[coldfire]: 2.6.31 support (WiP)
[15.05/openwrt.git] / target / linux / coldfire / patches / 007-mcfv4e_arch_mm_mods_1.patch
1 From 2bef1f8ce148cce9e782f75f9537767c1d8c0eea Mon Sep 17 00:00:00 2001
2 From: Kurt Mahan <kmahan@freescale.com>
3 Date: Wed, 31 Oct 2007 16:58:27 -0600
4 Subject: [PATCH] Core Coldfire/MCF5445x arch/mm changes.
5
6 LTIBName: mcfv4e-arch-mm-mods-1
7 Signed-off-by: Kurt Mahan <kmahan@freescale.com>
8 ---
9  arch/m68k/mm/Makefile |    1 +
10  arch/m68k/mm/cache.c  |   41 ++++++++
11  arch/m68k/mm/cf-mmu.c |  251 +++++++++++++++++++++++++++++++++++++++++++++++++
12  arch/m68k/mm/hwtest.c |    2 +
13  arch/m68k/mm/init.c   |    3 +-
14  arch/m68k/mm/kmap.c   |   13 +++
15  arch/m68k/mm/memory.c |   66 +++++++++++++-
16  7 files changed, 373 insertions(+), 4 deletions(-)
17  create mode 100644 arch/m68k/mm/cf-mmu.c
18
19 --- a/arch/m68k/mm/Makefile
20 +++ b/arch/m68k/mm/Makefile
21 @@ -6,3 +6,4 @@ obj-y           := cache.o init.o fault.o hwtest.
22  
23  obj-$(CONFIG_MMU_MOTOROLA)     += kmap.o memory.o motorola.o
24  obj-$(CONFIG_MMU_SUN3)         += sun3kmap.o sun3mmu.o
25 +obj-$(CONFIG_MMU_CFV4E)                += cf-mmu.o kmap.o memory.o
26 --- a/arch/m68k/mm/cache.c
27 +++ b/arch/m68k/mm/cache.c
28 @@ -10,7 +10,11 @@
29  #include <asm/pgalloc.h>
30  #include <asm/traps.h>
31  
32 +#ifdef CONFIG_COLDFIRE
33 +#include <asm/cfcache.h>
34 +#endif /* CONFIG_COLDFIRE */
35  
36 +#ifndef CONFIG_COLDFIRE
37  static unsigned long virt_to_phys_slow(unsigned long vaddr)
38  {
39         if (CPU_IS_060) {
40 @@ -69,11 +73,45 @@ static unsigned long virt_to_phys_slow(u
41         }
42         return 0;
43  }
44 +#endif /* CONFIG_COLDFIRE */
45 +
46  
47  /* Push n pages at kernel virtual address and clear the icache */
48  /* RZ: use cpush %bc instead of cpush %dc, cinv %ic */
49  void flush_icache_range(unsigned long address, unsigned long endaddr)
50  {
51 +#ifdef CONFIG_COLDFIRE
52 +       unsigned long set;
53 +       unsigned long start_set;
54 +       unsigned long end_set;
55 +
56 +       start_set = address & _ICACHE_SET_MASK;
57 +       end_set = endaddr & _ICACHE_SET_MASK;
58 +
59 +       if (start_set > end_set) {
60 +       /* from the begining to the lowest address */
61 +               for (set = 0; set <= end_set; set += (0x10 - 3))
62 +                       asm volatile ("cpushl %%ic,(%0)\n"
63 +                                     "\taddq%.l #1,%0\n"
64 +                                     "\tcpushl %%ic,(%0)\n"
65 +                                     "\taddq%.l #1,%0\n"
66 +                                     "\tcpushl %%ic,(%0)\n"
67 +                                     "\taddq%.l #1,%0\n"
68 +                                     "\tcpushl %%ic,(%0)" : : "a" (set));
69 +
70 +               /* next loop will finish the cache ie pass the hole */
71 +               end_set = LAST_ICACHE_ADDR;
72 +       }
73 +       for (set = start_set; set <= end_set; set += (0x10 - 3))
74 +               asm volatile ("cpushl %%ic,(%0)\n"
75 +                             "\taddq%.l #1,%0\n"
76 +                             "\tcpushl %%ic,(%0)\n"
77 +                             "\taddq%.l #1,%0\n"
78 +                             "\tcpushl %%ic,(%0)\n"
79 +                             "\taddq%.l #1,%0\n"
80 +                             "\tcpushl %%ic,(%0)" : : "a" (set));
81 +
82 +#else /* !CONFIG_COLDFIRE */
83  
84         if (CPU_IS_040_OR_060) {
85                 address &= PAGE_MASK;
86 @@ -94,9 +132,11 @@ void flush_icache_range(unsigned long ad
87                               : "=&d" (tmp)
88                               : "di" (FLUSH_I));
89         }
90 +#endif /* CONFIG_COLDFIRE */
91  }
92  EXPORT_SYMBOL(flush_icache_range);
93  
94 +#ifndef CONFIG_COLDFIRE
95  void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
96                              unsigned long addr, int len)
97  {
98 @@ -115,4 +155,5 @@ void flush_icache_user_range(struct vm_a
99                               : "di" (FLUSH_I));
100         }
101  }
102 +#endif /* CONFIG_COLDFIRE */
103  
104 --- /dev/null
105 +++ b/arch/m68k/mm/cf-mmu.c
106 @@ -0,0 +1,251 @@
107 +/*
108 + * linux/arch/m68k/mm/cf-mmu.c
109 + *
110 + * Based upon linux/arch/m68k/mm/sun3mmu.c
111 + * Based upon linux/arch/ppc/mm/mmu_context.c
112 + *
113 + * Implementations of mm routines specific to the Coldfire MMU.
114 + *
115 + * Copyright (c) 2008 Freescale Semiconductor, Inc.
116 + */
117 +
118 +#include <linux/signal.h>
119 +#include <linux/sched.h>
120 +#include <linux/mm.h>
121 +#include <linux/swap.h>
122 +#include <linux/kernel.h>
123 +#include <linux/string.h>
124 +#include <linux/types.h>
125 +#include <linux/init.h>
126 +#ifdef CONFIG_BLK_DEV_RAM
127 +#include <linux/blkdev.h>
128 +#endif
129 +#include <linux/bootmem.h>
130 +
131 +#include <asm/setup.h>
132 +#include <asm/uaccess.h>
133 +#include <asm/page.h>
134 +#include <asm/pgtable.h>
135 +#include <asm/system.h>
136 +#include <asm/machdep.h>
137 +#include <asm/io.h>
138 +#include <asm/mmu_context.h>
139 +#include <asm/cf_pgalloc.h>
140 +
141 +#include <asm/coldfire.h>
142 +#include <asm/tlbflush.h>
143 +
144 +mm_context_t next_mmu_context;
145 +unsigned long context_map[LAST_CONTEXT / BITS_PER_LONG + 1];
146 +
147 +atomic_t nr_free_contexts;
148 +struct mm_struct *context_mm[LAST_CONTEXT+1];
149 +void steal_context(void);
150 +
151 +
152 +const char bad_pmd_string[] = "Bad pmd in pte_alloc: %08lx\n";
153 +
154 +extern unsigned long empty_bad_page_table;
155 +extern unsigned long empty_bad_page;
156 +extern unsigned long num_pages;
157 +
158 +extern char __init_begin, __init_end;
159 +
160 +void free_initmem(void)
161 +{
162 +       unsigned long addr;
163 +       unsigned long start = (unsigned long)&__init_begin;
164 +       unsigned long end = (unsigned long)&__init_end;
165 +
166 +       printk(KERN_INFO "free_initmem: __init_begin = 0x%lx  __init_end = 0x%lx\n", start, end);
167 +
168 +       addr = (unsigned long)&__init_begin;
169 +       for (; addr < (unsigned long)&__init_end; addr += PAGE_SIZE) {
170 +               /* not currently used */
171 +               virt_to_page(addr)->flags &= ~(1 << PG_reserved);
172 +               init_page_count(virt_to_page(addr));
173 +               free_page(addr);
174 +               totalram_pages++;
175 +       }
176 +}
177 +
178 +/* Coldfire paging_init derived from sun3 */
179 +void __init paging_init(void)
180 +{
181 +       pgd_t * pg_dir;
182 +       pte_t * pg_table;
183 +       int i;
184 +       unsigned long address;
185 +       unsigned long next_pgtable;
186 +       unsigned long bootmem_end;
187 +       unsigned long zones_size[MAX_NR_ZONES];
188 +       unsigned long size;
189 +       enum zone_type zone;
190 +
191 +       empty_zero_page = (void *)alloc_bootmem_pages(PAGE_SIZE);
192 +       memset((void *)empty_zero_page, 0, PAGE_SIZE);
193 +
194 +       pg_dir = swapper_pg_dir;
195 +       memset(swapper_pg_dir, 0, sizeof (swapper_pg_dir));
196 +
197 +       size = num_pages * sizeof(pte_t);  
198 +       size = (size + PAGE_SIZE) & ~(PAGE_SIZE-1);
199 +       next_pgtable = (unsigned long)alloc_bootmem_pages(size);
200 +
201 +       bootmem_end = (next_pgtable + size + PAGE_SIZE) & PAGE_MASK; 
202 +       pg_dir += PAGE_OFFSET >> PGDIR_SHIFT; 
203 +
204 +       address = PAGE_OFFSET;
205 +       while (address < (unsigned long)high_memory) 
206 +       {
207 +               pg_table = (pte_t *)next_pgtable; 
208 +               next_pgtable += PTRS_PER_PTE * sizeof (pte_t);
209 +               pgd_val(*pg_dir) = (unsigned long) pg_table;
210 +               pg_dir++;
211 +
212 +               /* now change pg_table to kernel virtual addresses */
213 +               for (i=0; i<PTRS_PER_PTE; ++i, ++pg_table)
214 +               {
215 +                       pte_t pte = pfn_pte(virt_to_pfn(address), PAGE_INIT);
216 +                       if (address >= (unsigned long)high_memory)
217 +                               pte_val (pte) = 0;
218 +
219 +                       set_pte (pg_table, pte);
220 +                       address += PAGE_SIZE;
221 +               }
222 +       }
223 +       
224 +       current->mm = NULL;
225 +
226 +       /* clear zones */
227 +       for (zone = 0; zone < MAX_NR_ZONES; zone++)
228 +               zones_size[zone] = 0x0;
229 +
230 +       /* allocate the bottom 32M (0x40x 0x41x) to DMA - head.S marks them NO CACHE */
231 +       /* JKM - this should be changed to allocate from the TOP (0x4f,0x4e) but the
232 +        * allocator is being a bit challenging */
233 +        zones_size[ZONE_DMA] = (32*1024*1024) >> PAGE_SHIFT;
234 +
235 +       /* allocate the rest to NORMAL - head.S marks them CACHE */
236 +       zones_size[ZONE_NORMAL] = (((unsigned long)high_memory - PAGE_OFFSET) >> PAGE_SHIFT) - zones_size[0];
237 +
238 +       free_area_init(zones_size);
239 +}
240 +
241 +
242 +int cf_tlb_miss(struct pt_regs *regs, int write, int dtlb, int extension_word)
243 +{
244 +        struct mm_struct *mm;
245 +        pgd_t *pgd;
246 +        pmd_t *pmd;
247 +        pte_t *pte;
248 +        unsigned long mmuar;
249 +        int asid;
250 +       int flags;
251 +
252 +       local_save_flags(flags);
253 +       local_irq_disable();
254 +
255 +       mmuar = ( dtlb ) ? regs->mmuar
256 +                        : regs->pc + (extension_word * sizeof(long));
257 +
258 +        mm = (!user_mode(regs) && (mmuar >= PAGE_OFFSET)) ? &init_mm
259 +                                                         : current->mm;
260 +        if (!mm) {
261 +           local_irq_restore(flags);
262 +           return (-1);
263 +       }
264 +
265 +        pgd = pgd_offset(mm, mmuar);
266 +        if (pgd_none(*pgd))  {
267 +           local_irq_restore(flags);
268 +           return (-1);
269 +       }
270 +           
271 +       pmd = pmd_offset(pgd, mmuar);
272 +       if (pmd_none(*pmd)) {
273 +           local_irq_restore(flags);
274 +           return (-1);
275 +       }       
276 +       
277 +       pte = (mmuar >= PAGE_OFFSET) ? pte_offset_kernel(pmd, mmuar)
278 +                                    : pte_offset_map(pmd, mmuar);
279 +       if (pte_none(*pte) || !pte_present(*pte)) {
280 +           local_irq_restore(flags);
281 +           return (-1);                
282 +       }
283 +
284 +       if (write) {
285 +            if (!pte_write(*pte)) {
286 +               local_irq_restore(flags);
287 +               return (-1);
288 +           }
289 +           set_pte(pte, pte_mkdirty(*pte));
290 +       }
291 +       
292 +        set_pte(pte, pte_mkyoung(*pte));
293 +        asid = mm->context & 0xff;
294 +        if (!pte_dirty(*pte) && mmuar<=PAGE_OFFSET)
295 +           set_pte(pte, pte_wrprotect(*pte));
296 +
297 +        *MMUTR = (mmuar & PAGE_MASK) | (asid << CF_ASID_MMU_SHIFT)
298 +               | (((int)(pte->pte) & (int)CF_PAGE_MMUTR_MASK ) >> CF_PAGE_MMUTR_SHIFT)
299 +               | MMUTR_V;
300 +
301 +        *MMUDR = (pte_val(*pte) & PAGE_MASK) 
302 +              | ((pte->pte) & CF_PAGE_MMUDR_MASK)
303 +               | MMUDR_SZ8K | MMUDR_X;
304 +           
305 +       if ( dtlb )
306 +           *MMUOR = MMUOR_ACC | MMUOR_UAA;
307 +       else
308 +           *MMUOR = MMUOR_ITLB | MMUOR_ACC | MMUOR_UAA;
309 +
310 +       asm ("nop");
311 +       /*printk("cf_tlb_miss: va=%lx, pa=%lx\n", (mmuar & PAGE_MASK), 
312 +                 (pte_val(*pte)  & PAGE_MASK));*/
313 +       local_irq_restore(flags);
314 +        return (0);
315 +}
316 +
317 +
318 +/* The following was taken from arch/ppc/mmu_context.c
319 + *
320 + * Initialize the context management stuff.
321 + */
322 +void __init mmu_context_init(void)
323 +{
324 +       /*
325 +        * Some processors have too few contexts to reserve one for
326 +        * init_mm, and require using context 0 for a normal task.
327 +        * Other processors reserve the use of context zero for the kernel.
328 +        * This code assumes FIRST_CONTEXT < 32.
329 +        */
330 +       context_map[0] = (1 << FIRST_CONTEXT) - 1;
331 +       next_mmu_context = FIRST_CONTEXT;
332 +       atomic_set(&nr_free_contexts, LAST_CONTEXT - FIRST_CONTEXT + 1);
333 +}
334 +
335 +/*
336 + * Steal a context from a task that has one at the moment.
337 + * This is only used on 8xx and 4xx and we presently assume that
338 + * they don't do SMP.  If they do then thicfpgalloc.hs will have to check
339 + * whether the MM we steal is in use.
340 + * We also assume that this is only used on systems that don't
341 + * use an MMU hash table - this is true for 8xx and 4xx.
342 + * This isn't an LRU system, it just frees up each context in
343 + * turn (sort-of pseudo-random replacement :).  This would be the
344 + * place to implement an LRU scheme if anyone was motivated to do it.
345 + *  -- paulus
346 + */
347 +void steal_context(void)
348 +{
349 +       struct mm_struct *mm;
350 +       /* free up context `next_mmu_context' */
351 +       /* if we shouldn't free context 0, don't... */
352 +       if (next_mmu_context < FIRST_CONTEXT)
353 +               next_mmu_context = FIRST_CONTEXT;
354 +       mm = context_mm[next_mmu_context];
355 +       flush_tlb_mm(mm);
356 +       destroy_context(mm);
357 +}
358 --- a/arch/m68k/mm/hwtest.c
359 +++ b/arch/m68k/mm/hwtest.c
360 @@ -25,6 +25,7 @@
361  
362  #include <linux/module.h>
363  
364 +#ifndef CONFIG_COLDFIRE
365  int hwreg_present( volatile void *regp )
366  {
367      int        ret = 0;
368 @@ -82,4 +83,5 @@ int hwreg_write( volatile void *regp, un
369         return( ret );
370  }
371  EXPORT_SYMBOL(hwreg_write);
372 +#endif
373  
374 --- a/arch/m68k/mm/init.c
375 +++ b/arch/m68k/mm/init.c
376 @@ -122,7 +122,6 @@ void __init mem_init(void)
377         if (MACH_IS_ATARI)
378                 atari_stram_mem_init_hook();
379  #endif
380 -
381         /* this will put all memory onto the freelists */
382         totalram_pages = num_physpages = 0;
383         for_each_online_pgdat(pgdat) {
384 @@ -146,7 +145,7 @@ void __init mem_init(void)
385                 }
386         }
387  
388 -#ifndef CONFIG_SUN3
389 +#if !defined(CONFIG_SUN3) && !defined(CONFIG_COLDFIRE)
390         /* insert pointer tables allocated so far into the tablelist */
391         init_pointer_table((unsigned long)kernel_pg_dir);
392         for (i = 0; i < PTRS_PER_PGD; i++) {
393 --- a/arch/m68k/mm/kmap.c
394 +++ b/arch/m68k/mm/kmap.c
395 @@ -24,7 +24,11 @@
396  
397  #undef DEBUG
398  
399 +#ifndef CONFIG_COLDFIRE
400  #define PTRTREESIZE    (256*1024)
401 +#else
402 +#define PTRTREESIZE     PAGE_SIZE
403 +#endif
404  
405  /*
406   * For 040/060 we can use the virtual memory area like other architectures,
407 @@ -50,7 +54,11 @@ static inline void free_io_area(void *ad
408  
409  #else
410  
411 +#ifdef CONFIG_COLDFIRE
412 +#define IO_SIZE         PAGE_SIZE
413 +#else
414  #define IO_SIZE                (256*1024)
415 +#endif
416  
417  static struct vm_struct *iolist;
418  
419 @@ -170,7 +178,12 @@ void __iomem *__ioremap(unsigned long ph
420                         break;
421                 }
422         } else {
423 +#ifndef CONFIG_COLDFIRE
424                 physaddr |= (_PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_DIRTY);
425 +#else
426 +               physaddr |= (_PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_DIRTY | \
427 +                            _PAGE_READWRITE);
428 +#endif
429                 switch (cacheflag) {
430                 case IOMAP_NOCACHE_SER:
431                 case IOMAP_NOCACHE_NONSER:
432 --- a/arch/m68k/mm/memory.c
433 +++ b/arch/m68k/mm/memory.c
434 @@ -203,7 +203,38 @@ static inline void pushcl040(unsigned lo
435  
436  void cache_clear (unsigned long paddr, int len)
437  {
438 -    if (CPU_IS_040_OR_060) {
439 +    if (CPU_IS_CFV4E) {
440 +       unsigned long set;
441 +       unsigned long start_set;
442 +       unsigned long end_set;
443 +
444 +       start_set = paddr & _ICACHE_SET_MASK;
445 +       end_set = (paddr+len-1) & _ICACHE_SET_MASK;
446 +
447 +       if (start_set > end_set) {
448 +               /* from the begining to the lowest address */
449 +               for (set = 0; set <= end_set; set += (0x10 - 3))
450 +                       asm volatile("cpushl %%bc,(%0)\n"
451 +                                    "\taddq%.l #1,%0\n"
452 +                                    "\tcpushl %%bc,(%0)\n"
453 +                                    "\taddq%.l #1,%0\n"
454 +                                    "\tcpushl %%bc,(%0)\n"
455 +                                    "\taddq%.l #1,%0\n"
456 +                                    "\tcpushl %%bc,(%0)" : : "a" (set));
457 +
458 +               /* next loop will finish the cache ie pass the hole */
459 +               end_set = LAST_ICACHE_ADDR;
460 +       }
461 +       for (set = start_set; set <= end_set; set += (0x10 - 3))
462 +               asm volatile("cpushl %%bc,(%0)\n"
463 +                            "\taddq%.l #1,%0\n"
464 +                            "\tcpushl %%bc,(%0)\n"
465 +                            "\taddq%.l #1,%0\n"
466 +                            "\tcpushl %%bc,(%0)\n"
467 +                            "\taddq%.l #1,%0\n"
468 +                            "\tcpushl %%bc,(%0)" : : "a" (set));
469 +
470 +    } else if (CPU_IS_040_OR_060) {
471         int tmp;
472  
473         /*
474 @@ -250,7 +281,38 @@ EXPORT_SYMBOL(cache_clear);
475  
476  void cache_push (unsigned long paddr, int len)
477  {
478 -    if (CPU_IS_040_OR_060) {
479 +    if (CPU_IS_CFV4E) {
480 +       unsigned long set;
481 +       unsigned long start_set;
482 +       unsigned long end_set;
483 +
484 +       start_set = paddr & _ICACHE_SET_MASK;
485 +       end_set = (paddr+len-1) & _ICACHE_SET_MASK;
486 +
487 +       if (start_set > end_set) {
488 +               /* from the begining to the lowest address */
489 +               for (set = 0; set <= end_set; set += (0x10 - 3))
490 +                       asm volatile("cpushl %%bc,(%0)\n"
491 +                                    "\taddq%.l #1,%0\n"
492 +                                    "\tcpushl %%bc,(%0)\n"
493 +                                    "\taddq%.l #1,%0\n"
494 +                                    "\tcpushl %%bc,(%0)\n"
495 +                                    "\taddq%.l #1,%0\n"
496 +                                    "\tcpushl %%bc,(%0)" : : "a" (set));
497 +
498 +               /* next loop will finish the cache ie pass the hole */
499 +               end_set = LAST_ICACHE_ADDR;
500 +       }
501 +       for (set = start_set; set <= end_set; set += (0x10 - 3))
502 +               asm volatile("cpushl %%bc,(%0)\n"
503 +                            "\taddq%.l #1,%0\n"
504 +                            "\tcpushl %%bc,(%0)\n"
505 +                            "\taddq%.l #1,%0\n"
506 +                            "\tcpushl %%bc,(%0)\n"
507 +                            "\taddq%.l #1,%0\n"
508 +                            "\tcpushl %%bc,(%0)" : : "a" (set));
509 +
510 +    } else if (CPU_IS_040_OR_060) {
511         int tmp = PAGE_SIZE;
512  
513         /*