_booke_qremove(mmu_t mmu, vm_offset_t sva, int count)
1453 while (count-- > 0) {
1454 mmu_booke_kremove(mmu, va);
1460 * Map a wired page into kernel virtual address space.
1463 mmu_booke_kenter(mmu_t mmu, vm_offset_t va, vm_paddr_t pa)
1466 mmu_booke_kenter_attr(mmu, va, pa, VM_MEMATTR_DEFAULT);
1470 mmu_booke_kenter_attr(mmu_t mmu, vm_offset_t va, vm_paddr_t pa, vm_memattr_t ma)
1472 unsigned int pdir_idx = PDIR_IDX(va);
1473 unsigned int ptbl_idx = PTBL_IDX(va);
1477 KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1478 (va <= VM_MAX_KERNEL_ADDRESS)), ("mmu_booke_kenter: invalid va"));
1480 flags = PTE_SR | PTE_SW | PTE_SX | PTE_WIRED | PTE_VALID;
1481 flags |= tlb_calc_wimg(pa, ma);
1483 pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1485 mtx_lock_spin(&tlbivax_mutex);
1488 if (PTE_ISVALID(pte)) {
1490 CTR1(KTR_PMAP, "%s: replacing entry!", __func__);
1492 /* Flush entry from TLB0 */
1493 tlb0_flush_entry(va);
1496 pte->rpn = pa & ~PTE_PA_MASK;
1499 //debugf("mmu_booke_kenter: pdir_idx = %d ptbl_idx = %d va=0x%08x "
1500 // "pa=0x%08x rpn=0x%08x flags=0x%08x\n",
1501 // pdir_idx, ptbl_idx, va, pa, pte->rpn, pte->flags);
1503 /* Flush the real memory from the instruction cache. */
1504 if ((flags & (PTE_I | PTE_G)) == 0) {
1505 __syncicache((void *)va, PAGE_SIZE);
1509 mtx_unlock_spin(&tlbivax_mutex);
1513 * Remove a page from kernel page table.
1516 mmu_booke_kremove(mmu_t mmu, vm_offset_t va)
1518 unsigned int pdir_idx = PDIR_IDX(va);
1519 unsigned int ptbl_idx = PTBL_IDX(va);
1522 // CTR2(KTR_PMAP,("%s: s (va = 0x%08x)\n", __func__, va));
1524 KASSERT(((va >= VM_MIN_KERNEL_ADDRESS) &&
1525 (va <= VM_MAX_KERNEL_ADDRESS)),
1526 ("mmu_booke_kremove: invalid va"));
1528 pte = &(kernel_pmap->pm_pdir[pdir_idx][ptbl_idx]);
1530 if (!PTE_ISVALID(pte)) {
1532 CTR1(KTR_PMAP, "%s: invalid pte", __func__);
1537 mtx_lock_spin(&tlbivax_mutex);
1540 /* Invalidate entry in TLB0, update PTE. */
1541 tlb0_flush_entry(va);
1546 mtx_unlock_spin(&tlbivax_mutex);
1550 * Initialize pmap associated with process 0.
1553 mmu_booke_pinit0(mmu_t mmu, pmap_t pmap)
1556 PMAP_LOCK_INIT(pmap);
1557 mmu_booke_pinit(mmu, pmap);
1558 PCPU_SET(curpmap, pmap);
1562 * Initialize a preallocated and zeroed pmap structure,
1563 * such as one in a vmspace structure.
1566 mmu_booke_pinit(mmu_t mmu, pmap_t pmap)
1570 CTR4(KTR_PMAP, "%s: pmap = %p, proc %d '%s'", __func__, pmap,
1571 curthread->td_proc->p_pid, curthread->td_proc->p_comm);
1573 KASSERT((pmap != kernel_pmap), ("pmap_pinit: initializing kernel_pmap"));
1575 for (i = 0; i < MAXCPU; i++)
1576 pmap->pm_tid[i] = TID_NONE;
1577 CPU_ZERO(&kernel_pmap->pm_active);
1578 bzero(&pmap->pm_stats, sizeof(pmap->pm_stats));
1579 bzero(&pmap->pm_pdir, sizeof(pte_t *) * PDIR_NENTRIES);
1580 TAILQ_INIT(&pmap->pm_ptbl_list);
1584 * Release any resources held by the given physical map.
1585 * Called when a pmap initialized by mmu_booke_pinit is being released.
1586 * Should only be called if the map contains no valid mappings.
1589 mmu_booke_release(mmu_t mmu, pmap_t pmap)
1592 KASSERT(pmap->pm_stats.resident_count == 0,
1593 ("pmap_release: pmap resident count %ld != 0",
1594 pmap->pm_stats.resident_count));
1598 * Insert the given physical page at the specified virtual address in the
1599 * target physical map with the protection requested. If specified the page
1600 * will be wired down.
1603 mmu_booke_enter(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1604 vm_prot_t prot, u_int flags, int8_t psind)
1608 rw_wlock(&pvh_global_lock);
1610 error = mmu_booke_enter_locked(mmu, pmap, va, m, prot, flags, psind);
1611 rw_wunlock(&pvh_global_lock);
1617 mmu_booke_enter_locked(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1618 vm_prot_t prot, u_int pmap_flags, int8_t psind __unused)
1623 int error, su, sync;
1625 pa = VM_PAGE_TO_PHYS(m);
1626 su = (pmap == kernel_pmap);
1629 //debugf("mmu_booke_enter_locked: s (pmap=0x%08x su=%d tid=%d m=0x%08x va=0x%08x "
1630 // "pa=0x%08x prot=0x%08x flags=%#x)\n",
1631 // (u_int32_t)pmap, su, pmap->pm_tid,
1632 // (u_int32_t)m, va, pa, prot, flags);
1635 KASSERT(((va >= virtual_avail) &&
1636 (va <= VM_MAX_KERNEL_ADDRESS)),
1637 ("mmu_booke_enter_locked: kernel pmap, non kernel va"));
1639 KASSERT((va <= VM_MAXUSER_ADDRESS),
1640 ("mmu_booke_enter_locked: user pmap, non user va"));
1642 if ((m->oflags & VPO_UNMANAGED) == 0 && !vm_page_xbusied(m))
1643 VM_OBJECT_ASSERT_LOCKED(m->object);
1645 PMAP_LOCK_ASSERT(pmap, MA_OWNED);
1648 * If there is an existing mapping, and the physical address has not
1649 * changed, must be protection or wiring change.
1651 if (((pte = pte_find(mmu, pmap, va)) != NULL) &&
1652 (PTE_ISVALID(pte)) && (PTE_PA(pte) == pa)) {
1655 * Before actually updating pte->flags we calculate and
1656 * prepare its new value in a helper var.
1659 flags &= ~(PTE_UW | PTE_UX | PTE_SW | PTE_SX | PTE_MODIFIED);
1661 /* Wiring change, just update stats. */
1662 if ((pmap_flags & PMAP_ENTER_WIRED) != 0) {
1663 if (!PTE_ISWIRED(pte)) {
1665 pmap->pm_stats.wired_count++;
1668 if (PTE_ISWIRED(pte)) {
1669 flags &= ~PTE_WIRED;
1670 pmap->pm_stats.wired_count--;
1674 if (prot & VM_PROT_WRITE) {
1675 /* Add write permissions. */
1680 if ((flags & PTE_MANAGED) != 0)
1681 vm_page_aflag_set(m, PGA_WRITEABLE);
1683 /* Handle modified pages, sense modify status. */
1686 * The PTE_MODIFIED flag could be set by underlying
1687 * TLB misses since we last read it (above), possibly
1688 * other CPUs could update it so we check in the PTE
1689 * directly rather than rely on that saved local flags
1692 if (PTE_ISMODIFIED(pte))
1696 if (prot & VM_PROT_EXECUTE) {
1702 * Check existing flags for execute permissions: if we
1703 * are turning execute permissions on, icache should
1706 if ((pte->flags & (PTE_UX | PTE_SX)) == 0)
1710 flags &= ~PTE_REFERENCED;
1713 * The new flags value is all calculated -- only now actually
1716 mtx_lock_spin(&tlbivax_mutex);
1719 tlb0_flush_entry(va);
1723 mtx_unlock_spin(&tlbivax_mutex);
1727 * If there is an existing mapping, but it's for a different
1728 * physical address, pte_enter() will delete the old mapping.
1730 //if ((pte != NULL) && PTE_ISVALID(pte))
1731 // debugf("mmu_booke_enter_locked: replace\n");
1733 // debugf("mmu_booke_enter_locked: new\n");
1735 /* Now set up the flags and install the new mapping. */
1736 flags = (PTE_SR | PTE_VALID);
1742 if (prot & VM_PROT_WRITE) {
1747 if ((m->oflags & VPO_UNMANAGED) == 0)
1748 vm_page_aflag_set(m, PGA_WRITEABLE);
1751 if (prot & VM_PROT_EXECUTE) {
1757 /* If its wired update stats. */
1758 if ((pmap_flags & PMAP_ENTER_WIRED) != 0)
1761 error = pte_enter(mmu, pmap, m, va, flags,
1762 (pmap_flags & PMAP_ENTER_NOSLEEP) != 0);
1764 return (KERN_RESOURCE_SHORTAGE);
1766 if ((flags & PMAP_ENTER_WIRED) != 0)
1767 pmap->pm_stats.wired_count++;
1769 /* Flush the real memory from the instruction cache. */
1770 if (prot & VM_PROT_EXECUTE)
1774 if (sync && (su || pmap == PCPU_GET(curpmap))) {
1775 __syncicache((void *)va, PAGE_SIZE);
1779 return (KERN_SUCCESS);
1783 * Maps a sequence of resident pages belonging to the same object.
1784 * The sequence begins with the given page m_start. This page is
1785 * mapped at the given virtual address start. Each subsequent page is
1786 * mapped at a virtual address that is offset from start by the same
1787 * amount as the page is offset from m_start within the object. The
1788 * last page in the sequence is the page with the largest offset from
1789 * m_start that can be mapped at a virtual address less than the given
1790 * virtual address end. Not every virtual page between start and end
1791 * is mapped; only those for which a resident page exists with the
1792 * corresponding offset from m_start are mapped.
1795 mmu_booke_enter_object(mmu_t mmu, pmap_t pmap, vm_offset_t start,
1796 vm_offset_t end, vm_page_t m_start, vm_prot_t prot)
1799 vm_pindex_t diff, psize;
1801 VM_OBJECT_ASSERT_LOCKED(m_start->object);
1803 psize = atop(end - start);
1805 rw_wlock(&pvh_global_lock);
1807 while (m != NULL && (diff = m->pindex - m_start->pindex) < psize) {
1808 mmu_booke_enter_locked(mmu, pmap, start + ptoa(diff), m,
1809 prot & (VM_PROT_READ | VM_PROT_EXECUTE),
1810 PMAP_ENTER_NOSLEEP, 0);
1811 m = TAILQ_NEXT(m, listq);
1813 rw_wunlock(&pvh_global_lock);
1818 mmu_booke_enter_quick(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_page_t m,
1822 rw_wlock(&pvh_global_lock);
1824 mmu_booke_enter_locked(mmu, pmap, va, m,
1825 prot & (VM_PROT_READ | VM_PROT_EXECUTE), PMAP_ENTER_NOSLEEP,
1827 rw_wunlock(&pvh_global_lock);
1832 * Remove the given range of addresses from the specified map.
1834 * It is assumed that the start and end are properly rounded to the page size.
1837 mmu_booke_remove(mmu_t mmu, pmap_t pmap, vm_offset_t va, vm_offset_t endva)
1842 int su = (pmap == kernel_pmap);
1844 //debugf("mmu_booke_remove: s (su = %d pmap=0x%08x tid=%d va=0x%08x endva=0x%08x)\n",
1845 // su, (u_int32_t)pmap, pmap->pm_tid, va, endva);
1848 KASSERT(((va >= virtual_avail) &&
1849 (va <= VM_MAX_KERNEL_ADDRESS)),
1850 ("mmu_booke_remove: kernel pmap, non kernel va"));
1852 KASSERT((va <= VM_MAXUSER_ADDRESS),
1853 ("mmu_booke_remove: user pmap, non user va"));
1856 if (PMAP_REMOVE_DONE(pmap)) {
1857 //debugf("mmu_booke_remove: e (empty)\n");
1861 hold_flag = PTBL_HOLD_FLAG(pmap);
1862 //debugf("mmu_booke_remove: hold_flag = %d\n", hold_flag);
1864 rw_wlock(&pvh_global_lock);
1866 for (; va < endva; va += PAGE_SIZE) {
1867 pte = pte_find(mmu, pmap, va);
1868 if ((pte != NULL) && PTE_ISVALID(pte))
1869 pte_remove(mmu, pmap, va, hold_flag);
1872 rw_wunlock(&pvh_global_lock);
1874 //debugf("mmu_booke_remove: e\n");
1878 * Remove physical page from all pmaps in which it resides.
1881 mmu_booke_remove_all(mmu_t mmu, vm_page_t m)
1886 rw_wlock(&pvh_global_lock);
1887 for (pv = TAILQ_FIRST(&m->md.pv_list); pv != NULL; pv = pvn) {
1888 pvn = TAILQ_NEXT(pv, pv_link);
1890 PMAP_LOCK(pv->pv_pmap);
1891 hold_flag = PTBL_HOLD_FLAG(pv->pv_pmap);
1892 pte_remove(mmu, pv->pv_pmap, pv->pv_va, hold_flag);
1893 PMAP_UNLOCK(pv->pv_pmap);
1895 vm_page_aflag_clear(m, PGA_WRITEABLE);
1896 rw_wunlock(&pvh_global_lock);
1900 * Map a range of physical addresses into kernel virtual address space.
1903 mmu_booke_map(mmu_t mmu, vm_offset_t *virt, vm_paddr_t pa_start,
1904 vm_paddr_t pa_end, int prot)
1906 vm_offset_t sva = *virt;
1907 vm_offset_t va = sva;
1909 //debugf("mmu_booke_map: s (sva = 0x%08x pa_start = 0x%08x pa_end = 0x%08x)\n",
1910 // sva, pa_start, pa_end);
1912 while (pa_start < pa_end) {
1913 mmu_booke_kenter(mmu, va, pa_start);
1915 pa_start += PAGE_SIZE;
1919 //debugf("mmu_booke_map: e (va = 0x%08x)\n", va);
1924 * The pmap must be activated before it's address space can be accessed in any
1928 mmu_booke_activate(mmu_t mmu, struct thread *td)
1933 pmap = &td->td_proc->p_vmspace->vm_pmap;
1935 CTR5(KTR_PMAP, "%s: s (td = %p, proc = '%s', id = %d, pmap = 0x%08x)",
1936 __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1938 KASSERT((pmap != kernel_pmap), ("mmu_booke_activate: kernel_pmap!"));
1942 cpuid = PCPU_GET(cpuid);
1943 CPU_SET_ATOMIC(cpuid, &pmap->pm_active);
1944 PCPU_SET(curpmap, pmap);
1946 if (pmap->pm_tid[cpuid] == TID_NONE)
1949 /* Load PID0 register with pmap tid value. */
1950 mtspr(SPR_PID0, pmap->pm_tid[cpuid]);
1951 __asm __volatile("isync");
1953 mtspr(SPR_DBCR0, td->td_pcb->pcb_cpu.booke.dbcr0);
1957 CTR3(KTR_PMAP, "%s: e (tid = %d for '%s')", __func__,
1958 pmap->pm_tid[PCPU_GET(cpuid)], td->td_proc->p_comm);
1962 * Deactivate the specified process's address space.
1965 mmu_booke_deactivate(mmu_t mmu, struct thread *td)
1969 pmap = &td->td_proc->p_vmspace->vm_pmap;
1971 CTR5(KTR_PMAP, "%s: td=%p, proc = '%s', id = %d, pmap = 0x%08x",
1972 __func__, td, td->td_proc->p_comm, td->td_proc->p_pid, pmap);
1974 td->td_pcb->pcb_cpu.booke.dbcr0 = mfspr(SPR_DBCR0);
1976 CPU_CLR_ATOMIC(PCPU_GET(cpuid), &pmap->pm_active);
1977 PCPU_SET(curpmap, NULL);
1981 * Copy the range specified by src_addr/len
1982 * from the source map to the range dst_addr/len
1983 * in the destination map.
1985 * This routine is only advisory and need not do anything.
1988 mmu_booke_copy(mmu_t mmu, pmap_t dst_pmap, pmap_t src_pmap,
1989 vm_offset_t dst_addr, vm_size_t len, vm_offset_t src_addr)
1995 * Set the physical protection on the specified range of this map as requested.
1998 mmu_booke_protect(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
2005 if ((prot & VM_PROT_READ) == VM_PROT_NONE) {
2006 mmu_booke_remove(mmu, pmap, sva, eva);
2010 if (prot & VM_PROT_WRITE)
2014 for (va = sva; va < eva; va += PAGE_SIZE) {
2015 if ((pte = pte_find(mmu, pmap, va)) != NULL) {
2016 if (PTE_ISVALID(pte)) {
2017 m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2019 mtx_lock_spin(&tlbivax_mutex);
2022 /* Handle modified pages. */
2023 if (PTE_ISMODIFIED(pte) && PTE_ISMANAGED(pte))
2026 tlb0_flush_entry(va);
2027 pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2030 mtx_unlock_spin(&tlbivax_mutex);
2038 * Clear the write and modified bits in each of the given page's mappings.
2041 mmu_booke_remove_write(mmu_t mmu, vm_page_t m)
2046 KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2047 ("mmu_booke_remove_write: page %p is not managed", m));
2050 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2051 * set by another thread while the object is locked. Thus,
2052 * if PGA_WRITEABLE is clear, no page table entries need updating.
2054 VM_OBJECT_ASSERT_WLOCKED(m->object);
2055 if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2057 rw_wlock(&pvh_global_lock);
2058 TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2059 PMAP_LOCK(pv->pv_pmap);
2060 if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL) {
2061 if (PTE_ISVALID(pte)) {
2062 m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2064 mtx_lock_spin(&tlbivax_mutex);
2067 /* Handle modified pages. */
2068 if (PTE_ISMODIFIED(pte))
2071 /* Flush mapping from TLB0. */
2072 pte->flags &= ~(PTE_UW | PTE_SW | PTE_MODIFIED);
2075 mtx_unlock_spin(&tlbivax_mutex);
2078 PMAP_UNLOCK(pv->pv_pmap);
2080 vm_page_aflag_clear(m, PGA_WRITEABLE);
2081 rw_wunlock(&pvh_global_lock);
2085 mmu_booke_sync_icache(mmu_t mmu, pmap_t pm, vm_offset_t va, vm_size_t sz)
2094 va = trunc_page(va);
2095 sz = round_page(sz);
2097 rw_wlock(&pvh_global_lock);
2098 pmap = PCPU_GET(curpmap);
2099 active = (pm == kernel_pmap || pm == pmap) ? 1 : 0;
2102 pte = pte_find(mmu, pm, va);
2103 valid = (pte != NULL && PTE_ISVALID(pte)) ? 1 : 0;
2109 /* Create a mapping in the active pmap. */
2111 m = PHYS_TO_VM_PAGE(pa);
2113 pte_enter(mmu, pmap, m, addr,
2114 PTE_SR | PTE_VALID | PTE_UR, FALSE);
2115 __syncicache((void *)addr, PAGE_SIZE);
2116 pte_remove(mmu, pmap, addr, PTBL_UNHOLD);
2119 __syncicache((void *)va, PAGE_SIZE);
2124 rw_wunlock(&pvh_global_lock);
2128 * Atomically extract and hold the physical page with the given
2129 * pmap and virtual address pair if that mapping permits the given
2133 mmu_booke_extract_and_hold(mmu_t mmu, pmap_t pmap, vm_offset_t va,
2145 pte = pte_find(mmu, pmap, va);
2146 if ((pte != NULL) && PTE_ISVALID(pte)) {
2147 if (pmap == kernel_pmap)
2152 if ((pte->flags & pte_wbit) || ((prot & VM_PROT_WRITE) == 0)) {
2153 if (vm_page_pa_tryrelock(pmap, PTE_PA(pte), &pa))
2155 m = PHYS_TO_VM_PAGE(PTE_PA(pte));
2166 * Initialize a vm_page's machine-dependent fields.
2169 mmu_booke_page_init(mmu_t mmu, vm_page_t m)
2172 TAILQ_INIT(&m->md.pv_list);
2176 * mmu_booke_zero_page_area zeros the specified hardware page by
2177 * mapping it into virtual memory and using bzero to clear
2180 * off and size must reside within a single page.
2183 mmu_booke_zero_page_area(mmu_t mmu, vm_page_t m, int off, int size)
2187 /* XXX KASSERT off and size are within a single page? */
2189 mtx_lock(&zero_page_mutex);
2192 mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2193 bzero((caddr_t)va + off, size);
2194 mmu_booke_kremove(mmu, va);
2196 mtx_unlock(&zero_page_mutex);
2200 * mmu_booke_zero_page zeros the specified hardware page.
2203 mmu_booke_zero_page(mmu_t mmu, vm_page_t m)
2206 mmu_booke_zero_page_area(mmu, m, 0, PAGE_SIZE);
2210 * mmu_booke_copy_page copies the specified (machine independent) page by
2211 * mapping the page into virtual memory and using memcopy to copy the page,
2212 * one machine dependent page at a time.
2215 mmu_booke_copy_page(mmu_t mmu, vm_page_t sm, vm_page_t dm)
2217 vm_offset_t sva, dva;
2219 sva = copy_page_src_va;
2220 dva = copy_page_dst_va;
2222 mtx_lock(©_page_mutex);
2223 mmu_booke_kenter(mmu, sva, VM_PAGE_TO_PHYS(sm));
2224 mmu_booke_kenter(mmu, dva, VM_PAGE_TO_PHYS(dm));
2225 memcpy((caddr_t)dva, (caddr_t)sva, PAGE_SIZE);
2226 mmu_booke_kremove(mmu, dva);
2227 mmu_booke_kremove(mmu, sva);
2228 mtx_unlock(©_page_mutex);
2232 mmu_booke_copy_pages(mmu_t mmu, vm_page_t *ma, vm_offset_t a_offset,
2233 vm_page_t *mb, vm_offset_t b_offset, int xfersize)
2236 vm_offset_t a_pg_offset, b_pg_offset;
2239 mtx_lock(©_page_mutex);
2240 while (xfersize > 0) {
2241 a_pg_offset = a_offset & PAGE_MASK;
2242 cnt = min(xfersize, PAGE_SIZE - a_pg_offset);
2243 mmu_booke_kenter(mmu, copy_page_src_va,
2244 VM_PAGE_TO_PHYS(ma[a_offset >> PAGE_SHIFT]));
2245 a_cp = (char *)copy_page_src_va + a_pg_offset;
2246 b_pg_offset = b_offset & PAGE_MASK;
2247 cnt = min(cnt, PAGE_SIZE - b_pg_offset);
2248 mmu_booke_kenter(mmu, copy_page_dst_va,
2249 VM_PAGE_TO_PHYS(mb[b_offset >> PAGE_SHIFT]));
2250 b_cp = (char *)copy_page_dst_va + b_pg_offset;
2251 bcopy(a_cp, b_cp, cnt);
2252 mmu_booke_kremove(mmu, copy_page_dst_va);
2253 mmu_booke_kremove(mmu, copy_page_src_va);
2258 mtx_unlock(©_page_mutex);
2262 * mmu_booke_zero_page_idle zeros the specified hardware page by mapping it
2263 * into virtual memory and using bzero to clear its contents. This is intended
2264 * to be called from the vm_pagezero process only and outside of Giant. No
2268 mmu_booke_zero_page_idle(mmu_t mmu, vm_page_t m)
2272 va = zero_page_idle_va;
2273 mmu_booke_kenter(mmu, va, VM_PAGE_TO_PHYS(m));
2274 bzero((caddr_t)va, PAGE_SIZE);
2275 mmu_booke_kremove(mmu, va);
2279 * Return whether or not the specified physical page was modified
2280 * in any of physical maps.
2283 mmu_booke_is_modified(mmu_t mmu, vm_page_t m)
2289 KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2290 ("mmu_booke_is_modified: page %p is not managed", m));
2294 * If the page is not exclusive busied, then PGA_WRITEABLE cannot be
2295 * concurrently set while the object is locked. Thus, if PGA_WRITEABLE
2296 * is clear, no PTEs can be modified.
2298 VM_OBJECT_ASSERT_WLOCKED(m->object);
2299 if (!vm_page_xbusied(m) && (m->aflags & PGA_WRITEABLE) == 0)
2301 rw_wlock(&pvh_global_lock);
2302 TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2303 PMAP_LOCK(pv->pv_pmap);
2304 if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2306 if (PTE_ISMODIFIED(pte))
2309 PMAP_UNLOCK(pv->pv_pmap);
2313 rw_wunlock(&pvh_global_lock);
2318 * Return whether or not the specified virtual address is eligible
2322 mmu_booke_is_prefaultable(mmu_t mmu, pmap_t pmap, vm_offset_t addr)
2329 * Return whether or not the specified physical page was referenced
2330 * in any physical maps.
2333 mmu_booke_is_referenced(mmu_t mmu, vm_page_t m)
2339 KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2340 ("mmu_booke_is_referenced: page %p is not managed", m));
2342 rw_wlock(&pvh_global_lock);
2343 TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2344 PMAP_LOCK(pv->pv_pmap);
2345 if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2347 if (PTE_ISREFERENCED(pte))
2350 PMAP_UNLOCK(pv->pv_pmap);
2354 rw_wunlock(&pvh_global_lock);
2359 * Clear the modify bits on the specified physical page.
2362 mmu_booke_clear_modify(mmu_t mmu, vm_page_t m)
2367 KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2368 ("mmu_booke_clear_modify: page %p is not managed", m));
2369 VM_OBJECT_ASSERT_WLOCKED(m->object);
2370 KASSERT(!vm_page_xbusied(m),
2371 ("mmu_booke_clear_modify: page %p is exclusive busied", m));
2374 * If the page is not PG_AWRITEABLE, then no PTEs can be modified.
2375 * If the object containing the page is locked and the page is not
2376 * exclusive busied, then PG_AWRITEABLE cannot be concurrently set.
2378 if ((m->aflags & PGA_WRITEABLE) == 0)
2380 rw_wlock(&pvh_global_lock);
2381 TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2382 PMAP_LOCK(pv->pv_pmap);
2383 if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2385 mtx_lock_spin(&tlbivax_mutex);
2388 if (pte->flags & (PTE_SW | PTE_UW | PTE_MODIFIED)) {
2389 tlb0_flush_entry(pv->pv_va);
2390 pte->flags &= ~(PTE_SW | PTE_UW | PTE_MODIFIED |
2395 mtx_unlock_spin(&tlbivax_mutex);
2397 PMAP_UNLOCK(pv->pv_pmap);
2399 rw_wunlock(&pvh_global_lock);
2403 * Return a count of reference bits for a page, clearing those bits.
2404 * It is not necessary for every reference bit to be cleared, but it
2405 * is necessary that 0 only be returned when there are truly no
2406 * reference bits set.
2408 * XXX: The exact number of bits to check and clear is a matter that
2409 * should be tested and standardized at some point in the future for
2410 * optimal aging of shared pages.
2413 mmu_booke_ts_referenced(mmu_t mmu, vm_page_t m)
2419 KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2420 ("mmu_booke_ts_referenced: page %p is not managed", m));
2422 rw_wlock(&pvh_global_lock);
2423 TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2424 PMAP_LOCK(pv->pv_pmap);
2425 if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL &&
2427 if (PTE_ISREFERENCED(pte)) {
2428 mtx_lock_spin(&tlbivax_mutex);
2431 tlb0_flush_entry(pv->pv_va);
2432 pte->flags &= ~PTE_REFERENCED;
2435 mtx_unlock_spin(&tlbivax_mutex);
2438 PMAP_UNLOCK(pv->pv_pmap);
2443 PMAP_UNLOCK(pv->pv_pmap);
2445 rw_wunlock(&pvh_global_lock);
2450 * Clear the wired attribute from the mappings for the specified range of
2451 * addresses in the given pmap. Every valid mapping within that range must
2452 * have the wired attribute set. In contrast, invalid mappings cannot have
2453 * the wired attribute set, so they are ignored.
2455 * The wired attribute of the page table entry is not a hardware feature, so
2456 * there is no need to invalidate any TLB entries.
2459 mmu_booke_unwire(mmu_t mmu, pmap_t pmap, vm_offset_t sva, vm_offset_t eva)
2465 for (va = sva; va < eva; va += PAGE_SIZE) {
2466 if ((pte = pte_find(mmu, pmap, va)) != NULL &&
2468 if (!PTE_ISWIRED(pte))
2469 panic("mmu_booke_unwire: pte %p isn't wired",
2471 pte->flags &= ~PTE_WIRED;
2472 pmap->pm_stats.wired_count--;
2480 * Return true if the pmap's pv is one of the first 16 pvs linked to from this
2481 * page. This count may be changed upwards or downwards in the future; it is
2482 * only necessary that true be returned for a small subset of pmaps for proper
2486 mmu_booke_page_exists_quick(mmu_t mmu, pmap_t pmap, vm_page_t m)
2492 KASSERT((m->oflags & VPO_UNMANAGED) == 0,
2493 ("mmu_booke_page_exists_quick: page %p is not managed", m));
2496 rw_wlock(&pvh_global_lock);
2497 TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2498 if (pv->pv_pmap == pmap) {
2505 rw_wunlock(&pvh_global_lock);
2510 * Return the number of managed mappings to the given physical page that are
2514 mmu_booke_page_wired_mappings(mmu_t mmu, vm_page_t m)
2520 if ((m->oflags & VPO_UNMANAGED) != 0)
2522 rw_wlock(&pvh_global_lock);
2523 TAILQ_FOREACH(pv, &m->md.pv_list, pv_link) {
2524 PMAP_LOCK(pv->pv_pmap);
2525 if ((pte = pte_find(mmu, pv->pv_pmap, pv->pv_va)) != NULL)
2526 if (PTE_ISVALID(pte) && PTE_ISWIRED(pte))
2528 PMAP_UNLOCK(pv->pv_pmap);
2530 rw_wunlock(&pvh_global_lock);
2535 mmu_booke_dev_direct_mapped(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2541 * This currently does not work for entries that
2542 * overlap TLB1 entries.
2544 for (i = 0; i < tlb1_idx; i ++) {
2545 if (tlb1_iomapped(i, pa, size, &va) == 0)
2553 mmu_booke_dumpsys_map(mmu_t mmu, vm_paddr_t pa, size_t sz, void **va)
2559 /* Minidumps are based on virtual memory addresses. */
2565 /* Raw physical memory dumps don't have a virtual address. */
2566 /* We always map a 256MB page at 256M. */
2567 gran = 256 * 1024 * 1024;
2568 ppa = pa & ~(gran - 1);
2571 tlb1_set_entry((vm_offset_t)va, ppa, gran, _TLB_ENTRY_IO);
2573 if (sz > (gran - ofs))
2574 tlb1_set_entry((vm_offset_t)(va + gran), ppa + gran, gran,
2579 mmu_booke_dumpsys_unmap(mmu_t mmu, vm_paddr_t pa, size_t sz, void *va)
2585 /* Minidumps are based on virtual memory addresses. */
2586 /* Nothing to do... */
2590 /* Raw physical memory dumps don't have a virtual address. */
2592 tlb1[tlb1_idx].mas1 = 0;
2593 tlb1[tlb1_idx].mas2 = 0;
2594 tlb1[tlb1_idx].mas3 = 0;
2595 tlb1_write_entry(tlb1_idx);
2597 gran = 256 * 1024 * 1024;
2598 ppa = pa & ~(gran - 1);
2600 if (sz > (gran - ofs)) {
2602 tlb1[tlb1_idx].mas1 = 0;
2603 tlb1[tlb1_idx].mas2 = 0;
2604 tlb1[tlb1_idx].mas3 = 0;
2605 tlb1_write_entry(tlb1_idx);
2609 extern struct dump_pa dump_map[PHYS_AVAIL_SZ + 1];
2612 mmu_booke_scan_init(mmu_t mmu)
2619 /* Initialize phys. segments for dumpsys(). */
2620 memset(&dump_map, 0, sizeof(dump_map));
2621 mem_regions(&physmem_regions, &physmem_regions_sz, &availmem_regions,
2622 &availmem_regions_sz);
2623 for (i = 0; i < physmem_regions_sz; i++) {
2624 dump_map[i].pa_start = physmem_regions[i].mr_start;
2625 dump_map[i].pa_size = physmem_regions[i].mr_size;
2630 /* Virtual segments for minidumps: */
2631 memset(&dump_map, 0, sizeof(dump_map));
2633 /* 1st: kernel .data and .bss. */
2634 dump_map[0].pa_start = trunc_page((uintptr_t)_etext);
2635 dump_map[0].pa_size =
2636 round_page((uintptr_t)_end) - dump_map[0].pa_start;
2638 /* 2nd: msgbuf and tables (see pmap_bootstrap()). */
2639 dump_map[1].pa_start = data_start;
2640 dump_map[1].pa_size = data_end - data_start;
2642 /* 3rd: kernel VM. */
2643 va = dump_map[1].pa_start + dump_map[1].pa_size;
2644 /* Find start of next chunk (from va). */
2645 while (va < virtual_end) {
2646 /* Don't dump the buffer cache. */
2647 if (va >= kmi.buffer_sva && va < kmi.buffer_eva) {
2648 va = kmi.buffer_eva;
2651 pte = pte_find(mmu, kernel_pmap, va);
2652 if (pte != NULL && PTE_ISVALID(pte))
2656 if (va < virtual_end) {
2657 dump_map[2].pa_start = va;
2659 /* Find last page in chunk. */
2660 while (va < virtual_end) {
2661 /* Don't run into the buffer cache. */
2662 if (va == kmi.buffer_sva)
2664 pte = pte_find(mmu, kernel_pmap, va);
2665 if (pte == NULL || !PTE_ISVALID(pte))
2669 dump_map[2].pa_size = va - dump_map[2].pa_start;
2674 * Map a set of physical memory pages into the kernel virtual address space.
2675 * Return a pointer to where it is mapped. This routine is intended to be used
2676 * for mapping device memory, NOT real memory.
2679 mmu_booke_mapdev(mmu_t mmu, vm_paddr_t pa, vm_size_t size)
2682 return (mmu_booke_mapdev_attr(mmu, pa, size, VM_MEMATTR_DEFAULT));
2686 mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
2694 * Check if this is premapped in TLB1. Note: this should probably also
2695 * check whether a sequence of TLB1 entries exist that match the
2696 * requirement, but now only checks the easy case.
2698 if (ma == VM_MEMATTR_DEFAULT) {
2699 for (i = 0; i < tlb1_idx; i++) {
2700 if (!(tlb1[i].mas1 & MAS1_VALID))
2702 if (pa >= tlb1[i].phys &&
2703 (pa + size) <= (tlb1[i].phys + tlb1[i].size))
2704 return (void *)(tlb1[i].virt +
2705 (pa - tlb1[i].phys));
2709 size = roundup(size, PAGE_SIZE);
2712 * We leave a hole for device direct mapping between the maximum user
2713 * address (0x8000000) and the minimum KVA address (0xc0000000). If
2714 * devices are in there, just map them 1:1. If not, map them to the
2715 * device mapping area about VM_MAX_KERNEL_ADDRESS. These mapped
2716 * addresses should be pulled from an allocator, but since we do not
2717 * ever free TLB1 entries, it is safe just to increment a counter.
2718 * Note that there isn't a lot of address space here (128 MB) and it
2719 * is not at all difficult to imagine running out, since that is a 4:1
2720 * compression from the 0xc0000000 - 0xf0000000 address space that gets
2723 if (pa >= (VM_MAXUSER_ADDRESS + PAGE_SIZE) &&
2724 (pa + size - 1) < VM_MIN_KERNEL_ADDRESS)
2727 va = atomic_fetchadd_int(&tlb1_map_base, size);
2731 sz = 1 << (ilog2(size) & ~1);
2733 printf("Wiring VA=%x to PA=%x (size=%x), "
2734 "using TLB1[%d]\n", va, pa, sz, tlb1_idx);
2735 tlb1_set_entry(va, pa, sz, tlb_calc_wimg(pa, ma));
2745 * 'Unmap' a range mapped by mmu_booke_mapdev().
2748 mmu_booke_unmapdev(mmu_t mmu, vm_offset_t va, vm_size_t size)
2750 #ifdef SUPPORTS_SHRINKING_TLB1
2751 vm_offset_t base, offset;
2754 * Unmap only if this is inside kernel virtual space.
2756 if ((va >= VM_MIN_KERNEL_ADDRESS) && (va <= VM_MAX_KERNEL_ADDRESS)) {
2757 base = trunc_page(va);
2758 offset = va & PAGE_MASK;
2759 size = roundup(offset + size, PAGE_SIZE);
2760 kva_free(base, size);
2766 * mmu_booke_object_init_pt preloads the ptes for a given object into the
2767 * specified pmap. This eliminates the blast of soft faults on process startup
2768 * and immediately after an mmap.
2771 mmu_booke_object_init_pt(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2772 vm_object_t object, vm_pindex_t pindex, vm_size_t size)
2775 VM_OBJECT_ASSERT_WLOCKED(object);
2776 KASSERT(object->type == OBJT_DEVICE || object->type == OBJT_SG,
2777 ("mmu_booke_object_init_pt: non-device object"));
2781 * Perform the pmap work for mincore.
2784 mmu_booke_mincore(mmu_t mmu, pmap_t pmap, vm_offset_t addr,
2785 vm_paddr_t *locked_pa)
2788 /* XXX: this should be implemented at some point */
2792 /**************************************************************************/
2794 /**************************************************************************/
2797 * Allocate a TID. If necessary, steal one from someone else.
2798 * The new TID is flushed from the TLB before returning.
2801 tid_alloc(pmap_t pmap)
2806 KASSERT((pmap != kernel_pmap), ("tid_alloc: kernel pmap"));
2808 CTR2(KTR_PMAP, "%s: s (pmap = %p)", __func__, pmap);
2810 thiscpu = PCPU_GET(cpuid);
2812 tid = PCPU_GET(tid_next);
2815 PCPU_SET(tid_next, tid + 1);
2817 /* If we are stealing TID then clear the relevant pmap's field */
2818 if (tidbusy[thiscpu][tid] != NULL) {
2820 CTR2(KTR_PMAP, "%s: warning: stealing tid %d", __func__, tid);
2822 tidbusy[thiscpu][tid]->pm_tid[thiscpu] = TID_NONE;
2824 /* Flush all entries from TLB0 matching this TID. */
2825 tid_flush(tid, tlb0_ways, tlb0_entries_per_way);
2828 tidbusy[thiscpu][tid] = pmap;
2829 pmap->pm_tid[thiscpu] = tid;
2830 __asm __volatile("msync; isync");
2832 CTR3(KTR_PMAP, "%s: e (%02d next = %02d)", __func__, tid,
2833 PCPU_GET(tid_next));
2838 /**************************************************************************/
2840 /**************************************************************************/
2843 tlb_print_entry(int i, uint32_t mas1, uint32_t mas2, uint32_t mas3,
2853 if (mas1 & MAS1_VALID)
2858 if (mas1 & MAS1_IPROT)
2863 as = (mas1 & MAS1_TS_MASK) ? 1 : 0;
2864 tid = MAS1_GETTID(mas1);
2866 tsize = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
2869 size = tsize2size(tsize);
2871 debugf("%3d: (%s) [AS=%d] "
2872 "sz = 0x%08x tsz = %d tid = %d mas1 = 0x%08x "
2873 "mas2(va) = 0x%08x mas3(pa) = 0x%08x mas7 = 0x%08x\n",
2874 i, desc, as, size, tsize, tid, mas1, mas2, mas3, mas7);
2877 /* Convert TLB0 va and way number to tlb0[] table index. */
2878 static inline unsigned int
2879 tlb0_tableidx(vm_offset_t va, unsigned int way)
2883 idx = (way * TLB0_ENTRIES_PER_WAY);
2884 idx += (va & MAS2_TLB0_ENTRY_IDX_MASK) >> MAS2_TLB0_ENTRY_IDX_SHIFT;
2889 * Invalidate TLB0 entry.
2892 tlb0_flush_entry(vm_offset_t va)
2895 CTR2(KTR_PMAP, "%s: s va=0x%08x", __func__, va);
2897 mtx_assert(&tlbivax_mutex, MA_OWNED);
2899 __asm __volatile("tlbivax 0, %0" :: "r"(va & MAS2_EPN_MASK));
2900 __asm __volatile("isync; msync");
2901 __asm __volatile("tlbsync; msync");
2903 CTR1(KTR_PMAP, "%s: e", __func__);
2906 /* Print out contents of the MAS registers for each TLB0 entry */
2908 tlb0_print_tlbentries(void)
2910 uint32_t mas0, mas1, mas2, mas3, mas7;
2911 int entryidx, way, idx;
2913 debugf("TLB0 entries:\n");
2914 for (way = 0; way < TLB0_WAYS; way ++)
2915 for (entryidx = 0; entryidx < TLB0_ENTRIES_PER_WAY; entryidx++) {
2917 mas0 = MAS0_TLBSEL(0) | MAS0_ESEL(way);
2918 mtspr(SPR_MAS0, mas0);
2919 __asm __volatile("isync");
2921 mas2 = entryidx << MAS2_TLB0_ENTRY_IDX_SHIFT;
2922 mtspr(SPR_MAS2, mas2);
2924 __asm __volatile("isync; tlbre");
2926 mas1 = mfspr(SPR_MAS1);
2927 mas2 = mfspr(SPR_MAS2);
2928 mas3 = mfspr(SPR_MAS3);
2929 mas7 = mfspr(SPR_MAS7);
2931 idx = tlb0_tableidx(mas2, way);
2932 tlb_print_entry(idx, mas1, mas2, mas3, mas7);
2936 /**************************************************************************/
2938 /**************************************************************************/
2941 * TLB1 mapping notes:
2943 * TLB1[0] Kernel text and data.
2944 * TLB1[1-15] Additional kernel text and data mappings (if required), PCI
2945 * windows, other devices mappings.
2949 * Write given entry to TLB1 hardware.
2950 * Use 32 bit pa, clear 4 high-order bits of RPN (mas7).
2953 tlb1_write_entry(unsigned int idx)
2955 uint32_t mas0, mas7;
2957 //debugf("tlb1_write_entry: s\n");
2959 /* Clear high order RPN bits */
2963 mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(idx);
2964 //debugf("tlb1_write_entry: mas0 = 0x%08x\n", mas0);
2966 mtspr(SPR_MAS0, mas0);
2967 __asm __volatile("isync");
2968 mtspr(SPR_MAS1, tlb1[idx].mas1);
2969 __asm __volatile("isync");
2970 mtspr(SPR_MAS2, tlb1[idx].mas2);
2971 __asm __volatile("isync");
2972 mtspr(SPR_MAS3, tlb1[idx].mas3);
2973 __asm __volatile("isync");
2974 mtspr(SPR_MAS7, mas7);
2975 __asm __volatile("isync; tlbwe; isync; msync");
2977 //debugf("tlb1_write_entry: e\n");
2981 * Return the largest uint value log such that 2^log <= num.
2984 ilog2(unsigned int num)
2988 __asm ("cntlzw %0, %1" : "=r" (lz) : "r" (num));
2993 * Convert TLB TSIZE value to mapped region size.
2996 tsize2size(unsigned int tsize)
3001 * size = 4^tsize * 2^10 = 2^(2 * tsize - 10)
3004 return ((1 << (2 * tsize)) * 1024);
3008 * Convert region size (must be power of 4) to TLB TSIZE value.
3011 size2tsize(vm_size_t size)
3014 return (ilog2(size) / 2 - 5);
3018 * Register permanent kernel mapping in TLB1.
3020 * Entries are created starting from index 0 (current free entry is
3021 * kept in tlb1_idx) and are not supposed to be invalidated.
3024 tlb1_set_entry(vm_offset_t va, vm_offset_t pa, vm_size_t size,
3030 index = atomic_fetchadd_int(&tlb1_idx, 1);
3031 if (index >= TLB1_ENTRIES) {
3032 printf("tlb1_set_entry: TLB1 full!\n");
3036 /* Convert size to TSIZE */
3037 tsize = size2tsize(size);
3039 tid = (TID_KERNEL << MAS1_TID_SHIFT) & MAS1_TID_MASK;
3040 /* XXX TS is hard coded to 0 for now as we only use single address space */
3041 ts = (0 << MAS1_TS_SHIFT) & MAS1_TS_MASK;
3044 * Atomicity is preserved by the atomic increment above since nothing
3045 * is ever removed from tlb1.
3048 tlb1[index].phys = pa;
3049 tlb1[index].virt = va;
3050 tlb1[index].size = size;
3051 tlb1[index].mas1 = MAS1_VALID | MAS1_IPROT | ts | tid;
3052 tlb1[index].mas1 |= ((tsize << MAS1_TSIZE_SHIFT) & MAS1_TSIZE_MASK);
3053 tlb1[index].mas2 = (va & MAS2_EPN_MASK) | flags;
3055 /* Set supervisor RWX permission bits */
3056 tlb1[index].mas3 = (pa & MAS3_RPN) | MAS3_SR | MAS3_SW | MAS3_SX;
3058 tlb1_write_entry(index);
3061 * XXX in general TLB1 updates should be propagated between CPUs,
3062 * since current design assumes to have the same TLB1 set-up on all
3069 * Map in contiguous RAM region into the TLB1 using maximum of
3070 * KERNEL_REGION_MAX_TLB_ENTRIES entries.
3072 * If necessary round up last entry size and return total size
3073 * used by all allocated entries.
3076 tlb1_mapin_region(vm_offset_t va, vm_paddr_t pa, vm_size_t size)
3078 vm_size_t pgs[KERNEL_REGION_MAX_TLB_ENTRIES];
3079 vm_size_t mapped, pgsz, base, mask;
3082 /* Round up to the next 1M */
3083 size = (size + (1 << 20) - 1) & ~((1 << 20) - 1);
3088 pgsz = 64*1024*1024;
3089 while (mapped < size) {
3090 while (mapped < size && idx < KERNEL_REGION_MAX_TLB_ENTRIES) {
3091 while (pgsz > (size - mapped))
3097 /* We under-map. Correct for this. */
3098 if (mapped < size) {
3099 while (pgs[idx - 1] == pgsz) {
3103 /* XXX We may increase beyond out starting point. */
3112 /* Align address to the boundary */
3114 va = (va + mask) & ~mask;
3115 pa = (pa + mask) & ~mask;
3118 for (idx = 0; idx < nents; idx++) {
3120 debugf("%u: %x -> %x, size=%x\n", idx, pa, va, pgsz);
3121 tlb1_set_entry(va, pa, pgsz, _TLB_ENTRY_MEM);
3126 mapped = (va - base);
3127 printf("mapped size 0x%08x (wasted space 0x%08x)\n",
3128 mapped, mapped - size);
3133 * TLB1 initialization routine, to be called after the very first
3134 * assembler level setup done in locore.S.
3139 uint32_t mas0, mas1, mas2, mas3;
3143 if (bootinfo != NULL && bootinfo[0] != 1) {
3144 tlb1_idx = *((uint16_t *)(bootinfo + 8));
3148 /* The first entry/entries are used to map the kernel. */
3149 for (i = 0; i < tlb1_idx; i++) {
3150 mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3151 mtspr(SPR_MAS0, mas0);
3152 __asm __volatile("isync; tlbre");
3154 mas1 = mfspr(SPR_MAS1);
3155 if ((mas1 & MAS1_VALID) == 0)
3158 mas2 = mfspr(SPR_MAS2);
3159 mas3 = mfspr(SPR_MAS3);
3161 tlb1[i].mas1 = mas1;
3162 tlb1[i].mas2 = mfspr(SPR_MAS2);
3163 tlb1[i].mas3 = mas3;
3164 tlb1[i].virt = mas2 & MAS2_EPN_MASK;
3165 tlb1[i].phys = mas3 & MAS3_RPN;
3168 kernload = mas3 & MAS3_RPN;
3170 tsz = (mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3171 tlb1[i].size = (tsz > 0) ? tsize2size(tsz) : 0;
3172 kernsize += tlb1[i].size;
3176 bp_ntlb1s = tlb1_idx;
3179 /* Purge the remaining entries */
3180 for (i = tlb1_idx; i < TLB1_ENTRIES; i++)
3181 tlb1_write_entry(i);
3183 /* Setup TLB miss defaults */
3184 set_mas4_defaults();
3188 pmap_early_io_map(vm_paddr_t pa, vm_size_t size)
3194 KASSERT(!pmap_bootstrapped, ("Do not use after PMAP is up!"));
3196 for (i = 0; i < tlb1_idx; i++) {
3197 if (!(tlb1[i].mas1 & MAS1_VALID))
3199 if (pa >= tlb1[i].phys && (pa + size) <=
3200 (tlb1[i].phys + tlb1[i].size))
3201 return (tlb1[i].virt + (pa - tlb1[i].phys));
3204 pa_base = trunc_page(pa);
3205 size = roundup(size + (pa - pa_base), PAGE_SIZE);
3206 tlb1_map_base = roundup2(tlb1_map_base, 1 << (ilog2(size) & ~1));
3207 va = tlb1_map_base + (pa - pa_base);
3210 sz = 1 << (ilog2(size) & ~1);
3211 tlb1_set_entry(tlb1_map_base, pa_base, sz, _TLB_ENTRY_IO);
3214 tlb1_map_base += sz;
3218 bp_ntlb1s = tlb1_idx;
3225 * Setup MAS4 defaults.
3226 * These values are loaded to MAS0-2 on a TLB miss.
3229 set_mas4_defaults(void)
3233 /* Defaults: TLB0, PID0, TSIZED=4K */
3234 mas4 = MAS4_TLBSELD0;
3235 mas4 |= (TLB_SIZE_4K << MAS4_TSIZED_SHIFT) & MAS4_TSIZED_MASK;
3239 mtspr(SPR_MAS4, mas4);
3240 __asm __volatile("isync");
3244 * Print out contents of the MAS registers for each TLB1 entry
3247 tlb1_print_tlbentries(void)
3249 uint32_t mas0, mas1, mas2, mas3, mas7;
3252 debugf("TLB1 entries:\n");
3253 for (i = 0; i < TLB1_ENTRIES; i++) {
3255 mas0 = MAS0_TLBSEL(1) | MAS0_ESEL(i);
3256 mtspr(SPR_MAS0, mas0);
3258 __asm __volatile("isync; tlbre");
3260 mas1 = mfspr(SPR_MAS1);
3261 mas2 = mfspr(SPR_MAS2);
3262 mas3 = mfspr(SPR_MAS3);
3263 mas7 = mfspr(SPR_MAS7);
3265 tlb_print_entry(i, mas1, mas2, mas3, mas7);
3270 * Print out contents of the in-ram tlb1 table.
3273 tlb1_print_entries(void)
3277 debugf("tlb1[] table entries:\n");
3278 for (i = 0; i < TLB1_ENTRIES; i++)
3279 tlb_print_entry(i, tlb1[i].mas1, tlb1[i].mas2, tlb1[i].mas3, 0);
3283 * Return 0 if the physical IO range is encompassed by one of the
3284 * the TLB1 entries, otherwise return related error code.
3287 tlb1_iomapped(int i, vm_paddr_t pa, vm_size_t size, vm_offset_t *va)
3290 vm_paddr_t pa_start;
3292 unsigned int entry_tsize;
3293 vm_size_t entry_size;
3295 *va = (vm_offset_t)NULL;
3297 /* Skip invalid entries */
3298 if (!(tlb1[i].mas1 & MAS1_VALID))
3302 * The entry must be cache-inhibited, guarded, and r/w
3303 * so it can function as an i/o page
3305 prot = tlb1[i].mas2 & (MAS2_I | MAS2_G);
3306 if (prot != (MAS2_I | MAS2_G))
3309 prot = tlb1[i].mas3 & (MAS3_SR | MAS3_SW);
3310 if (prot != (MAS3_SR | MAS3_SW))
3313 /* The address should be within the entry range. */
3314 entry_tsize = (tlb1[i].mas1 & MAS1_TSIZE_MASK) >> MAS1_TSIZE_SHIFT;
3315 KASSERT((entry_tsize), ("tlb1_iomapped: invalid entry tsize"));
3317 entry_size = tsize2size(entry_tsize);
3318 pa_start = tlb1[i].mas3 & MAS3_RPN;
3319 pa_end = pa_start + entry_size - 1;
3321 if ((pa < pa_start) || ((pa + size) > pa_end))
3324 /* Return virtual address of this mapping. */
3325 *va = (tlb1[i].mas2 & MAS2_EPN_MASK) + (pa - pa_start);