sparc: Move of_console_{device,path,options} info prom_common.c
[linux-2.6] / arch / sparc / mm / sun4c.c
1 /* sun4c.c: Doing in software what should be done in hardware.
2  *
3  * Copyright (C) 1996 David S. Miller (davem@davemloft.net)
4  * Copyright (C) 1996 Eddie C. Dost (ecd@skynet.be)
5  * Copyright (C) 1996 Andrew Tridgell (Andrew.Tridgell@anu.edu.au)
6  * Copyright (C) 1997-2000 Anton Blanchard (anton@samba.org)
7  * Copyright (C) 1998 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
8  */
9
10 #define NR_TASK_BUCKETS 512
11
12 #include <linux/kernel.h>
13 #include <linux/mm.h>
14 #include <linux/init.h>
15 #include <linux/bootmem.h>
16 #include <linux/highmem.h>
17 #include <linux/fs.h>
18 #include <linux/seq_file.h>
19 #include <linux/scatterlist.h>
20
21 #include <asm/page.h>
22 #include <asm/pgalloc.h>
23 #include <asm/pgtable.h>
24 #include <asm/vaddrs.h>
25 #include <asm/idprom.h>
26 #include <asm/machines.h>
27 #include <asm/memreg.h>
28 #include <asm/processor.h>
29 #include <asm/auxio.h>
30 #include <asm/io.h>
31 #include <asm/oplib.h>
32 #include <asm/openprom.h>
33 #include <asm/mmu_context.h>
34 #include <asm/highmem.h>
35 #include <asm/btfixup.h>
36 #include <asm/cacheflush.h>
37 #include <asm/tlbflush.h>
38
39 /* Because of our dynamic kernel TLB miss strategy, and how
40  * our DVMA mapping allocation works, you _MUST_:
41  *
42  * 1) Disable interrupts _and_ not touch any dynamic kernel
43  *    memory while messing with kernel MMU state.  By
44  *    dynamic memory I mean any object which is not in
45  *    the kernel image itself or a thread_union (both of
46  *    which are locked into the MMU).
47  * 2) Disable interrupts while messing with user MMU state.
48  */
49
50 extern int num_segmaps, num_contexts;
51
52 extern unsigned long page_kernel;
53
54 /* That's it, we prom_halt() on sun4c if the cache size is something other than 65536.
55  * So let's save some cycles and just use that everywhere except for that bootup
56  * sanity check.
57  */
58 #define SUN4C_VAC_SIZE 65536
59
60 #define SUN4C_KERNEL_BUCKETS 32
61
62 /* Flushing the cache. */
63 struct sun4c_vac_props sun4c_vacinfo;
64 unsigned long sun4c_kernel_faults;
65
66 /* Invalidate every sun4c cache line tag. */
67 static void __init sun4c_flush_all(void)
68 {
69         unsigned long begin, end;
70
71         if (sun4c_vacinfo.on)
72                 panic("SUN4C: AIEEE, trying to invalidate vac while it is on.");
73
74         /* Clear 'valid' bit in all cache line tags */
75         begin = AC_CACHETAGS;
76         end = (AC_CACHETAGS + SUN4C_VAC_SIZE);
77         while (begin < end) {
78                 __asm__ __volatile__("sta %%g0, [%0] %1\n\t" : :
79                                      "r" (begin), "i" (ASI_CONTROL));
80                 begin += sun4c_vacinfo.linesize;
81         }
82 }
83
84 static void sun4c_flush_context_hw(void)
85 {
86         unsigned long end = SUN4C_VAC_SIZE;
87
88         __asm__ __volatile__(
89                 "1:     addcc   %0, -4096, %0\n\t"
90                 "       bne     1b\n\t"
91                 "        sta    %%g0, [%0] %2"
92         : "=&r" (end)
93         : "0" (end), "i" (ASI_HWFLUSHCONTEXT)
94         : "cc");
95 }
96
97 /* Must be called minimally with IRQs disabled. */
98 static void sun4c_flush_segment_hw(unsigned long addr)
99 {
100         if (sun4c_get_segmap(addr) != invalid_segment) {
101                 unsigned long vac_size = SUN4C_VAC_SIZE;
102
103                 __asm__ __volatile__(
104                         "1:     addcc   %0, -4096, %0\n\t"
105                         "       bne     1b\n\t"
106                         "        sta    %%g0, [%2 + %0] %3"
107                         : "=&r" (vac_size)
108                         : "0" (vac_size), "r" (addr), "i" (ASI_HWFLUSHSEG)
109                         : "cc");
110         }
111 }
112
113 /* File local boot time fixups. */
114 BTFIXUPDEF_CALL(void, sun4c_flush_page, unsigned long)
115 BTFIXUPDEF_CALL(void, sun4c_flush_segment, unsigned long)
116 BTFIXUPDEF_CALL(void, sun4c_flush_context, void)
117
118 #define sun4c_flush_page(addr) BTFIXUP_CALL(sun4c_flush_page)(addr)
119 #define sun4c_flush_segment(addr) BTFIXUP_CALL(sun4c_flush_segment)(addr)
120 #define sun4c_flush_context() BTFIXUP_CALL(sun4c_flush_context)()
121
122 /* Must be called minimally with interrupts disabled. */
123 static void sun4c_flush_page_hw(unsigned long addr)
124 {
125         addr &= PAGE_MASK;
126         if ((int)sun4c_get_pte(addr) < 0)
127                 __asm__ __volatile__("sta %%g0, [%0] %1"
128                                      : : "r" (addr), "i" (ASI_HWFLUSHPAGE));
129 }
130
131 /* Don't inline the software version as it eats too many cache lines if expanded. */
132 static void sun4c_flush_context_sw(void)
133 {
134         unsigned long nbytes = SUN4C_VAC_SIZE;
135         unsigned long lsize = sun4c_vacinfo.linesize;
136
137         __asm__ __volatile__(
138         "add    %2, %2, %%g1\n\t"
139         "add    %2, %%g1, %%g2\n\t"
140         "add    %2, %%g2, %%g3\n\t"
141         "add    %2, %%g3, %%g4\n\t"
142         "add    %2, %%g4, %%g5\n\t"
143         "add    %2, %%g5, %%o4\n\t"
144         "add    %2, %%o4, %%o5\n"
145         "1:\n\t"
146         "subcc  %0, %%o5, %0\n\t"
147         "sta    %%g0, [%0] %3\n\t"
148         "sta    %%g0, [%0 + %2] %3\n\t"
149         "sta    %%g0, [%0 + %%g1] %3\n\t"
150         "sta    %%g0, [%0 + %%g2] %3\n\t"
151         "sta    %%g0, [%0 + %%g3] %3\n\t"
152         "sta    %%g0, [%0 + %%g4] %3\n\t"
153         "sta    %%g0, [%0 + %%g5] %3\n\t"
154         "bg     1b\n\t"
155         " sta   %%g0, [%1 + %%o4] %3\n"
156         : "=&r" (nbytes)
157         : "0" (nbytes), "r" (lsize), "i" (ASI_FLUSHCTX)
158         : "g1", "g2", "g3", "g4", "g5", "o4", "o5", "cc");
159 }
160
161 /* Don't inline the software version as it eats too many cache lines if expanded. */
162 static void sun4c_flush_segment_sw(unsigned long addr)
163 {
164         if (sun4c_get_segmap(addr) != invalid_segment) {
165                 unsigned long nbytes = SUN4C_VAC_SIZE;
166                 unsigned long lsize = sun4c_vacinfo.linesize;
167
168                 __asm__ __volatile__(
169                 "add    %2, %2, %%g1\n\t"
170                 "add    %2, %%g1, %%g2\n\t"
171                 "add    %2, %%g2, %%g3\n\t"
172                 "add    %2, %%g3, %%g4\n\t"
173                 "add    %2, %%g4, %%g5\n\t"
174                 "add    %2, %%g5, %%o4\n\t"
175                 "add    %2, %%o4, %%o5\n"
176                 "1:\n\t"
177                 "subcc  %1, %%o5, %1\n\t"
178                 "sta    %%g0, [%0] %6\n\t"
179                 "sta    %%g0, [%0 + %2] %6\n\t"
180                 "sta    %%g0, [%0 + %%g1] %6\n\t"
181                 "sta    %%g0, [%0 + %%g2] %6\n\t"
182                 "sta    %%g0, [%0 + %%g3] %6\n\t"
183                 "sta    %%g0, [%0 + %%g4] %6\n\t"
184                 "sta    %%g0, [%0 + %%g5] %6\n\t"
185                 "sta    %%g0, [%0 + %%o4] %6\n\t"
186                 "bg     1b\n\t"
187                 " add   %0, %%o5, %0\n"
188                 : "=&r" (addr), "=&r" (nbytes), "=&r" (lsize)
189                 : "0" (addr), "1" (nbytes), "2" (lsize),
190                   "i" (ASI_FLUSHSEG)
191                 : "g1", "g2", "g3", "g4", "g5", "o4", "o5", "cc");
192         }
193 }
194
195 /* Don't inline the software version as it eats too many cache lines if expanded. */
196 static void sun4c_flush_page_sw(unsigned long addr)
197 {
198         addr &= PAGE_MASK;
199         if ((sun4c_get_pte(addr) & (_SUN4C_PAGE_NOCACHE | _SUN4C_PAGE_VALID)) ==
200             _SUN4C_PAGE_VALID) {
201                 unsigned long left = PAGE_SIZE;
202                 unsigned long lsize = sun4c_vacinfo.linesize;
203
204                 __asm__ __volatile__(
205                 "add    %2, %2, %%g1\n\t"
206                 "add    %2, %%g1, %%g2\n\t"
207                 "add    %2, %%g2, %%g3\n\t"
208                 "add    %2, %%g3, %%g4\n\t"
209                 "add    %2, %%g4, %%g5\n\t"
210                 "add    %2, %%g5, %%o4\n\t"
211                 "add    %2, %%o4, %%o5\n"
212                 "1:\n\t"
213                 "subcc  %1, %%o5, %1\n\t"
214                 "sta    %%g0, [%0] %6\n\t"
215                 "sta    %%g0, [%0 + %2] %6\n\t"
216                 "sta    %%g0, [%0 + %%g1] %6\n\t"
217                 "sta    %%g0, [%0 + %%g2] %6\n\t"
218                 "sta    %%g0, [%0 + %%g3] %6\n\t"
219                 "sta    %%g0, [%0 + %%g4] %6\n\t"
220                 "sta    %%g0, [%0 + %%g5] %6\n\t"
221                 "sta    %%g0, [%0 + %%o4] %6\n\t"
222                 "bg     1b\n\t"
223                 " add   %0, %%o5, %0\n"
224                 : "=&r" (addr), "=&r" (left), "=&r" (lsize)
225                 : "0" (addr), "1" (left), "2" (lsize),
226                   "i" (ASI_FLUSHPG)
227                 : "g1", "g2", "g3", "g4", "g5", "o4", "o5", "cc");
228         }
229 }
230
231 /* The sun4c's do have an on chip store buffer.  And the way you
232  * clear them out isn't so obvious.  The only way I can think of
233  * to accomplish this is to read the current context register,
234  * store the same value there, then read an external hardware
235  * register.
236  */
237 void sun4c_complete_all_stores(void)
238 {
239         volatile int _unused;
240
241         _unused = sun4c_get_context();
242         sun4c_set_context(_unused);
243         _unused = get_auxio();
244 }
245
246 /* Bootup utility functions. */
247 static inline void sun4c_init_clean_segmap(unsigned char pseg)
248 {
249         unsigned long vaddr;
250
251         sun4c_put_segmap(0, pseg);
252         for (vaddr = 0; vaddr < SUN4C_REAL_PGDIR_SIZE; vaddr += PAGE_SIZE)
253                 sun4c_put_pte(vaddr, 0);
254         sun4c_put_segmap(0, invalid_segment);
255 }
256
257 static inline void sun4c_init_clean_mmu(unsigned long kernel_end)
258 {
259         unsigned long vaddr;
260         unsigned char savectx, ctx;
261
262         savectx = sun4c_get_context();
263         for (ctx = 0; ctx < num_contexts; ctx++) {
264                 sun4c_set_context(ctx);
265                 for (vaddr = 0; vaddr < 0x20000000; vaddr += SUN4C_REAL_PGDIR_SIZE)
266                         sun4c_put_segmap(vaddr, invalid_segment);
267                 for (vaddr = 0xe0000000; vaddr < KERNBASE; vaddr += SUN4C_REAL_PGDIR_SIZE)
268                         sun4c_put_segmap(vaddr, invalid_segment);
269                 for (vaddr = kernel_end; vaddr < KADB_DEBUGGER_BEGVM; vaddr += SUN4C_REAL_PGDIR_SIZE)
270                         sun4c_put_segmap(vaddr, invalid_segment);
271                 for (vaddr = LINUX_OPPROM_ENDVM; vaddr; vaddr += SUN4C_REAL_PGDIR_SIZE)
272                         sun4c_put_segmap(vaddr, invalid_segment);
273         }
274         sun4c_set_context(savectx);
275 }
276
277 void __init sun4c_probe_vac(void)
278 {
279         sun4c_disable_vac();
280
281         if ((idprom->id_machtype == (SM_SUN4C | SM_4C_SS1)) ||
282             (idprom->id_machtype == (SM_SUN4C | SM_4C_SS1PLUS))) {
283                 /* PROM on SS1 lacks this info, to be super safe we
284                  * hard code it here since this arch is cast in stone.
285                  */
286                 sun4c_vacinfo.num_bytes = 65536;
287                 sun4c_vacinfo.linesize = 16;
288         } else {
289                 sun4c_vacinfo.num_bytes =
290                  prom_getintdefault(prom_root_node, "vac-size", 65536);
291                 sun4c_vacinfo.linesize =
292                  prom_getintdefault(prom_root_node, "vac-linesize", 16);
293         }
294         sun4c_vacinfo.do_hwflushes =
295          prom_getintdefault(prom_root_node, "vac-hwflush", 0);
296
297         if (sun4c_vacinfo.do_hwflushes == 0)
298                 sun4c_vacinfo.do_hwflushes =
299                  prom_getintdefault(prom_root_node, "vac_hwflush", 0);
300
301         if (sun4c_vacinfo.num_bytes != 65536) {
302                 prom_printf("WEIRD Sun4C VAC cache size, "
303                             "tell sparclinux@vger.kernel.org");
304                 prom_halt();
305         }
306
307         switch (sun4c_vacinfo.linesize) {
308         case 16:
309                 sun4c_vacinfo.log2lsize = 4;
310                 break;
311         case 32:
312                 sun4c_vacinfo.log2lsize = 5;
313                 break;
314         default:
315                 prom_printf("probe_vac: Didn't expect vac-linesize of %d, halting\n",
316                             sun4c_vacinfo.linesize);
317                 prom_halt();
318         };
319
320         sun4c_flush_all();
321         sun4c_enable_vac();
322 }
323
324 /* Patch instructions for the low level kernel fault handler. */
325 extern unsigned long invalid_segment_patch1, invalid_segment_patch1_ff;
326 extern unsigned long invalid_segment_patch2, invalid_segment_patch2_ff;
327 extern unsigned long invalid_segment_patch1_1ff, invalid_segment_patch2_1ff;
328 extern unsigned long num_context_patch1, num_context_patch1_16;
329 extern unsigned long num_context_patch2_16;
330 extern unsigned long vac_linesize_patch, vac_linesize_patch_32;
331 extern unsigned long vac_hwflush_patch1, vac_hwflush_patch1_on;
332 extern unsigned long vac_hwflush_patch2, vac_hwflush_patch2_on;
333
334 #define PATCH_INSN(src, dst) do {       \
335                 daddr = &(dst);         \
336                 iaddr = &(src);         \
337                 *daddr = *iaddr;        \
338         } while (0)
339
340 static void __init patch_kernel_fault_handler(void)
341 {
342         unsigned long *iaddr, *daddr;
343
344         switch (num_segmaps) {
345                 case 128:
346                         /* Default, nothing to do. */
347                         break;
348                 case 256:
349                         PATCH_INSN(invalid_segment_patch1_ff,
350                                    invalid_segment_patch1);
351                         PATCH_INSN(invalid_segment_patch2_ff,
352                                    invalid_segment_patch2);
353                         break;
354                 case 512:
355                         PATCH_INSN(invalid_segment_patch1_1ff,
356                                    invalid_segment_patch1);
357                         PATCH_INSN(invalid_segment_patch2_1ff,
358                                    invalid_segment_patch2);
359                         break;
360                 default:
361                         prom_printf("Unhandled number of segmaps: %d\n",
362                                     num_segmaps);
363                         prom_halt();
364         };
365         switch (num_contexts) {
366                 case 8:
367                         /* Default, nothing to do. */
368                         break;
369                 case 16:
370                         PATCH_INSN(num_context_patch1_16,
371                                    num_context_patch1);
372                         break;
373                 default:
374                         prom_printf("Unhandled number of contexts: %d\n",
375                                     num_contexts);
376                         prom_halt();
377         };
378
379         if (sun4c_vacinfo.do_hwflushes != 0) {
380                 PATCH_INSN(vac_hwflush_patch1_on, vac_hwflush_patch1);
381                 PATCH_INSN(vac_hwflush_patch2_on, vac_hwflush_patch2);
382         } else {
383                 switch (sun4c_vacinfo.linesize) {
384                 case 16:
385                         /* Default, nothing to do. */
386                         break;
387                 case 32:
388                         PATCH_INSN(vac_linesize_patch_32, vac_linesize_patch);
389                         break;
390                 default:
391                         prom_printf("Impossible VAC linesize %d, halting...\n",
392                                     sun4c_vacinfo.linesize);
393                         prom_halt();
394                 };
395         }
396 }
397
398 static void __init sun4c_probe_mmu(void)
399 {
400         if ((idprom->id_machtype == (SM_SUN4C | SM_4C_SS1)) ||
401             (idprom->id_machtype == (SM_SUN4C | SM_4C_SS1PLUS))) {
402                 /* Hardcode these just to be safe, PROM on SS1 does
403                 * not have this info available in the root node.
404                 */
405                 num_segmaps = 128;
406                 num_contexts = 8;
407         } else {
408                 num_segmaps =
409                     prom_getintdefault(prom_root_node, "mmu-npmg", 128);
410                 num_contexts =
411                     prom_getintdefault(prom_root_node, "mmu-nctx", 0x8);
412         }
413         patch_kernel_fault_handler();
414 }
415
416 volatile unsigned long __iomem *sun4c_memerr_reg = NULL;
417
418 void __init sun4c_probe_memerr_reg(void)
419 {
420         int node;
421         struct linux_prom_registers regs[1];
422
423         node = prom_getchild(prom_root_node);
424         node = prom_searchsiblings(prom_root_node, "memory-error");
425         if (!node)
426                 return;
427         if (prom_getproperty(node, "reg", (char *)regs, sizeof(regs)) <= 0)
428                 return;
429         /* hmm I think regs[0].which_io is zero here anyways */
430         sun4c_memerr_reg = ioremap(regs[0].phys_addr, regs[0].reg_size);
431 }
432
433 static inline void sun4c_init_ss2_cache_bug(void)
434 {
435         extern unsigned long start;
436
437         if ((idprom->id_machtype == (SM_SUN4C | SM_4C_SS2)) ||
438             (idprom->id_machtype == (SM_SUN4C | SM_4C_IPX)) ||
439             (idprom->id_machtype == (SM_SUN4C | SM_4C_ELC))) {
440                 /* Whee.. */
441                 printk("SS2 cache bug detected, uncaching trap table page\n");
442                 sun4c_flush_page((unsigned int) &start);
443                 sun4c_put_pte(((unsigned long) &start),
444                         (sun4c_get_pte((unsigned long) &start) | _SUN4C_PAGE_NOCACHE));
445         }
446 }
447
448 /* Addr is always aligned on a page boundary for us already. */
449 static int sun4c_map_dma_area(struct device *dev, dma_addr_t *pba, unsigned long va,
450                               unsigned long addr, int len)
451 {
452         unsigned long page, end;
453
454         *pba = addr;
455
456         end = PAGE_ALIGN((addr + len));
457         while (addr < end) {
458                 page = va;
459                 sun4c_flush_page(page);
460                 page -= PAGE_OFFSET;
461                 page >>= PAGE_SHIFT;
462                 page |= (_SUN4C_PAGE_VALID | _SUN4C_PAGE_DIRTY |
463                          _SUN4C_PAGE_NOCACHE | _SUN4C_PAGE_PRIV);
464                 sun4c_put_pte(addr, page);
465                 addr += PAGE_SIZE;
466                 va += PAGE_SIZE;
467         }
468
469         return 0;
470 }
471
472 static void sun4c_unmap_dma_area(struct device *dev, unsigned long busa, int len)
473 {
474         /* Fortunately for us, bus_addr == uncached_virt in sun4c. */
475         /* XXX Implement this */
476 }
477
478 /* TLB management. */
479
480 /* Don't change this struct without changing entry.S. This is used
481  * in the in-window kernel fault handler, and you don't want to mess
482  * with that. (See sun4c_fault in entry.S).
483  */
484 struct sun4c_mmu_entry {
485         struct sun4c_mmu_entry *next;
486         struct sun4c_mmu_entry *prev;
487         unsigned long vaddr;
488         unsigned char pseg;
489         unsigned char locked;
490
491         /* For user mappings only, and completely hidden from kernel
492          * TLB miss code.
493          */
494         unsigned char ctx;
495         struct sun4c_mmu_entry *lru_next;
496         struct sun4c_mmu_entry *lru_prev;
497 };
498
499 static struct sun4c_mmu_entry mmu_entry_pool[SUN4C_MAX_SEGMAPS];
500
501 static void __init sun4c_init_mmu_entry_pool(void)
502 {
503         int i;
504
505         for (i=0; i < SUN4C_MAX_SEGMAPS; i++) {
506                 mmu_entry_pool[i].pseg = i;
507                 mmu_entry_pool[i].next = NULL;
508                 mmu_entry_pool[i].prev = NULL;
509                 mmu_entry_pool[i].vaddr = 0;
510                 mmu_entry_pool[i].locked = 0;
511                 mmu_entry_pool[i].ctx = 0;
512                 mmu_entry_pool[i].lru_next = NULL;
513                 mmu_entry_pool[i].lru_prev = NULL;
514         }
515         mmu_entry_pool[invalid_segment].locked = 1;
516 }
517
518 static inline void fix_permissions(unsigned long vaddr, unsigned long bits_on,
519                                    unsigned long bits_off)
520 {
521         unsigned long start, end;
522
523         end = vaddr + SUN4C_REAL_PGDIR_SIZE;
524         for (start = vaddr; start < end; start += PAGE_SIZE)
525                 if (sun4c_get_pte(start) & _SUN4C_PAGE_VALID)
526                         sun4c_put_pte(start, (sun4c_get_pte(start) | bits_on) &
527                                       ~bits_off);
528 }
529
530 static inline void sun4c_init_map_kernelprom(unsigned long kernel_end)
531 {
532         unsigned long vaddr;
533         unsigned char pseg, ctx;
534
535         for (vaddr = KADB_DEBUGGER_BEGVM;
536              vaddr < LINUX_OPPROM_ENDVM;
537              vaddr += SUN4C_REAL_PGDIR_SIZE) {
538                 pseg = sun4c_get_segmap(vaddr);
539                 if (pseg != invalid_segment) {
540                         mmu_entry_pool[pseg].locked = 1;
541                         for (ctx = 0; ctx < num_contexts; ctx++)
542                                 prom_putsegment(ctx, vaddr, pseg);
543                         fix_permissions(vaddr, _SUN4C_PAGE_PRIV, 0);
544                 }
545         }
546
547         for (vaddr = KERNBASE; vaddr < kernel_end; vaddr += SUN4C_REAL_PGDIR_SIZE) {
548                 pseg = sun4c_get_segmap(vaddr);
549                 mmu_entry_pool[pseg].locked = 1;
550                 for (ctx = 0; ctx < num_contexts; ctx++)
551                         prom_putsegment(ctx, vaddr, pseg);
552                 fix_permissions(vaddr, _SUN4C_PAGE_PRIV, _SUN4C_PAGE_NOCACHE);
553         }
554 }
555
556 static void __init sun4c_init_lock_area(unsigned long start, unsigned long end)
557 {
558         int i, ctx;
559
560         while (start < end) {
561                 for (i = 0; i < invalid_segment; i++)
562                         if (!mmu_entry_pool[i].locked)
563                                 break;
564                 mmu_entry_pool[i].locked = 1;
565                 sun4c_init_clean_segmap(i);
566                 for (ctx = 0; ctx < num_contexts; ctx++)
567                         prom_putsegment(ctx, start, mmu_entry_pool[i].pseg);
568                 start += SUN4C_REAL_PGDIR_SIZE;
569         }
570 }
571
572 /* Don't change this struct without changing entry.S. This is used
573  * in the in-window kernel fault handler, and you don't want to mess
574  * with that. (See sun4c_fault in entry.S).
575  */
576 struct sun4c_mmu_ring {
577         struct sun4c_mmu_entry ringhd;
578         int num_entries;
579 };
580
581 static struct sun4c_mmu_ring sun4c_context_ring[SUN4C_MAX_CONTEXTS]; /* used user entries */
582 static struct sun4c_mmu_ring sun4c_ufree_ring;       /* free user entries */
583 static struct sun4c_mmu_ring sun4c_ulru_ring;        /* LRU user entries */
584 struct sun4c_mmu_ring sun4c_kernel_ring;      /* used kernel entries */
585 struct sun4c_mmu_ring sun4c_kfree_ring;       /* free kernel entries */
586
587 static inline void sun4c_init_rings(void)
588 {
589         int i;
590
591         for (i = 0; i < SUN4C_MAX_CONTEXTS; i++) {
592                 sun4c_context_ring[i].ringhd.next =
593                         sun4c_context_ring[i].ringhd.prev =
594                         &sun4c_context_ring[i].ringhd;
595                 sun4c_context_ring[i].num_entries = 0;
596         }
597         sun4c_ufree_ring.ringhd.next = sun4c_ufree_ring.ringhd.prev =
598                 &sun4c_ufree_ring.ringhd;
599         sun4c_ufree_ring.num_entries = 0;
600         sun4c_ulru_ring.ringhd.lru_next = sun4c_ulru_ring.ringhd.lru_prev =
601                 &sun4c_ulru_ring.ringhd;
602         sun4c_ulru_ring.num_entries = 0;
603         sun4c_kernel_ring.ringhd.next = sun4c_kernel_ring.ringhd.prev =
604                 &sun4c_kernel_ring.ringhd;
605         sun4c_kernel_ring.num_entries = 0;
606         sun4c_kfree_ring.ringhd.next = sun4c_kfree_ring.ringhd.prev =
607                 &sun4c_kfree_ring.ringhd;
608         sun4c_kfree_ring.num_entries = 0;
609 }
610
611 static void add_ring(struct sun4c_mmu_ring *ring,
612                      struct sun4c_mmu_entry *entry)
613 {
614         struct sun4c_mmu_entry *head = &ring->ringhd;
615
616         entry->prev = head;
617         (entry->next = head->next)->prev = entry;
618         head->next = entry;
619         ring->num_entries++;
620 }
621
622 static inline void add_lru(struct sun4c_mmu_entry *entry)
623 {
624         struct sun4c_mmu_ring *ring = &sun4c_ulru_ring;
625         struct sun4c_mmu_entry *head = &ring->ringhd;
626
627         entry->lru_next = head;
628         (entry->lru_prev = head->lru_prev)->lru_next = entry;
629         head->lru_prev = entry;
630 }
631
632 static void add_ring_ordered(struct sun4c_mmu_ring *ring,
633                              struct sun4c_mmu_entry *entry)
634 {
635         struct sun4c_mmu_entry *head = &ring->ringhd;
636         unsigned long addr = entry->vaddr;
637
638         while ((head->next != &ring->ringhd) && (head->next->vaddr < addr))
639                 head = head->next;
640
641         entry->prev = head;
642         (entry->next = head->next)->prev = entry;
643         head->next = entry;
644         ring->num_entries++;
645
646         add_lru(entry);
647 }
648
649 static inline void remove_ring(struct sun4c_mmu_ring *ring,
650                                    struct sun4c_mmu_entry *entry)
651 {
652         struct sun4c_mmu_entry *next = entry->next;
653
654         (next->prev = entry->prev)->next = next;
655         ring->num_entries--;
656 }
657
658 static void remove_lru(struct sun4c_mmu_entry *entry)
659 {
660         struct sun4c_mmu_entry *next = entry->lru_next;
661
662         (next->lru_prev = entry->lru_prev)->lru_next = next;
663 }
664
665 static void free_user_entry(int ctx, struct sun4c_mmu_entry *entry)
666 {
667         remove_ring(sun4c_context_ring+ctx, entry);
668         remove_lru(entry);
669         add_ring(&sun4c_ufree_ring, entry);
670 }
671
672 static void free_kernel_entry(struct sun4c_mmu_entry *entry,
673                               struct sun4c_mmu_ring *ring)
674 {
675         remove_ring(ring, entry);
676         add_ring(&sun4c_kfree_ring, entry);
677 }
678
679 static void __init sun4c_init_fill_kernel_ring(int howmany)
680 {
681         int i;
682
683         while (howmany) {
684                 for (i = 0; i < invalid_segment; i++)
685                         if (!mmu_entry_pool[i].locked)
686                                 break;
687                 mmu_entry_pool[i].locked = 1;
688                 sun4c_init_clean_segmap(i);
689                 add_ring(&sun4c_kfree_ring, &mmu_entry_pool[i]);
690                 howmany--;
691         }
692 }
693
694 static void __init sun4c_init_fill_user_ring(void)
695 {
696         int i;
697
698         for (i = 0; i < invalid_segment; i++) {
699                 if (mmu_entry_pool[i].locked)
700                         continue;
701                 sun4c_init_clean_segmap(i);
702                 add_ring(&sun4c_ufree_ring, &mmu_entry_pool[i]);
703         }
704 }
705
706 static void sun4c_kernel_unmap(struct sun4c_mmu_entry *kentry)
707 {
708         int savectx, ctx;
709
710         savectx = sun4c_get_context();
711         for (ctx = 0; ctx < num_contexts; ctx++) {
712                 sun4c_set_context(ctx);
713                 sun4c_put_segmap(kentry->vaddr, invalid_segment);
714         }
715         sun4c_set_context(savectx);
716 }
717
718 static void sun4c_kernel_map(struct sun4c_mmu_entry *kentry)
719 {
720         int savectx, ctx;
721
722         savectx = sun4c_get_context();
723         for (ctx = 0; ctx < num_contexts; ctx++) {
724                 sun4c_set_context(ctx);
725                 sun4c_put_segmap(kentry->vaddr, kentry->pseg);
726         }
727         sun4c_set_context(savectx);
728 }
729
730 #define sun4c_user_unmap(__entry) \
731         sun4c_put_segmap((__entry)->vaddr, invalid_segment)
732
733 static void sun4c_demap_context(struct sun4c_mmu_ring *crp, unsigned char ctx)
734 {
735         struct sun4c_mmu_entry *head = &crp->ringhd;
736         unsigned long flags;
737
738         local_irq_save(flags);
739         if (head->next != head) {
740                 struct sun4c_mmu_entry *entry = head->next;
741                 int savectx = sun4c_get_context();
742
743                 flush_user_windows();
744                 sun4c_set_context(ctx);
745                 sun4c_flush_context();
746                 do {
747                         struct sun4c_mmu_entry *next = entry->next;
748
749                         sun4c_user_unmap(entry);
750                         free_user_entry(ctx, entry);
751
752                         entry = next;
753                 } while (entry != head);
754                 sun4c_set_context(savectx);
755         }
756         local_irq_restore(flags);
757 }
758
759 static int sun4c_user_taken_entries;  /* This is how much we have.             */
760 static int max_user_taken_entries;    /* This limits us and prevents deadlock. */
761
762 static struct sun4c_mmu_entry *sun4c_kernel_strategy(void)
763 {
764         struct sun4c_mmu_entry *this_entry;
765
766         /* If some are free, return first one. */
767         if (sun4c_kfree_ring.num_entries) {
768                 this_entry = sun4c_kfree_ring.ringhd.next;
769                 return this_entry;
770         }
771
772         /* Else free one up. */
773         this_entry = sun4c_kernel_ring.ringhd.prev;
774         sun4c_flush_segment(this_entry->vaddr);
775         sun4c_kernel_unmap(this_entry);
776         free_kernel_entry(this_entry, &sun4c_kernel_ring);
777         this_entry = sun4c_kfree_ring.ringhd.next;
778
779         return this_entry;
780 }
781
782 /* Using this method to free up mmu entries eliminates a lot of
783  * potential races since we have a kernel that incurs tlb
784  * replacement faults.  There may be performance penalties.
785  *
786  * NOTE: Must be called with interrupts disabled.
787  */
788 static struct sun4c_mmu_entry *sun4c_user_strategy(void)
789 {
790         struct sun4c_mmu_entry *entry;
791         unsigned char ctx;
792         int savectx;
793
794         /* If some are free, return first one. */
795         if (sun4c_ufree_ring.num_entries) {
796                 entry = sun4c_ufree_ring.ringhd.next;
797                 goto unlink_out;
798         }
799
800         if (sun4c_user_taken_entries) {
801                 entry = sun4c_kernel_strategy();
802                 sun4c_user_taken_entries--;
803                 goto kunlink_out;
804         }
805
806         /* Grab from the beginning of the LRU list. */
807         entry = sun4c_ulru_ring.ringhd.lru_next;
808         ctx = entry->ctx;
809
810         savectx = sun4c_get_context();
811         flush_user_windows();
812         sun4c_set_context(ctx);
813         sun4c_flush_segment(entry->vaddr);
814         sun4c_user_unmap(entry);
815         remove_ring(sun4c_context_ring + ctx, entry);
816         remove_lru(entry);
817         sun4c_set_context(savectx);
818
819         return entry;
820
821 unlink_out:
822         remove_ring(&sun4c_ufree_ring, entry);
823         return entry;
824 kunlink_out:
825         remove_ring(&sun4c_kfree_ring, entry);
826         return entry;
827 }
828
829 /* NOTE: Must be called with interrupts disabled. */
830 void sun4c_grow_kernel_ring(void)
831 {
832         struct sun4c_mmu_entry *entry;
833
834         /* Prevent deadlock condition. */
835         if (sun4c_user_taken_entries >= max_user_taken_entries)
836                 return;
837
838         if (sun4c_ufree_ring.num_entries) {
839                 entry = sun4c_ufree_ring.ringhd.next;
840                 remove_ring(&sun4c_ufree_ring, entry);
841                 add_ring(&sun4c_kfree_ring, entry);
842                 sun4c_user_taken_entries++;
843         }
844 }
845
846 /* 2 page buckets for task struct and kernel stack allocation.
847  *
848  * TASK_STACK_BEGIN
849  * bucket[0]
850  * bucket[1]
851  *   [ ... ]
852  * bucket[NR_TASK_BUCKETS-1]
853  * TASK_STACK_BEGIN + (sizeof(struct task_bucket) * NR_TASK_BUCKETS)
854  *
855  * Each slot looks like:
856  *
857  *  page 1 --  task struct + beginning of kernel stack
858  *  page 2 --  rest of kernel stack
859  */
860
861 union task_union *sun4c_bucket[NR_TASK_BUCKETS];
862
863 static int sun4c_lowbucket_avail;
864
865 #define BUCKET_EMPTY     ((union task_union *) 0)
866 #define BUCKET_SHIFT     (PAGE_SHIFT + 1)        /* log2(sizeof(struct task_bucket)) */
867 #define BUCKET_SIZE      (1 << BUCKET_SHIFT)
868 #define BUCKET_NUM(addr) ((((addr) - SUN4C_LOCK_VADDR) >> BUCKET_SHIFT))
869 #define BUCKET_ADDR(num) (((num) << BUCKET_SHIFT) + SUN4C_LOCK_VADDR)
870 #define BUCKET_PTE(page)       \
871         ((((page) - PAGE_OFFSET) >> PAGE_SHIFT) | pgprot_val(SUN4C_PAGE_KERNEL))
872 #define BUCKET_PTE_PAGE(pte)   \
873         (PAGE_OFFSET + (((pte) & SUN4C_PFN_MASK) << PAGE_SHIFT))
874
875 static void get_locked_segment(unsigned long addr)
876 {
877         struct sun4c_mmu_entry *stolen;
878         unsigned long flags;
879
880         local_irq_save(flags);
881         addr &= SUN4C_REAL_PGDIR_MASK;
882         stolen = sun4c_user_strategy();
883         max_user_taken_entries--;
884         stolen->vaddr = addr;
885         flush_user_windows();
886         sun4c_kernel_map(stolen);
887         local_irq_restore(flags);
888 }
889
890 static void free_locked_segment(unsigned long addr)
891 {
892         struct sun4c_mmu_entry *entry;
893         unsigned long flags;
894         unsigned char pseg;
895
896         local_irq_save(flags);
897         addr &= SUN4C_REAL_PGDIR_MASK;
898         pseg = sun4c_get_segmap(addr);
899         entry = &mmu_entry_pool[pseg];
900
901         flush_user_windows();
902         sun4c_flush_segment(addr);
903         sun4c_kernel_unmap(entry);
904         add_ring(&sun4c_ufree_ring, entry);
905         max_user_taken_entries++;
906         local_irq_restore(flags);
907 }
908
909 static inline void garbage_collect(int entry)
910 {
911         int start, end;
912
913         /* 32 buckets per segment... */
914         entry &= ~31;
915         start = entry;
916         for (end = (start + 32); start < end; start++)
917                 if (sun4c_bucket[start] != BUCKET_EMPTY)
918                         return;
919
920         /* Entire segment empty, release it. */
921         free_locked_segment(BUCKET_ADDR(entry));
922 }
923
924 static struct thread_info *sun4c_alloc_thread_info(void)
925 {
926         unsigned long addr, pages;
927         int entry;
928
929         pages = __get_free_pages(GFP_KERNEL, THREAD_INFO_ORDER);
930         if (!pages)
931                 return NULL;
932
933         for (entry = sun4c_lowbucket_avail; entry < NR_TASK_BUCKETS; entry++)
934                 if (sun4c_bucket[entry] == BUCKET_EMPTY)
935                         break;
936         if (entry == NR_TASK_BUCKETS) {
937                 free_pages(pages, THREAD_INFO_ORDER);
938                 return NULL;
939         }
940         if (entry >= sun4c_lowbucket_avail)
941                 sun4c_lowbucket_avail = entry + 1;
942
943         addr = BUCKET_ADDR(entry);
944         sun4c_bucket[entry] = (union task_union *) addr;
945         if(sun4c_get_segmap(addr) == invalid_segment)
946                 get_locked_segment(addr);
947
948         /* We are changing the virtual color of the page(s)
949          * so we must flush the cache to guarantee consistency.
950          */
951         sun4c_flush_page(pages);
952         sun4c_flush_page(pages + PAGE_SIZE);
953
954         sun4c_put_pte(addr, BUCKET_PTE(pages));
955         sun4c_put_pte(addr + PAGE_SIZE, BUCKET_PTE(pages + PAGE_SIZE));
956
957 #ifdef CONFIG_DEBUG_STACK_USAGE
958         memset((void *)addr, 0, PAGE_SIZE << THREAD_INFO_ORDER);
959 #endif /* DEBUG_STACK_USAGE */
960
961         return (struct thread_info *) addr;
962 }
963
964 static void sun4c_free_thread_info(struct thread_info *ti)
965 {
966         unsigned long tiaddr = (unsigned long) ti;
967         unsigned long pages = BUCKET_PTE_PAGE(sun4c_get_pte(tiaddr));
968         int entry = BUCKET_NUM(tiaddr);
969
970         /* We are deleting a mapping, so the flush here is mandatory. */
971         sun4c_flush_page(tiaddr);
972         sun4c_flush_page(tiaddr + PAGE_SIZE);
973
974         sun4c_put_pte(tiaddr, 0);
975         sun4c_put_pte(tiaddr + PAGE_SIZE, 0);
976
977         sun4c_bucket[entry] = BUCKET_EMPTY;
978         if (entry < sun4c_lowbucket_avail)
979                 sun4c_lowbucket_avail = entry;
980
981         free_pages(pages, THREAD_INFO_ORDER);
982         garbage_collect(entry);
983 }
984
985 static void __init sun4c_init_buckets(void)
986 {
987         int entry;
988
989         if (sizeof(union thread_union) != (PAGE_SIZE << THREAD_INFO_ORDER)) {
990                 extern void thread_info_size_is_bolixed_pete(void);
991                 thread_info_size_is_bolixed_pete();
992         }
993
994         for (entry = 0; entry < NR_TASK_BUCKETS; entry++)
995                 sun4c_bucket[entry] = BUCKET_EMPTY;
996         sun4c_lowbucket_avail = 0;
997 }
998
999 static unsigned long sun4c_iobuffer_start;
1000 static unsigned long sun4c_iobuffer_end;
1001 static unsigned long sun4c_iobuffer_high;
1002 static unsigned long *sun4c_iobuffer_map;
1003 static int iobuffer_map_size;
1004
1005 /*
1006  * Alias our pages so they do not cause a trap.
1007  * Also one page may be aliased into several I/O areas and we may
1008  * finish these I/O separately.
1009  */
1010 static char *sun4c_lockarea(char *vaddr, unsigned long size)
1011 {
1012         unsigned long base, scan;
1013         unsigned long npages;
1014         unsigned long vpage;
1015         unsigned long pte;
1016         unsigned long apage;
1017         unsigned long high;
1018         unsigned long flags;
1019
1020         npages = (((unsigned long)vaddr & ~PAGE_MASK) +
1021                   size + (PAGE_SIZE-1)) >> PAGE_SHIFT;
1022
1023         scan = 0;
1024         local_irq_save(flags);
1025         for (;;) {
1026                 scan = find_next_zero_bit(sun4c_iobuffer_map,
1027                                           iobuffer_map_size, scan);
1028                 if ((base = scan) + npages > iobuffer_map_size) goto abend;
1029                 for (;;) {
1030                         if (scan >= base + npages) goto found;
1031                         if (test_bit(scan, sun4c_iobuffer_map)) break;
1032                         scan++;
1033                 }
1034         }
1035
1036 found:
1037         high = ((base + npages) << PAGE_SHIFT) + sun4c_iobuffer_start;
1038         high = SUN4C_REAL_PGDIR_ALIGN(high);
1039         while (high > sun4c_iobuffer_high) {
1040                 get_locked_segment(sun4c_iobuffer_high);
1041                 sun4c_iobuffer_high += SUN4C_REAL_PGDIR_SIZE;
1042         }
1043
1044         vpage = ((unsigned long) vaddr) & PAGE_MASK;
1045         for (scan = base; scan < base+npages; scan++) {
1046                 pte = ((vpage-PAGE_OFFSET) >> PAGE_SHIFT);
1047                 pte |= pgprot_val(SUN4C_PAGE_KERNEL);
1048                 pte |= _SUN4C_PAGE_NOCACHE;
1049                 set_bit(scan, sun4c_iobuffer_map);
1050                 apage = (scan << PAGE_SHIFT) + sun4c_iobuffer_start;
1051
1052                 /* Flush original mapping so we see the right things later. */
1053                 sun4c_flush_page(vpage);
1054
1055                 sun4c_put_pte(apage, pte);
1056                 vpage += PAGE_SIZE;
1057         }
1058         local_irq_restore(flags);
1059         return (char *) ((base << PAGE_SHIFT) + sun4c_iobuffer_start +
1060                          (((unsigned long) vaddr) & ~PAGE_MASK));
1061
1062 abend:
1063         local_irq_restore(flags);
1064         printk("DMA vaddr=0x%p size=%08lx\n", vaddr, size);
1065         panic("Out of iobuffer table");
1066         return NULL;
1067 }
1068
1069 static void sun4c_unlockarea(char *vaddr, unsigned long size)
1070 {
1071         unsigned long vpage, npages;
1072         unsigned long flags;
1073         int scan, high;
1074
1075         vpage = (unsigned long)vaddr & PAGE_MASK;
1076         npages = (((unsigned long)vaddr & ~PAGE_MASK) +
1077                   size + (PAGE_SIZE-1)) >> PAGE_SHIFT;
1078
1079         local_irq_save(flags);
1080         while (npages != 0) {
1081                 --npages;
1082
1083                 /* This mapping is marked non-cachable, no flush necessary. */
1084                 sun4c_put_pte(vpage, 0);
1085                 clear_bit((vpage - sun4c_iobuffer_start) >> PAGE_SHIFT,
1086                           sun4c_iobuffer_map);
1087                 vpage += PAGE_SIZE;
1088         }
1089
1090         /* garbage collect */
1091         scan = (sun4c_iobuffer_high - sun4c_iobuffer_start) >> PAGE_SHIFT;
1092         while (scan >= 0 && !sun4c_iobuffer_map[scan >> 5])
1093                 scan -= 32;
1094         scan += 32;
1095         high = sun4c_iobuffer_start + (scan << PAGE_SHIFT);
1096         high = SUN4C_REAL_PGDIR_ALIGN(high) + SUN4C_REAL_PGDIR_SIZE;
1097         while (high < sun4c_iobuffer_high) {
1098                 sun4c_iobuffer_high -= SUN4C_REAL_PGDIR_SIZE;
1099                 free_locked_segment(sun4c_iobuffer_high);
1100         }
1101         local_irq_restore(flags);
1102 }
1103
1104 /* Note the scsi code at init time passes to here buffers
1105  * which sit on the kernel stack, those are already locked
1106  * by implication and fool the page locking code above
1107  * if passed to by mistake.
1108  */
1109 static __u32 sun4c_get_scsi_one(struct device *dev, char *bufptr, unsigned long len)
1110 {
1111         unsigned long page;
1112
1113         page = ((unsigned long)bufptr) & PAGE_MASK;
1114         if (!virt_addr_valid(page)) {
1115                 sun4c_flush_page(page);
1116                 return (__u32)bufptr; /* already locked */
1117         }
1118         return (__u32)sun4c_lockarea(bufptr, len);
1119 }
1120
1121 static void sun4c_get_scsi_sgl(struct device *dev, struct scatterlist *sg, int sz)
1122 {
1123         while (sz != 0) {
1124                 --sz;
1125                 sg->dvma_address = (__u32)sun4c_lockarea(sg_virt(sg), sg->length);
1126                 sg->dvma_length = sg->length;
1127                 sg = sg_next(sg);
1128         }
1129 }
1130
1131 static void sun4c_release_scsi_one(struct device *dev, __u32 bufptr, unsigned long len)
1132 {
1133         if (bufptr < sun4c_iobuffer_start)
1134                 return; /* On kernel stack or similar, see above */
1135         sun4c_unlockarea((char *)bufptr, len);
1136 }
1137
1138 static void sun4c_release_scsi_sgl(struct device *dev, struct scatterlist *sg, int sz)
1139 {
1140         while (sz != 0) {
1141                 --sz;
1142                 sun4c_unlockarea((char *)sg->dvma_address, sg->length);
1143                 sg = sg_next(sg);
1144         }
1145 }
1146
1147 #define TASK_ENTRY_SIZE    BUCKET_SIZE /* see above */
1148 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1149
1150 struct vm_area_struct sun4c_kstack_vma;
1151
1152 static void __init sun4c_init_lock_areas(void)
1153 {
1154         unsigned long sun4c_taskstack_start;
1155         unsigned long sun4c_taskstack_end;
1156         int bitmap_size;
1157
1158         sun4c_init_buckets();
1159         sun4c_taskstack_start = SUN4C_LOCK_VADDR;
1160         sun4c_taskstack_end = (sun4c_taskstack_start +
1161                                (TASK_ENTRY_SIZE * NR_TASK_BUCKETS));
1162         if (sun4c_taskstack_end >= SUN4C_LOCK_END) {
1163                 prom_printf("Too many tasks, decrease NR_TASK_BUCKETS please.\n");
1164                 prom_halt();
1165         }
1166
1167         sun4c_iobuffer_start = sun4c_iobuffer_high =
1168                                 SUN4C_REAL_PGDIR_ALIGN(sun4c_taskstack_end);
1169         sun4c_iobuffer_end = SUN4C_LOCK_END;
1170         bitmap_size = (sun4c_iobuffer_end - sun4c_iobuffer_start) >> PAGE_SHIFT;
1171         bitmap_size = (bitmap_size + 7) >> 3;
1172         bitmap_size = LONG_ALIGN(bitmap_size);
1173         iobuffer_map_size = bitmap_size << 3;
1174         sun4c_iobuffer_map = __alloc_bootmem(bitmap_size, SMP_CACHE_BYTES, 0UL);
1175         memset((void *) sun4c_iobuffer_map, 0, bitmap_size);
1176
1177         sun4c_kstack_vma.vm_mm = &init_mm;
1178         sun4c_kstack_vma.vm_start = sun4c_taskstack_start;
1179         sun4c_kstack_vma.vm_end = sun4c_taskstack_end;
1180         sun4c_kstack_vma.vm_page_prot = PAGE_SHARED;
1181         sun4c_kstack_vma.vm_flags = VM_READ | VM_WRITE | VM_EXEC;
1182         insert_vm_struct(&init_mm, &sun4c_kstack_vma);
1183 }
1184
1185 /* Cache flushing on the sun4c. */
1186 static void sun4c_flush_cache_all(void)
1187 {
1188         unsigned long begin, end;
1189
1190         flush_user_windows();
1191         begin = (KERNBASE + SUN4C_REAL_PGDIR_SIZE);
1192         end = (begin + SUN4C_VAC_SIZE);
1193
1194         if (sun4c_vacinfo.linesize == 32) {
1195                 while (begin < end) {
1196                         __asm__ __volatile__(
1197                         "ld     [%0 + 0x00], %%g0\n\t"
1198                         "ld     [%0 + 0x20], %%g0\n\t"
1199                         "ld     [%0 + 0x40], %%g0\n\t"
1200                         "ld     [%0 + 0x60], %%g0\n\t"
1201                         "ld     [%0 + 0x80], %%g0\n\t"
1202                         "ld     [%0 + 0xa0], %%g0\n\t"
1203                         "ld     [%0 + 0xc0], %%g0\n\t"
1204                         "ld     [%0 + 0xe0], %%g0\n\t"
1205                         "ld     [%0 + 0x100], %%g0\n\t"
1206                         "ld     [%0 + 0x120], %%g0\n\t"
1207                         "ld     [%0 + 0x140], %%g0\n\t"
1208                         "ld     [%0 + 0x160], %%g0\n\t"
1209                         "ld     [%0 + 0x180], %%g0\n\t"
1210                         "ld     [%0 + 0x1a0], %%g0\n\t"
1211                         "ld     [%0 + 0x1c0], %%g0\n\t"
1212                         "ld     [%0 + 0x1e0], %%g0\n"
1213                         : : "r" (begin));
1214                         begin += 512;
1215                 }
1216         } else {
1217                 while (begin < end) {
1218                         __asm__ __volatile__(
1219                         "ld     [%0 + 0x00], %%g0\n\t"
1220                         "ld     [%0 + 0x10], %%g0\n\t"
1221                         "ld     [%0 + 0x20], %%g0\n\t"
1222                         "ld     [%0 + 0x30], %%g0\n\t"
1223                         "ld     [%0 + 0x40], %%g0\n\t"
1224                         "ld     [%0 + 0x50], %%g0\n\t"
1225                         "ld     [%0 + 0x60], %%g0\n\t"
1226                         "ld     [%0 + 0x70], %%g0\n\t"
1227                         "ld     [%0 + 0x80], %%g0\n\t"
1228                         "ld     [%0 + 0x90], %%g0\n\t"
1229                         "ld     [%0 + 0xa0], %%g0\n\t"
1230                         "ld     [%0 + 0xb0], %%g0\n\t"
1231                         "ld     [%0 + 0xc0], %%g0\n\t"
1232                         "ld     [%0 + 0xd0], %%g0\n\t"
1233                         "ld     [%0 + 0xe0], %%g0\n\t"
1234                         "ld     [%0 + 0xf0], %%g0\n"
1235                         : : "r" (begin));
1236                         begin += 256;
1237                 }
1238         }
1239 }
1240
1241 static void sun4c_flush_cache_mm(struct mm_struct *mm)
1242 {
1243         int new_ctx = mm->context;
1244
1245         if (new_ctx != NO_CONTEXT) {
1246                 flush_user_windows();
1247
1248                 if (sun4c_context_ring[new_ctx].num_entries) {
1249                         struct sun4c_mmu_entry *head = &sun4c_context_ring[new_ctx].ringhd;
1250                         unsigned long flags;
1251
1252                         local_irq_save(flags);
1253                         if (head->next != head) {
1254                                 struct sun4c_mmu_entry *entry = head->next;
1255                                 int savectx = sun4c_get_context();
1256
1257                                 sun4c_set_context(new_ctx);
1258                                 sun4c_flush_context();
1259                                 do {
1260                                         struct sun4c_mmu_entry *next = entry->next;
1261
1262                                         sun4c_user_unmap(entry);
1263                                         free_user_entry(new_ctx, entry);
1264
1265                                         entry = next;
1266                                 } while (entry != head);
1267                                 sun4c_set_context(savectx);
1268                         }
1269                         local_irq_restore(flags);
1270                 }
1271         }
1272 }
1273
1274 static void sun4c_flush_cache_range(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1275 {
1276         struct mm_struct *mm = vma->vm_mm;
1277         int new_ctx = mm->context;
1278
1279         if (new_ctx != NO_CONTEXT) {
1280                 struct sun4c_mmu_entry *head = &sun4c_context_ring[new_ctx].ringhd;
1281                 struct sun4c_mmu_entry *entry;
1282                 unsigned long flags;
1283
1284                 flush_user_windows();
1285
1286                 local_irq_save(flags);
1287                 /* All user segmap chains are ordered on entry->vaddr. */
1288                 for (entry = head->next;
1289                      (entry != head) && ((entry->vaddr+SUN4C_REAL_PGDIR_SIZE) < start);
1290                      entry = entry->next)
1291                         ;
1292
1293                 /* Tracing various job mixtures showed that this conditional
1294                  * only passes ~35% of the time for most worse case situations,
1295                  * therefore we avoid all of this gross overhead ~65% of the time.
1296                  */
1297                 if ((entry != head) && (entry->vaddr < end)) {
1298                         int octx = sun4c_get_context();
1299                         sun4c_set_context(new_ctx);
1300
1301                         /* At this point, always, (start >= entry->vaddr) and
1302                          * (entry->vaddr < end), once the latter condition
1303                          * ceases to hold, or we hit the end of the list, we
1304                          * exit the loop.  The ordering of all user allocated
1305                          * segmaps makes this all work out so beautifully.
1306                          */
1307                         do {
1308                                 struct sun4c_mmu_entry *next = entry->next;
1309                                 unsigned long realend;
1310
1311                                 /* "realstart" is always >= entry->vaddr */
1312                                 realend = entry->vaddr + SUN4C_REAL_PGDIR_SIZE;
1313                                 if (end < realend)
1314                                         realend = end;
1315                                 if ((realend - entry->vaddr) <= (PAGE_SIZE << 3)) {
1316                                         unsigned long page = entry->vaddr;
1317                                         while (page < realend) {
1318                                                 sun4c_flush_page(page);
1319                                                 page += PAGE_SIZE;
1320                                         }
1321                                 } else {
1322                                         sun4c_flush_segment(entry->vaddr);
1323                                         sun4c_user_unmap(entry);
1324                                         free_user_entry(new_ctx, entry);
1325                                 }
1326                                 entry = next;
1327                         } while ((entry != head) && (entry->vaddr < end));
1328                         sun4c_set_context(octx);
1329                 }
1330                 local_irq_restore(flags);
1331         }
1332 }
1333
1334 static void sun4c_flush_cache_page(struct vm_area_struct *vma, unsigned long page)
1335 {
1336         struct mm_struct *mm = vma->vm_mm;
1337         int new_ctx = mm->context;
1338
1339         /* Sun4c has no separate I/D caches so cannot optimize for non
1340          * text page flushes.
1341          */
1342         if (new_ctx != NO_CONTEXT) {
1343                 int octx = sun4c_get_context();
1344                 unsigned long flags;
1345
1346                 flush_user_windows();
1347                 local_irq_save(flags);
1348                 sun4c_set_context(new_ctx);
1349                 sun4c_flush_page(page);
1350                 sun4c_set_context(octx);
1351                 local_irq_restore(flags);
1352         }
1353 }
1354
1355 static void sun4c_flush_page_to_ram(unsigned long page)
1356 {
1357         unsigned long flags;
1358
1359         local_irq_save(flags);
1360         sun4c_flush_page(page);
1361         local_irq_restore(flags);
1362 }
1363
1364 /* Sun4c cache is unified, both instructions and data live there, so
1365  * no need to flush the on-stack instructions for new signal handlers.
1366  */
1367 static void sun4c_flush_sig_insns(struct mm_struct *mm, unsigned long insn_addr)
1368 {
1369 }
1370
1371 /* TLB flushing on the sun4c.  These routines count on the cache
1372  * flushing code to flush the user register windows so that we need
1373  * not do so when we get here.
1374  */
1375
1376 static void sun4c_flush_tlb_all(void)
1377 {
1378         struct sun4c_mmu_entry *this_entry, *next_entry;
1379         unsigned long flags;
1380         int savectx, ctx;
1381
1382         local_irq_save(flags);
1383         this_entry = sun4c_kernel_ring.ringhd.next;
1384         savectx = sun4c_get_context();
1385         flush_user_windows();
1386         while (sun4c_kernel_ring.num_entries) {
1387                 next_entry = this_entry->next;
1388                 sun4c_flush_segment(this_entry->vaddr);
1389                 for (ctx = 0; ctx < num_contexts; ctx++) {
1390                         sun4c_set_context(ctx);
1391                         sun4c_put_segmap(this_entry->vaddr, invalid_segment);
1392                 }
1393                 free_kernel_entry(this_entry, &sun4c_kernel_ring);
1394                 this_entry = next_entry;
1395         }
1396         sun4c_set_context(savectx);
1397         local_irq_restore(flags);
1398 }
1399
1400 static void sun4c_flush_tlb_mm(struct mm_struct *mm)
1401 {
1402         int new_ctx = mm->context;
1403
1404         if (new_ctx != NO_CONTEXT) {
1405                 struct sun4c_mmu_entry *head = &sun4c_context_ring[new_ctx].ringhd;
1406                 unsigned long flags;
1407
1408                 local_irq_save(flags);
1409                 if (head->next != head) {
1410                         struct sun4c_mmu_entry *entry = head->next;
1411                         int savectx = sun4c_get_context();
1412
1413                         sun4c_set_context(new_ctx);
1414                         sun4c_flush_context();
1415                         do {
1416                                 struct sun4c_mmu_entry *next = entry->next;
1417
1418                                 sun4c_user_unmap(entry);
1419                                 free_user_entry(new_ctx, entry);
1420
1421                                 entry = next;
1422                         } while (entry != head);
1423                         sun4c_set_context(savectx);
1424                 }
1425                 local_irq_restore(flags);
1426         }
1427 }
1428
1429 static void sun4c_flush_tlb_range(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1430 {
1431         struct mm_struct *mm = vma->vm_mm;
1432         int new_ctx = mm->context;
1433
1434         if (new_ctx != NO_CONTEXT) {
1435                 struct sun4c_mmu_entry *head = &sun4c_context_ring[new_ctx].ringhd;
1436                 struct sun4c_mmu_entry *entry;
1437                 unsigned long flags;
1438
1439                 local_irq_save(flags);
1440                 /* See commentary in sun4c_flush_cache_range(). */
1441                 for (entry = head->next;
1442                      (entry != head) && ((entry->vaddr+SUN4C_REAL_PGDIR_SIZE) < start);
1443                      entry = entry->next)
1444                         ;
1445
1446                 if ((entry != head) && (entry->vaddr < end)) {
1447                         int octx = sun4c_get_context();
1448
1449                         sun4c_set_context(new_ctx);
1450                         do {
1451                                 struct sun4c_mmu_entry *next = entry->next;
1452
1453                                 sun4c_flush_segment(entry->vaddr);
1454                                 sun4c_user_unmap(entry);
1455                                 free_user_entry(new_ctx, entry);
1456
1457                                 entry = next;
1458                         } while ((entry != head) && (entry->vaddr < end));
1459                         sun4c_set_context(octx);
1460                 }
1461                 local_irq_restore(flags);
1462         }
1463 }
1464
1465 static void sun4c_flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
1466 {
1467         struct mm_struct *mm = vma->vm_mm;
1468         int new_ctx = mm->context;
1469
1470         if (new_ctx != NO_CONTEXT) {
1471                 int savectx = sun4c_get_context();
1472                 unsigned long flags;
1473
1474                 local_irq_save(flags);
1475                 sun4c_set_context(new_ctx);
1476                 page &= PAGE_MASK;
1477                 sun4c_flush_page(page);
1478                 sun4c_put_pte(page, 0);
1479                 sun4c_set_context(savectx);
1480                 local_irq_restore(flags);
1481         }
1482 }
1483
1484 static inline void sun4c_mapioaddr(unsigned long physaddr, unsigned long virt_addr)
1485 {
1486         unsigned long page_entry, pg_iobits;
1487
1488         pg_iobits = _SUN4C_PAGE_PRESENT | _SUN4C_READABLE | _SUN4C_WRITEABLE |
1489                     _SUN4C_PAGE_IO | _SUN4C_PAGE_NOCACHE;
1490
1491         page_entry = ((physaddr >> PAGE_SHIFT) & SUN4C_PFN_MASK);
1492         page_entry |= ((pg_iobits | _SUN4C_PAGE_PRIV) & ~(_SUN4C_PAGE_PRESENT));
1493         sun4c_put_pte(virt_addr, page_entry);
1494 }
1495
1496 static void sun4c_mapiorange(unsigned int bus, unsigned long xpa,
1497     unsigned long xva, unsigned int len)
1498 {
1499         while (len != 0) {
1500                 len -= PAGE_SIZE;
1501                 sun4c_mapioaddr(xpa, xva);
1502                 xva += PAGE_SIZE;
1503                 xpa += PAGE_SIZE;
1504         }
1505 }
1506
1507 static void sun4c_unmapiorange(unsigned long virt_addr, unsigned int len)
1508 {
1509         while (len != 0) {
1510                 len -= PAGE_SIZE;
1511                 sun4c_put_pte(virt_addr, 0);
1512                 virt_addr += PAGE_SIZE;
1513         }
1514 }
1515
1516 static void sun4c_alloc_context(struct mm_struct *old_mm, struct mm_struct *mm)
1517 {
1518         struct ctx_list *ctxp;
1519
1520         ctxp = ctx_free.next;
1521         if (ctxp != &ctx_free) {
1522                 remove_from_ctx_list(ctxp);
1523                 add_to_used_ctxlist(ctxp);
1524                 mm->context = ctxp->ctx_number;
1525                 ctxp->ctx_mm = mm;
1526                 return;
1527         }
1528         ctxp = ctx_used.next;
1529         if (ctxp->ctx_mm == old_mm)
1530                 ctxp = ctxp->next;
1531         remove_from_ctx_list(ctxp);
1532         add_to_used_ctxlist(ctxp);
1533         ctxp->ctx_mm->context = NO_CONTEXT;
1534         ctxp->ctx_mm = mm;
1535         mm->context = ctxp->ctx_number;
1536         sun4c_demap_context(&sun4c_context_ring[ctxp->ctx_number],
1537                                ctxp->ctx_number);
1538 }
1539
1540 /* Switch the current MM context. */
1541 static void sun4c_switch_mm(struct mm_struct *old_mm, struct mm_struct *mm, struct task_struct *tsk, int cpu)
1542 {
1543         struct ctx_list *ctx;
1544         int dirty = 0;
1545
1546         if (mm->context == NO_CONTEXT) {
1547                 dirty = 1;
1548                 sun4c_alloc_context(old_mm, mm);
1549         } else {
1550                 /* Update the LRU ring of contexts. */
1551                 ctx = ctx_list_pool + mm->context;
1552                 remove_from_ctx_list(ctx);
1553                 add_to_used_ctxlist(ctx);
1554         }
1555         if (dirty || old_mm != mm)
1556                 sun4c_set_context(mm->context);
1557 }
1558
1559 static void sun4c_destroy_context(struct mm_struct *mm)
1560 {
1561         struct ctx_list *ctx_old;
1562
1563         if (mm->context != NO_CONTEXT) {
1564                 sun4c_demap_context(&sun4c_context_ring[mm->context], mm->context);
1565                 ctx_old = ctx_list_pool + mm->context;
1566                 remove_from_ctx_list(ctx_old);
1567                 add_to_free_ctxlist(ctx_old);
1568                 mm->context = NO_CONTEXT;
1569         }
1570 }
1571
1572 static void sun4c_mmu_info(struct seq_file *m)
1573 {
1574         int used_user_entries, i;
1575
1576         used_user_entries = 0;
1577         for (i = 0; i < num_contexts; i++)
1578                 used_user_entries += sun4c_context_ring[i].num_entries;
1579
1580         seq_printf(m, 
1581                    "vacsize\t\t: %d bytes\n"
1582                    "vachwflush\t: %s\n"
1583                    "vaclinesize\t: %d bytes\n"
1584                    "mmuctxs\t\t: %d\n"
1585                    "mmupsegs\t: %d\n"
1586                    "kernelpsegs\t: %d\n"
1587                    "kfreepsegs\t: %d\n"
1588                    "usedpsegs\t: %d\n"
1589                    "ufreepsegs\t: %d\n"
1590                    "user_taken\t: %d\n"
1591                    "max_taken\t: %d\n",
1592                    sun4c_vacinfo.num_bytes,
1593                    (sun4c_vacinfo.do_hwflushes ? "yes" : "no"),
1594                    sun4c_vacinfo.linesize,
1595                    num_contexts,
1596                    (invalid_segment + 1),
1597                    sun4c_kernel_ring.num_entries,
1598                    sun4c_kfree_ring.num_entries,
1599                    used_user_entries,
1600                    sun4c_ufree_ring.num_entries,
1601                    sun4c_user_taken_entries,
1602                    max_user_taken_entries);
1603 }
1604
1605 /* Nothing below here should touch the mmu hardware nor the mmu_entry
1606  * data structures.
1607  */
1608
1609 /* First the functions which the mid-level code uses to directly
1610  * manipulate the software page tables.  Some defines since we are
1611  * emulating the i386 page directory layout.
1612  */
1613 #define PGD_PRESENT  0x001
1614 #define PGD_RW       0x002
1615 #define PGD_USER     0x004
1616 #define PGD_ACCESSED 0x020
1617 #define PGD_DIRTY    0x040
1618 #define PGD_TABLE    (PGD_PRESENT | PGD_RW | PGD_USER | PGD_ACCESSED | PGD_DIRTY)
1619
1620 static void sun4c_set_pte(pte_t *ptep, pte_t pte)
1621 {
1622         *ptep = pte;
1623 }
1624
1625 static void sun4c_pgd_set(pgd_t * pgdp, pmd_t * pmdp)
1626 {
1627 }
1628
1629 static void sun4c_pmd_set(pmd_t * pmdp, pte_t * ptep)
1630 {
1631         pmdp->pmdv[0] = PGD_TABLE | (unsigned long) ptep;
1632 }
1633
1634 static void sun4c_pmd_populate(pmd_t * pmdp, struct page * ptep)
1635 {
1636         if (page_address(ptep) == NULL) BUG();  /* No highmem on sun4c */
1637         pmdp->pmdv[0] = PGD_TABLE | (unsigned long) page_address(ptep);
1638 }
1639
1640 static int sun4c_pte_present(pte_t pte)
1641 {
1642         return ((pte_val(pte) & (_SUN4C_PAGE_PRESENT | _SUN4C_PAGE_PRIV)) != 0);
1643 }
1644 static void sun4c_pte_clear(pte_t *ptep)        { *ptep = __pte(0); }
1645
1646 static int sun4c_pmd_bad(pmd_t pmd)
1647 {
1648         return (((pmd_val(pmd) & ~PAGE_MASK) != PGD_TABLE) ||
1649                 (!virt_addr_valid(pmd_val(pmd))));
1650 }
1651
1652 static int sun4c_pmd_present(pmd_t pmd)
1653 {
1654         return ((pmd_val(pmd) & PGD_PRESENT) != 0);
1655 }
1656
1657 #if 0 /* if PMD takes one word */
1658 static void sun4c_pmd_clear(pmd_t *pmdp)        { *pmdp = __pmd(0); }
1659 #else /* if pmd_t is a longish aggregate */
1660 static void sun4c_pmd_clear(pmd_t *pmdp) {
1661         memset((void *)pmdp, 0, sizeof(pmd_t));
1662 }
1663 #endif
1664
1665 static int sun4c_pgd_none(pgd_t pgd)            { return 0; }
1666 static int sun4c_pgd_bad(pgd_t pgd)             { return 0; }
1667 static int sun4c_pgd_present(pgd_t pgd)         { return 1; }
1668 static void sun4c_pgd_clear(pgd_t * pgdp)       { }
1669
1670 /*
1671  * The following only work if pte_present() is true.
1672  * Undefined behaviour if not..
1673  */
1674 static pte_t sun4c_pte_mkwrite(pte_t pte)
1675 {
1676         pte = __pte(pte_val(pte) | _SUN4C_PAGE_WRITE);
1677         if (pte_val(pte) & _SUN4C_PAGE_MODIFIED)
1678                 pte = __pte(pte_val(pte) | _SUN4C_PAGE_SILENT_WRITE);
1679         return pte;
1680 }
1681
1682 static pte_t sun4c_pte_mkdirty(pte_t pte)
1683 {
1684         pte = __pte(pte_val(pte) | _SUN4C_PAGE_MODIFIED);
1685         if (pte_val(pte) & _SUN4C_PAGE_WRITE)
1686                 pte = __pte(pte_val(pte) | _SUN4C_PAGE_SILENT_WRITE);
1687         return pte;
1688 }
1689
1690 static pte_t sun4c_pte_mkyoung(pte_t pte)
1691 {
1692         pte = __pte(pte_val(pte) | _SUN4C_PAGE_ACCESSED);
1693         if (pte_val(pte) & _SUN4C_PAGE_READ)
1694                 pte = __pte(pte_val(pte) | _SUN4C_PAGE_SILENT_READ);
1695         return pte;
1696 }
1697
1698 /*
1699  * Conversion functions: convert a page and protection to a page entry,
1700  * and a page entry and page directory to the page they refer to.
1701  */
1702 static pte_t sun4c_mk_pte(struct page *page, pgprot_t pgprot)
1703 {
1704         return __pte(page_to_pfn(page) | pgprot_val(pgprot));
1705 }
1706
1707 static pte_t sun4c_mk_pte_phys(unsigned long phys_page, pgprot_t pgprot)
1708 {
1709         return __pte((phys_page >> PAGE_SHIFT) | pgprot_val(pgprot));
1710 }
1711
1712 static pte_t sun4c_mk_pte_io(unsigned long page, pgprot_t pgprot, int space)
1713 {
1714         return __pte(((page - PAGE_OFFSET) >> PAGE_SHIFT) | pgprot_val(pgprot));
1715 }
1716
1717 static unsigned long sun4c_pte_pfn(pte_t pte)
1718 {
1719         return pte_val(pte) & SUN4C_PFN_MASK;
1720 }
1721
1722 static pte_t sun4c_pgoff_to_pte(unsigned long pgoff)
1723 {
1724         return __pte(pgoff | _SUN4C_PAGE_FILE);
1725 }
1726
1727 static unsigned long sun4c_pte_to_pgoff(pte_t pte)
1728 {
1729         return pte_val(pte) & ((1UL << PTE_FILE_MAX_BITS) - 1);
1730 }
1731
1732
1733 static inline unsigned long sun4c_pmd_page_v(pmd_t pmd)
1734 {
1735         return (pmd_val(pmd) & PAGE_MASK);
1736 }
1737
1738 static struct page *sun4c_pmd_page(pmd_t pmd)
1739 {
1740         return virt_to_page(sun4c_pmd_page_v(pmd));
1741 }
1742
1743 static unsigned long sun4c_pgd_page(pgd_t pgd) { return 0; }
1744
1745 /* to find an entry in a page-table-directory */
1746 static inline pgd_t *sun4c_pgd_offset(struct mm_struct * mm, unsigned long address)
1747 {
1748         return mm->pgd + (address >> SUN4C_PGDIR_SHIFT);
1749 }
1750
1751 /* Find an entry in the second-level page table.. */
1752 static pmd_t *sun4c_pmd_offset(pgd_t * dir, unsigned long address)
1753 {
1754         return (pmd_t *) dir;
1755 }
1756
1757 /* Find an entry in the third-level page table.. */ 
1758 pte_t *sun4c_pte_offset_kernel(pmd_t * dir, unsigned long address)
1759 {
1760         return (pte_t *) sun4c_pmd_page_v(*dir) +
1761                         ((address >> PAGE_SHIFT) & (SUN4C_PTRS_PER_PTE - 1));
1762 }
1763
1764 static unsigned long sun4c_swp_type(swp_entry_t entry)
1765 {
1766         return (entry.val & SUN4C_SWP_TYPE_MASK);
1767 }
1768
1769 static unsigned long sun4c_swp_offset(swp_entry_t entry)
1770 {
1771         return (entry.val >> SUN4C_SWP_OFF_SHIFT) & SUN4C_SWP_OFF_MASK;
1772 }
1773
1774 static swp_entry_t sun4c_swp_entry(unsigned long type, unsigned long offset)
1775 {
1776         return (swp_entry_t) {
1777                   (offset & SUN4C_SWP_OFF_MASK) << SUN4C_SWP_OFF_SHIFT
1778                 | (type & SUN4C_SWP_TYPE_MASK) };
1779 }
1780
1781 static void sun4c_free_pte_slow(pte_t *pte)
1782 {
1783         free_page((unsigned long)pte);
1784 }
1785
1786 static void sun4c_free_pgd_slow(pgd_t *pgd)
1787 {
1788         free_page((unsigned long)pgd);
1789 }
1790
1791 static pgd_t *sun4c_get_pgd_fast(void)
1792 {
1793         unsigned long *ret;
1794
1795         if ((ret = pgd_quicklist) != NULL) {
1796                 pgd_quicklist = (unsigned long *)(*ret);
1797                 ret[0] = ret[1];
1798                 pgtable_cache_size--;
1799         } else {
1800                 pgd_t *init;
1801                 
1802                 ret = (unsigned long *)__get_free_page(GFP_KERNEL);
1803                 memset (ret, 0, (KERNBASE / SUN4C_PGDIR_SIZE) * sizeof(pgd_t));
1804                 init = sun4c_pgd_offset(&init_mm, 0);
1805                 memcpy (((pgd_t *)ret) + USER_PTRS_PER_PGD, init + USER_PTRS_PER_PGD,
1806                         (PTRS_PER_PGD - USER_PTRS_PER_PGD) * sizeof(pgd_t));
1807         }
1808         return (pgd_t *)ret;
1809 }
1810
1811 static void sun4c_free_pgd_fast(pgd_t *pgd)
1812 {
1813         *(unsigned long *)pgd = (unsigned long) pgd_quicklist;
1814         pgd_quicklist = (unsigned long *) pgd;
1815         pgtable_cache_size++;
1816 }
1817
1818
1819 static inline pte_t *
1820 sun4c_pte_alloc_one_fast(struct mm_struct *mm, unsigned long address)
1821 {
1822         unsigned long *ret;
1823
1824         if ((ret = (unsigned long *)pte_quicklist) != NULL) {
1825                 pte_quicklist = (unsigned long *)(*ret);
1826                 ret[0] = ret[1];
1827                 pgtable_cache_size--;
1828         }
1829         return (pte_t *)ret;
1830 }
1831
1832 static pte_t *sun4c_pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
1833 {
1834         pte_t *pte;
1835
1836         if ((pte = sun4c_pte_alloc_one_fast(mm, address)) != NULL)
1837                 return pte;
1838
1839         pte = (pte_t *)get_zeroed_page(GFP_KERNEL|__GFP_REPEAT);
1840         return pte;
1841 }
1842
1843 static pgtable_t sun4c_pte_alloc_one(struct mm_struct *mm, unsigned long address)
1844 {
1845         pte_t *pte;
1846         struct page *page;
1847
1848         pte = sun4c_pte_alloc_one_kernel(mm, address);
1849         if (pte == NULL)
1850                 return NULL;
1851         page = virt_to_page(pte);
1852         pgtable_page_ctor(page);
1853         return page;
1854 }
1855
1856 static inline void sun4c_free_pte_fast(pte_t *pte)
1857 {
1858         *(unsigned long *)pte = (unsigned long) pte_quicklist;
1859         pte_quicklist = (unsigned long *) pte;
1860         pgtable_cache_size++;
1861 }
1862
1863 static void sun4c_pte_free(pgtable_t pte)
1864 {
1865         pgtable_page_dtor(pte);
1866         sun4c_free_pte_fast(page_address(pte));
1867 }
1868
1869 /*
1870  * allocating and freeing a pmd is trivial: the 1-entry pmd is
1871  * inside the pgd, so has no extra memory associated with it.
1872  */
1873 static pmd_t *sun4c_pmd_alloc_one(struct mm_struct *mm, unsigned long address)
1874 {
1875         BUG();
1876         return NULL;
1877 }
1878
1879 static void sun4c_free_pmd_fast(pmd_t * pmd) { }
1880
1881 static void sun4c_check_pgt_cache(int low, int high)
1882 {
1883         if (pgtable_cache_size > high) {
1884                 do {
1885                         if (pgd_quicklist)
1886                                 sun4c_free_pgd_slow(sun4c_get_pgd_fast());
1887                         if (pte_quicklist)
1888                                 sun4c_free_pte_slow(sun4c_pte_alloc_one_fast(NULL, 0));
1889                 } while (pgtable_cache_size > low);
1890         }
1891 }
1892
1893 /* An experiment, turn off by default for now... -DaveM */
1894 #define SUN4C_PRELOAD_PSEG
1895
1896 void sun4c_update_mmu_cache(struct vm_area_struct *vma, unsigned long address, pte_t pte)
1897 {
1898         unsigned long flags;
1899         int pseg;
1900
1901         if (vma->vm_mm->context == NO_CONTEXT)
1902                 return;
1903
1904         local_irq_save(flags);
1905         address &= PAGE_MASK;
1906         if ((pseg = sun4c_get_segmap(address)) == invalid_segment) {
1907                 struct sun4c_mmu_entry *entry = sun4c_user_strategy();
1908                 struct mm_struct *mm = vma->vm_mm;
1909                 unsigned long start, end;
1910
1911                 entry->vaddr = start = (address & SUN4C_REAL_PGDIR_MASK);
1912                 entry->ctx = mm->context;
1913                 add_ring_ordered(sun4c_context_ring + mm->context, entry);
1914                 sun4c_put_segmap(entry->vaddr, entry->pseg);
1915                 end = start + SUN4C_REAL_PGDIR_SIZE;
1916                 while (start < end) {
1917 #ifdef SUN4C_PRELOAD_PSEG
1918                         pgd_t *pgdp = sun4c_pgd_offset(mm, start);
1919                         pte_t *ptep;
1920
1921                         if (!pgdp)
1922                                 goto no_mapping;
1923                         ptep = sun4c_pte_offset_kernel((pmd_t *) pgdp, start);
1924                         if (!ptep || !(pte_val(*ptep) & _SUN4C_PAGE_PRESENT))
1925                                 goto no_mapping;
1926                         sun4c_put_pte(start, pte_val(*ptep));
1927                         goto next;
1928
1929                 no_mapping:
1930 #endif
1931                         sun4c_put_pte(start, 0);
1932 #ifdef SUN4C_PRELOAD_PSEG
1933                 next:
1934 #endif
1935                         start += PAGE_SIZE;
1936                 }
1937 #ifndef SUN4C_PRELOAD_PSEG
1938                 sun4c_put_pte(address, pte_val(pte));
1939 #endif
1940                 local_irq_restore(flags);
1941                 return;
1942         } else {
1943                 struct sun4c_mmu_entry *entry = &mmu_entry_pool[pseg];
1944
1945                 remove_lru(entry);
1946                 add_lru(entry);
1947         }
1948
1949         sun4c_put_pte(address, pte_val(pte));
1950         local_irq_restore(flags);
1951 }
1952
1953 extern void sparc_context_init(int);
1954 extern unsigned long end;
1955 extern unsigned long bootmem_init(unsigned long *pages_avail);
1956 extern unsigned long last_valid_pfn;
1957
1958 void __init sun4c_paging_init(void)
1959 {
1960         int i, cnt;
1961         unsigned long kernel_end, vaddr;
1962         extern struct resource sparc_iomap;
1963         unsigned long end_pfn, pages_avail;
1964
1965         kernel_end = (unsigned long) &end;
1966         kernel_end = SUN4C_REAL_PGDIR_ALIGN(kernel_end);
1967
1968         pages_avail = 0;
1969         last_valid_pfn = bootmem_init(&pages_avail);
1970         end_pfn = last_valid_pfn;
1971
1972         sun4c_probe_mmu();
1973         invalid_segment = (num_segmaps - 1);
1974         sun4c_init_mmu_entry_pool();
1975         sun4c_init_rings();
1976         sun4c_init_map_kernelprom(kernel_end);
1977         sun4c_init_clean_mmu(kernel_end);
1978         sun4c_init_fill_kernel_ring(SUN4C_KERNEL_BUCKETS);
1979         sun4c_init_lock_area(sparc_iomap.start, IOBASE_END);
1980         sun4c_init_lock_area(DVMA_VADDR, DVMA_END);
1981         sun4c_init_lock_areas();
1982         sun4c_init_fill_user_ring();
1983
1984         sun4c_set_context(0);
1985         memset(swapper_pg_dir, 0, PAGE_SIZE);
1986         memset(pg0, 0, PAGE_SIZE);
1987         memset(pg1, 0, PAGE_SIZE);
1988         memset(pg2, 0, PAGE_SIZE);
1989         memset(pg3, 0, PAGE_SIZE);
1990
1991         /* Save work later. */
1992         vaddr = VMALLOC_START;
1993         swapper_pg_dir[vaddr>>SUN4C_PGDIR_SHIFT] = __pgd(PGD_TABLE | (unsigned long) pg0);
1994         vaddr += SUN4C_PGDIR_SIZE;
1995         swapper_pg_dir[vaddr>>SUN4C_PGDIR_SHIFT] = __pgd(PGD_TABLE | (unsigned long) pg1);
1996         vaddr += SUN4C_PGDIR_SIZE;
1997         swapper_pg_dir[vaddr>>SUN4C_PGDIR_SHIFT] = __pgd(PGD_TABLE | (unsigned long) pg2);
1998         vaddr += SUN4C_PGDIR_SIZE;
1999         swapper_pg_dir[vaddr>>SUN4C_PGDIR_SHIFT] = __pgd(PGD_TABLE | (unsigned long) pg3);
2000         sun4c_init_ss2_cache_bug();
2001         sparc_context_init(num_contexts);
2002
2003         {
2004                 unsigned long zones_size[MAX_NR_ZONES];
2005                 unsigned long zholes_size[MAX_NR_ZONES];
2006                 unsigned long npages;
2007                 int znum;
2008
2009                 for (znum = 0; znum < MAX_NR_ZONES; znum++)
2010                         zones_size[znum] = zholes_size[znum] = 0;
2011
2012                 npages = max_low_pfn - pfn_base;
2013
2014                 zones_size[ZONE_DMA] = npages;
2015                 zholes_size[ZONE_DMA] = npages - pages_avail;
2016
2017                 npages = highend_pfn - max_low_pfn;
2018                 zones_size[ZONE_HIGHMEM] = npages;
2019                 zholes_size[ZONE_HIGHMEM] = npages - calc_highpages();
2020
2021                 free_area_init_node(0, zones_size, pfn_base, zholes_size);
2022         }
2023
2024         cnt = 0;
2025         for (i = 0; i < num_segmaps; i++)
2026                 if (mmu_entry_pool[i].locked)
2027                         cnt++;
2028
2029         max_user_taken_entries = num_segmaps - cnt - 40 - 1;
2030
2031         printk("SUN4C: %d mmu entries for the kernel\n", cnt);
2032 }
2033
2034 static pgprot_t sun4c_pgprot_noncached(pgprot_t prot)
2035 {
2036         prot |= __pgprot(_SUN4C_PAGE_IO | _SUN4C_PAGE_NOCACHE);
2037
2038         return prot;
2039 }
2040
2041 /* Load up routines and constants for sun4c mmu */
2042 void __init ld_mmu_sun4c(void)
2043 {
2044         extern void ___xchg32_sun4c(void);
2045         
2046         printk("Loading sun4c MMU routines\n");
2047
2048         /* First the constants */
2049         BTFIXUPSET_SIMM13(pgdir_shift, SUN4C_PGDIR_SHIFT);
2050         BTFIXUPSET_SETHI(pgdir_size, SUN4C_PGDIR_SIZE);
2051         BTFIXUPSET_SETHI(pgdir_mask, SUN4C_PGDIR_MASK);
2052
2053         BTFIXUPSET_SIMM13(ptrs_per_pmd, SUN4C_PTRS_PER_PMD);
2054         BTFIXUPSET_SIMM13(ptrs_per_pgd, SUN4C_PTRS_PER_PGD);
2055         BTFIXUPSET_SIMM13(user_ptrs_per_pgd, KERNBASE / SUN4C_PGDIR_SIZE);
2056
2057         BTFIXUPSET_INT(page_none, pgprot_val(SUN4C_PAGE_NONE));
2058         PAGE_SHARED = pgprot_val(SUN4C_PAGE_SHARED);
2059         BTFIXUPSET_INT(page_copy, pgprot_val(SUN4C_PAGE_COPY));
2060         BTFIXUPSET_INT(page_readonly, pgprot_val(SUN4C_PAGE_READONLY));
2061         BTFIXUPSET_INT(page_kernel, pgprot_val(SUN4C_PAGE_KERNEL));
2062         page_kernel = pgprot_val(SUN4C_PAGE_KERNEL);
2063
2064         /* Functions */
2065         BTFIXUPSET_CALL(pgprot_noncached, sun4c_pgprot_noncached, BTFIXUPCALL_NORM);
2066         BTFIXUPSET_CALL(___xchg32, ___xchg32_sun4c, BTFIXUPCALL_NORM);
2067         BTFIXUPSET_CALL(do_check_pgt_cache, sun4c_check_pgt_cache, BTFIXUPCALL_NORM);
2068         
2069         BTFIXUPSET_CALL(flush_cache_all, sun4c_flush_cache_all, BTFIXUPCALL_NORM);
2070
2071         if (sun4c_vacinfo.do_hwflushes) {
2072                 BTFIXUPSET_CALL(sun4c_flush_page, sun4c_flush_page_hw, BTFIXUPCALL_NORM);
2073                 BTFIXUPSET_CALL(sun4c_flush_segment, sun4c_flush_segment_hw, BTFIXUPCALL_NORM);
2074                 BTFIXUPSET_CALL(sun4c_flush_context, sun4c_flush_context_hw, BTFIXUPCALL_NORM);
2075         } else {
2076                 BTFIXUPSET_CALL(sun4c_flush_page, sun4c_flush_page_sw, BTFIXUPCALL_NORM);
2077                 BTFIXUPSET_CALL(sun4c_flush_segment, sun4c_flush_segment_sw, BTFIXUPCALL_NORM);
2078                 BTFIXUPSET_CALL(sun4c_flush_context, sun4c_flush_context_sw, BTFIXUPCALL_NORM);
2079         }
2080
2081         BTFIXUPSET_CALL(flush_tlb_mm, sun4c_flush_tlb_mm, BTFIXUPCALL_NORM);
2082         BTFIXUPSET_CALL(flush_cache_mm, sun4c_flush_cache_mm, BTFIXUPCALL_NORM);
2083         BTFIXUPSET_CALL(destroy_context, sun4c_destroy_context, BTFIXUPCALL_NORM);
2084         BTFIXUPSET_CALL(switch_mm, sun4c_switch_mm, BTFIXUPCALL_NORM);
2085         BTFIXUPSET_CALL(flush_cache_page, sun4c_flush_cache_page, BTFIXUPCALL_NORM);
2086         BTFIXUPSET_CALL(flush_tlb_page, sun4c_flush_tlb_page, BTFIXUPCALL_NORM);
2087         BTFIXUPSET_CALL(flush_tlb_range, sun4c_flush_tlb_range, BTFIXUPCALL_NORM);
2088         BTFIXUPSET_CALL(flush_cache_range, sun4c_flush_cache_range, BTFIXUPCALL_NORM);
2089         BTFIXUPSET_CALL(__flush_page_to_ram, sun4c_flush_page_to_ram, BTFIXUPCALL_NORM);
2090         BTFIXUPSET_CALL(flush_tlb_all, sun4c_flush_tlb_all, BTFIXUPCALL_NORM);
2091
2092         BTFIXUPSET_CALL(flush_sig_insns, sun4c_flush_sig_insns, BTFIXUPCALL_NOP);
2093
2094         BTFIXUPSET_CALL(set_pte, sun4c_set_pte, BTFIXUPCALL_STO1O0);
2095
2096         /* The 2.4.18 code does not set this on sun4c, how does it work? XXX */
2097         /* BTFIXUPSET_SETHI(none_mask, 0x00000000); */  /* Defaults to zero? */
2098
2099         BTFIXUPSET_CALL(pte_pfn, sun4c_pte_pfn, BTFIXUPCALL_NORM);
2100 #if 0 /* PAGE_SHIFT <= 12 */ /* Eek. Investigate. XXX */
2101         BTFIXUPSET_CALL(pmd_page, sun4c_pmd_page, BTFIXUPCALL_ANDNINT(PAGE_SIZE - 1));
2102 #else
2103         BTFIXUPSET_CALL(pmd_page, sun4c_pmd_page, BTFIXUPCALL_NORM);
2104 #endif
2105         BTFIXUPSET_CALL(pmd_set, sun4c_pmd_set, BTFIXUPCALL_NORM);
2106         BTFIXUPSET_CALL(pmd_populate, sun4c_pmd_populate, BTFIXUPCALL_NORM);
2107
2108         BTFIXUPSET_CALL(pte_present, sun4c_pte_present, BTFIXUPCALL_NORM);
2109         BTFIXUPSET_CALL(pte_clear, sun4c_pte_clear, BTFIXUPCALL_STG0O0);
2110
2111         BTFIXUPSET_CALL(pmd_bad, sun4c_pmd_bad, BTFIXUPCALL_NORM);
2112         BTFIXUPSET_CALL(pmd_present, sun4c_pmd_present, BTFIXUPCALL_NORM);
2113         BTFIXUPSET_CALL(pmd_clear, sun4c_pmd_clear, BTFIXUPCALL_STG0O0);
2114
2115         BTFIXUPSET_CALL(pgd_none, sun4c_pgd_none, BTFIXUPCALL_RETINT(0));
2116         BTFIXUPSET_CALL(pgd_bad, sun4c_pgd_bad, BTFIXUPCALL_RETINT(0));
2117         BTFIXUPSET_CALL(pgd_present, sun4c_pgd_present, BTFIXUPCALL_RETINT(1));
2118         BTFIXUPSET_CALL(pgd_clear, sun4c_pgd_clear, BTFIXUPCALL_NOP);
2119
2120         BTFIXUPSET_CALL(mk_pte, sun4c_mk_pte, BTFIXUPCALL_NORM);
2121         BTFIXUPSET_CALL(mk_pte_phys, sun4c_mk_pte_phys, BTFIXUPCALL_NORM);
2122         BTFIXUPSET_CALL(mk_pte_io, sun4c_mk_pte_io, BTFIXUPCALL_NORM);
2123
2124         BTFIXUPSET_INT(pte_modify_mask, _SUN4C_PAGE_CHG_MASK);
2125         BTFIXUPSET_CALL(pmd_offset, sun4c_pmd_offset, BTFIXUPCALL_NORM);
2126         BTFIXUPSET_CALL(pte_offset_kernel, sun4c_pte_offset_kernel, BTFIXUPCALL_NORM);
2127         BTFIXUPSET_CALL(free_pte_fast, sun4c_free_pte_fast, BTFIXUPCALL_NORM);
2128         BTFIXUPSET_CALL(pte_free, sun4c_pte_free, BTFIXUPCALL_NORM);
2129         BTFIXUPSET_CALL(pte_alloc_one_kernel, sun4c_pte_alloc_one_kernel, BTFIXUPCALL_NORM);
2130         BTFIXUPSET_CALL(pte_alloc_one, sun4c_pte_alloc_one, BTFIXUPCALL_NORM);
2131         BTFIXUPSET_CALL(free_pmd_fast, sun4c_free_pmd_fast, BTFIXUPCALL_NOP);
2132         BTFIXUPSET_CALL(pmd_alloc_one, sun4c_pmd_alloc_one, BTFIXUPCALL_RETO0);
2133         BTFIXUPSET_CALL(free_pgd_fast, sun4c_free_pgd_fast, BTFIXUPCALL_NORM);
2134         BTFIXUPSET_CALL(get_pgd_fast, sun4c_get_pgd_fast, BTFIXUPCALL_NORM);
2135
2136         BTFIXUPSET_HALF(pte_writei, _SUN4C_PAGE_WRITE);
2137         BTFIXUPSET_HALF(pte_dirtyi, _SUN4C_PAGE_MODIFIED);
2138         BTFIXUPSET_HALF(pte_youngi, _SUN4C_PAGE_ACCESSED);
2139         BTFIXUPSET_HALF(pte_filei, _SUN4C_PAGE_FILE);
2140         BTFIXUPSET_HALF(pte_wrprotecti, _SUN4C_PAGE_WRITE|_SUN4C_PAGE_SILENT_WRITE);
2141         BTFIXUPSET_HALF(pte_mkcleani, _SUN4C_PAGE_MODIFIED|_SUN4C_PAGE_SILENT_WRITE);
2142         BTFIXUPSET_HALF(pte_mkoldi, _SUN4C_PAGE_ACCESSED|_SUN4C_PAGE_SILENT_READ);
2143         BTFIXUPSET_CALL(pte_mkwrite, sun4c_pte_mkwrite, BTFIXUPCALL_NORM);
2144         BTFIXUPSET_CALL(pte_mkdirty, sun4c_pte_mkdirty, BTFIXUPCALL_NORM);
2145         BTFIXUPSET_CALL(pte_mkyoung, sun4c_pte_mkyoung, BTFIXUPCALL_NORM);
2146         BTFIXUPSET_CALL(update_mmu_cache, sun4c_update_mmu_cache, BTFIXUPCALL_NORM);
2147
2148         BTFIXUPSET_CALL(pte_to_pgoff, sun4c_pte_to_pgoff, BTFIXUPCALL_NORM);
2149         BTFIXUPSET_CALL(pgoff_to_pte, sun4c_pgoff_to_pte, BTFIXUPCALL_NORM);
2150
2151         BTFIXUPSET_CALL(mmu_lockarea, sun4c_lockarea, BTFIXUPCALL_NORM);
2152         BTFIXUPSET_CALL(mmu_unlockarea, sun4c_unlockarea, BTFIXUPCALL_NORM);
2153
2154         BTFIXUPSET_CALL(mmu_get_scsi_one, sun4c_get_scsi_one, BTFIXUPCALL_NORM);
2155         BTFIXUPSET_CALL(mmu_get_scsi_sgl, sun4c_get_scsi_sgl, BTFIXUPCALL_NORM);
2156         BTFIXUPSET_CALL(mmu_release_scsi_one, sun4c_release_scsi_one, BTFIXUPCALL_NORM);
2157         BTFIXUPSET_CALL(mmu_release_scsi_sgl, sun4c_release_scsi_sgl, BTFIXUPCALL_NORM);
2158
2159         BTFIXUPSET_CALL(mmu_map_dma_area, sun4c_map_dma_area, BTFIXUPCALL_NORM);
2160         BTFIXUPSET_CALL(mmu_unmap_dma_area, sun4c_unmap_dma_area, BTFIXUPCALL_NORM);
2161
2162         BTFIXUPSET_CALL(sparc_mapiorange, sun4c_mapiorange, BTFIXUPCALL_NORM);
2163         BTFIXUPSET_CALL(sparc_unmapiorange, sun4c_unmapiorange, BTFIXUPCALL_NORM);
2164
2165         BTFIXUPSET_CALL(__swp_type, sun4c_swp_type, BTFIXUPCALL_NORM);
2166         BTFIXUPSET_CALL(__swp_offset, sun4c_swp_offset, BTFIXUPCALL_NORM);
2167         BTFIXUPSET_CALL(__swp_entry, sun4c_swp_entry, BTFIXUPCALL_NORM);
2168
2169         BTFIXUPSET_CALL(alloc_thread_info, sun4c_alloc_thread_info, BTFIXUPCALL_NORM);
2170         BTFIXUPSET_CALL(free_thread_info, sun4c_free_thread_info, BTFIXUPCALL_NORM);
2171
2172         BTFIXUPSET_CALL(mmu_info, sun4c_mmu_info, BTFIXUPCALL_NORM);
2173
2174         /* These should _never_ get called with two level tables. */
2175         BTFIXUPSET_CALL(pgd_set, sun4c_pgd_set, BTFIXUPCALL_NOP);
2176         BTFIXUPSET_CALL(pgd_page_vaddr, sun4c_pgd_page, BTFIXUPCALL_RETO0);
2177 }