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