aboutsummaryrefslogtreecommitdiffstats
path: root/arch/sparc64/mm/init.c
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@g5.osdl.org>2006-03-20 14:57:50 -0500
committerLinus Torvalds <torvalds@g5.osdl.org>2006-03-20 14:57:50 -0500
commitc4a1745aa09fc110afdefea0e5d025043e348bae (patch)
tree6d28dc3a0c1bf18437b3d49f28e5c81b850cdb2f /arch/sparc64/mm/init.c
parent88dcb91177cfa5b26143a29074389a2aa259c7cf (diff)
parentac0eb3eb7e54b700386068be025a43d2a3958ee5 (diff)
Merge master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6
* master.kernel.org:/pub/scm/linux/kernel/git/davem/sparc-2.6: (230 commits) [SPARC64]: Update defconfig. [SPARC64]: Fix 2 bugs in huge page support. [SPARC64]: CONFIG_BLK_DEV_RAM fix [SPARC64]: Optimized TSB table initialization. [SPARC64]: Allow CONFIG_MEMORY_HOTPLUG to build. [SPARC64]: Use SLAB caches for TSB tables. [SPARC64]: Don't kill the page allocator when growing a TSB. [SPARC64]: Randomize mm->mmap_base when PF_RANDOMIZE is set. [SPARC64]: Increase top of 32-bit process stack. [SPARC64]: Top-down address space allocation for 32-bit tasks. [SPARC64] bbc_i2c: Fix cpu check and add missing module license. [SPARC64]: Fix and re-enable dynamic TSB sizing. [SUNSU]: Fix missing spinlock initialization. [TG3]: Do not try to access NIC_SRAM_DATA_SIG on Sun parts. [SPARC64]: First cut at VIS simulator for Niagara. [SPARC64]: Fix system type in /proc/cpuinfo and remove bogus OBP check. [SPARC64]: Add SMT scheduling support for Niagara. [SPARC64]: Fix 32-bit truncation which broke sparsemem. [SPARC64]: Move over to sparsemem. [SPARC64]: Fix new context version SMP handling. ...
Diffstat (limited to 'arch/sparc64/mm/init.c')
-rw-r--r--arch/sparc64/mm/init.c1431
1 files changed, 789 insertions, 642 deletions
diff --git a/arch/sparc64/mm/init.c b/arch/sparc64/mm/init.c
index 1e44ee26cee8..c2b556106fc1 100644
--- a/arch/sparc64/mm/init.c
+++ b/arch/sparc64/mm/init.c
@@ -6,6 +6,7 @@
6 */ 6 */
7 7
8#include <linux/config.h> 8#include <linux/config.h>
9#include <linux/module.h>
9#include <linux/kernel.h> 10#include <linux/kernel.h>
10#include <linux/sched.h> 11#include <linux/sched.h>
11#include <linux/string.h> 12#include <linux/string.h>
@@ -39,9 +40,27 @@
39#include <asm/tlb.h> 40#include <asm/tlb.h>
40#include <asm/spitfire.h> 41#include <asm/spitfire.h>
41#include <asm/sections.h> 42#include <asm/sections.h>
43#include <asm/tsb.h>
44#include <asm/hypervisor.h>
42 45
43extern void device_scan(void); 46extern void device_scan(void);
44 47
48#define MAX_PHYS_ADDRESS (1UL << 42UL)
49#define KPTE_BITMAP_CHUNK_SZ (256UL * 1024UL * 1024UL)
50#define KPTE_BITMAP_BYTES \
51 ((MAX_PHYS_ADDRESS / KPTE_BITMAP_CHUNK_SZ) / 8)
52
53unsigned long kern_linear_pte_xor[2] __read_mostly;
54
55/* A bitmap, one bit for every 256MB of physical memory. If the bit
56 * is clear, we should use a 4MB page (via kern_linear_pte_xor[0]) else
57 * if set we should use a 256MB page (via kern_linear_pte_xor[1]).
58 */
59unsigned long kpte_linear_bitmap[KPTE_BITMAP_BYTES / sizeof(unsigned long)];
60
61/* A special kernel TSB for 4MB and 256MB linear mappings. */
62struct tsb swapper_4m_tsb[KERNEL_TSB4M_NENTRIES];
63
45#define MAX_BANKS 32 64#define MAX_BANKS 32
46 65
47static struct linux_prom64_registers pavail[MAX_BANKS] __initdata; 66static struct linux_prom64_registers pavail[MAX_BANKS] __initdata;
@@ -111,11 +130,9 @@ static void __init read_obp_memory(const char *property,
111 130
112unsigned long *sparc64_valid_addr_bitmap __read_mostly; 131unsigned long *sparc64_valid_addr_bitmap __read_mostly;
113 132
114/* Ugly, but necessary... -DaveM */ 133/* Kernel physical address base and size in bytes. */
115unsigned long phys_base __read_mostly;
116unsigned long kern_base __read_mostly; 134unsigned long kern_base __read_mostly;
117unsigned long kern_size __read_mostly; 135unsigned long kern_size __read_mostly;
118unsigned long pfn_base __read_mostly;
119 136
120/* get_new_mmu_context() uses "cache + 1". */ 137/* get_new_mmu_context() uses "cache + 1". */
121DEFINE_SPINLOCK(ctx_alloc_lock); 138DEFINE_SPINLOCK(ctx_alloc_lock);
@@ -141,24 +158,28 @@ unsigned long sparc64_kern_sec_context __read_mostly;
141 158
142int bigkernel = 0; 159int bigkernel = 0;
143 160
144/* XXX Tune this... */ 161kmem_cache_t *pgtable_cache __read_mostly;
145#define PGT_CACHE_LOW 25 162
146#define PGT_CACHE_HIGH 50 163static void zero_ctor(void *addr, kmem_cache_t *cache, unsigned long flags)
164{
165 clear_page(addr);
166}
167
168extern void tsb_cache_init(void);
147 169
148void check_pgt_cache(void) 170void pgtable_cache_init(void)
149{ 171{
150 preempt_disable(); 172 pgtable_cache = kmem_cache_create("pgtable_cache",
151 if (pgtable_cache_size > PGT_CACHE_HIGH) { 173 PAGE_SIZE, PAGE_SIZE,
152 do { 174 SLAB_HWCACHE_ALIGN |
153 if (pgd_quicklist) 175 SLAB_MUST_HWCACHE_ALIGN,
154 free_pgd_slow(get_pgd_fast()); 176 zero_ctor,
155 if (pte_quicklist[0]) 177 NULL);
156 free_pte_slow(pte_alloc_one_fast(NULL, 0)); 178 if (!pgtable_cache) {
157 if (pte_quicklist[1]) 179 prom_printf("Could not create pgtable_cache\n");
158 free_pte_slow(pte_alloc_one_fast(NULL, 1 << (PAGE_SHIFT + 10))); 180 prom_halt();
159 } while (pgtable_cache_size > PGT_CACHE_LOW);
160 } 181 }
161 preempt_enable(); 182 tsb_cache_init();
162} 183}
163 184
164#ifdef CONFIG_DEBUG_DCFLUSH 185#ifdef CONFIG_DEBUG_DCFLUSH
@@ -168,8 +189,9 @@ atomic_t dcpage_flushes_xcall = ATOMIC_INIT(0);
168#endif 189#endif
169#endif 190#endif
170 191
171__inline__ void flush_dcache_page_impl(struct page *page) 192inline void flush_dcache_page_impl(struct page *page)
172{ 193{
194 BUG_ON(tlb_type == hypervisor);
173#ifdef CONFIG_DEBUG_DCFLUSH 195#ifdef CONFIG_DEBUG_DCFLUSH
174 atomic_inc(&dcpage_flushes); 196 atomic_inc(&dcpage_flushes);
175#endif 197#endif
@@ -186,8 +208,8 @@ __inline__ void flush_dcache_page_impl(struct page *page)
186} 208}
187 209
188#define PG_dcache_dirty PG_arch_1 210#define PG_dcache_dirty PG_arch_1
189#define PG_dcache_cpu_shift 24 211#define PG_dcache_cpu_shift 24UL
190#define PG_dcache_cpu_mask (256 - 1) 212#define PG_dcache_cpu_mask (256UL - 1UL)
191 213
192#if NR_CPUS > 256 214#if NR_CPUS > 256
193#error D-cache dirty tracking and thread_info->cpu need fixing for > 256 cpus 215#error D-cache dirty tracking and thread_info->cpu need fixing for > 256 cpus
@@ -243,32 +265,61 @@ static __inline__ void clear_dcache_dirty_cpu(struct page *page, unsigned long c
243 : "g1", "g7"); 265 : "g1", "g7");
244} 266}
245 267
268static inline void tsb_insert(struct tsb *ent, unsigned long tag, unsigned long pte)
269{
270 unsigned long tsb_addr = (unsigned long) ent;
271
272 if (tlb_type == cheetah_plus || tlb_type == hypervisor)
273 tsb_addr = __pa(tsb_addr);
274
275 __tsb_insert(tsb_addr, tag, pte);
276}
277
278unsigned long _PAGE_ALL_SZ_BITS __read_mostly;
279unsigned long _PAGE_SZBITS __read_mostly;
280
246void update_mmu_cache(struct vm_area_struct *vma, unsigned long address, pte_t pte) 281void update_mmu_cache(struct vm_area_struct *vma, unsigned long address, pte_t pte)
247{ 282{
248 struct page *page; 283 struct mm_struct *mm;
249 unsigned long pfn; 284 struct tsb *tsb;
250 unsigned long pg_flags; 285 unsigned long tag, flags;
251 286
252 pfn = pte_pfn(pte); 287 if (tlb_type != hypervisor) {
253 if (pfn_valid(pfn) && 288 unsigned long pfn = pte_pfn(pte);
254 (page = pfn_to_page(pfn), page_mapping(page)) && 289 unsigned long pg_flags;
255 ((pg_flags = page->flags) & (1UL << PG_dcache_dirty))) { 290 struct page *page;
256 int cpu = ((pg_flags >> PG_dcache_cpu_shift) & 291
257 PG_dcache_cpu_mask); 292 if (pfn_valid(pfn) &&
258 int this_cpu = get_cpu(); 293 (page = pfn_to_page(pfn), page_mapping(page)) &&
259 294 ((pg_flags = page->flags) & (1UL << PG_dcache_dirty))) {
260 /* This is just to optimize away some function calls 295 int cpu = ((pg_flags >> PG_dcache_cpu_shift) &
261 * in the SMP case. 296 PG_dcache_cpu_mask);
262 */ 297 int this_cpu = get_cpu();
263 if (cpu == this_cpu) 298
264 flush_dcache_page_impl(page); 299 /* This is just to optimize away some function calls
265 else 300 * in the SMP case.
266 smp_flush_dcache_page_impl(page, cpu); 301 */
302 if (cpu == this_cpu)
303 flush_dcache_page_impl(page);
304 else
305 smp_flush_dcache_page_impl(page, cpu);
267 306
268 clear_dcache_dirty_cpu(page, cpu); 307 clear_dcache_dirty_cpu(page, cpu);
269 308
270 put_cpu(); 309 put_cpu();
310 }
271 } 311 }
312
313 mm = vma->vm_mm;
314
315 spin_lock_irqsave(&mm->context.lock, flags);
316
317 tsb = &mm->context.tsb[(address >> PAGE_SHIFT) &
318 (mm->context.tsb_nentries - 1UL)];
319 tag = (address >> 22UL);
320 tsb_insert(tsb, tag, pte_val(pte));
321
322 spin_unlock_irqrestore(&mm->context.lock, flags);
272} 323}
273 324
274void flush_dcache_page(struct page *page) 325void flush_dcache_page(struct page *page)
@@ -276,6 +327,9 @@ void flush_dcache_page(struct page *page)
276 struct address_space *mapping; 327 struct address_space *mapping;
277 int this_cpu; 328 int this_cpu;
278 329
330 if (tlb_type == hypervisor)
331 return;
332
279 /* Do not bother with the expensive D-cache flush if it 333 /* Do not bother with the expensive D-cache flush if it
280 * is merely the zero page. The 'bigcore' testcase in GDB 334 * is merely the zero page. The 'bigcore' testcase in GDB
281 * causes this case to run millions of times. 335 * causes this case to run millions of times.
@@ -311,7 +365,7 @@ out:
311 365
312void __kprobes flush_icache_range(unsigned long start, unsigned long end) 366void __kprobes flush_icache_range(unsigned long start, unsigned long end)
313{ 367{
314 /* Cheetah has coherent I-cache. */ 368 /* Cheetah and Hypervisor platform cpus have coherent I-cache. */
315 if (tlb_type == spitfire) { 369 if (tlb_type == spitfire) {
316 unsigned long kaddr; 370 unsigned long kaddr;
317 371
@@ -320,16 +374,6 @@ void __kprobes flush_icache_range(unsigned long start, unsigned long end)
320 } 374 }
321} 375}
322 376
323unsigned long page_to_pfn(struct page *page)
324{
325 return (unsigned long) ((page - mem_map) + pfn_base);
326}
327
328struct page *pfn_to_page(unsigned long pfn)
329{
330 return (mem_map + (pfn - pfn_base));
331}
332
333void show_mem(void) 377void show_mem(void)
334{ 378{
335 printk("Mem-info:\n"); 379 printk("Mem-info:\n");
@@ -338,7 +382,6 @@ void show_mem(void)
338 nr_swap_pages << (PAGE_SHIFT-10)); 382 nr_swap_pages << (PAGE_SHIFT-10));
339 printk("%ld pages of RAM\n", num_physpages); 383 printk("%ld pages of RAM\n", num_physpages);
340 printk("%d free pages\n", nr_free_pages()); 384 printk("%d free pages\n", nr_free_pages());
341 printk("%d pages in page table cache\n",pgtable_cache_size);
342} 385}
343 386
344void mmu_info(struct seq_file *m) 387void mmu_info(struct seq_file *m)
@@ -349,6 +392,8 @@ void mmu_info(struct seq_file *m)
349 seq_printf(m, "MMU Type\t: Cheetah+\n"); 392 seq_printf(m, "MMU Type\t: Cheetah+\n");
350 else if (tlb_type == spitfire) 393 else if (tlb_type == spitfire)
351 seq_printf(m, "MMU Type\t: Spitfire\n"); 394 seq_printf(m, "MMU Type\t: Spitfire\n");
395 else if (tlb_type == hypervisor)
396 seq_printf(m, "MMU Type\t: Hypervisor (sun4v)\n");
352 else 397 else
353 seq_printf(m, "MMU Type\t: ???\n"); 398 seq_printf(m, "MMU Type\t: ???\n");
354 399
@@ -371,45 +416,13 @@ struct linux_prom_translation {
371/* Exported for kernel TLB miss handling in ktlb.S */ 416/* Exported for kernel TLB miss handling in ktlb.S */
372struct linux_prom_translation prom_trans[512] __read_mostly; 417struct linux_prom_translation prom_trans[512] __read_mostly;
373unsigned int prom_trans_ents __read_mostly; 418unsigned int prom_trans_ents __read_mostly;
374unsigned int swapper_pgd_zero __read_mostly;
375
376extern unsigned long prom_boot_page;
377extern void prom_remap(unsigned long physpage, unsigned long virtpage, int mmu_ihandle);
378extern int prom_get_mmu_ihandle(void);
379extern void register_prom_callbacks(void);
380 419
381/* Exported for SMP bootup purposes. */ 420/* Exported for SMP bootup purposes. */
382unsigned long kern_locked_tte_data; 421unsigned long kern_locked_tte_data;
383 422
384/*
385 * Translate PROM's mapping we capture at boot time into physical address.
386 * The second parameter is only set from prom_callback() invocations.
387 */
388unsigned long prom_virt_to_phys(unsigned long promva, int *error)
389{
390 int i;
391
392 for (i = 0; i < prom_trans_ents; i++) {
393 struct linux_prom_translation *p = &prom_trans[i];
394
395 if (promva >= p->virt &&
396 promva < (p->virt + p->size)) {
397 unsigned long base = p->data & _PAGE_PADDR;
398
399 if (error)
400 *error = 0;
401 return base + (promva & (8192 - 1));
402 }
403 }
404 if (error)
405 *error = 1;
406 return 0UL;
407}
408
409/* The obp translations are saved based on 8k pagesize, since obp can 423/* The obp translations are saved based on 8k pagesize, since obp can
410 * use a mixture of pagesizes. Misses to the LOW_OBP_ADDRESS -> 424 * use a mixture of pagesizes. Misses to the LOW_OBP_ADDRESS ->
411 * HI_OBP_ADDRESS range are handled in ktlb.S and do not use the vpte 425 * HI_OBP_ADDRESS range are handled in ktlb.S.
412 * scheme (also, see rant in inherit_locked_prom_mappings()).
413 */ 426 */
414static inline int in_obp_range(unsigned long vaddr) 427static inline int in_obp_range(unsigned long vaddr)
415{ 428{
@@ -490,6 +503,36 @@ static void __init read_obp_translations(void)
490 } 503 }
491} 504}
492 505
506static void __init hypervisor_tlb_lock(unsigned long vaddr,
507 unsigned long pte,
508 unsigned long mmu)
509{
510 register unsigned long func asm("%o5");
511 register unsigned long arg0 asm("%o0");
512 register unsigned long arg1 asm("%o1");
513 register unsigned long arg2 asm("%o2");
514 register unsigned long arg3 asm("%o3");
515
516 func = HV_FAST_MMU_MAP_PERM_ADDR;
517 arg0 = vaddr;
518 arg1 = 0;
519 arg2 = pte;
520 arg3 = mmu;
521 __asm__ __volatile__("ta 0x80"
522 : "=&r" (func), "=&r" (arg0),
523 "=&r" (arg1), "=&r" (arg2),
524 "=&r" (arg3)
525 : "0" (func), "1" (arg0), "2" (arg1),
526 "3" (arg2), "4" (arg3));
527 if (arg0 != 0) {
528 prom_printf("hypervisor_tlb_lock[%lx:%lx:%lx:%lx]: "
529 "errors with %lx\n", vaddr, 0, pte, mmu, arg0);
530 prom_halt();
531 }
532}
533
534static unsigned long kern_large_tte(unsigned long paddr);
535
493static void __init remap_kernel(void) 536static void __init remap_kernel(void)
494{ 537{
495 unsigned long phys_page, tte_vaddr, tte_data; 538 unsigned long phys_page, tte_vaddr, tte_data;
@@ -497,25 +540,34 @@ static void __init remap_kernel(void)
497 540
498 tte_vaddr = (unsigned long) KERNBASE; 541 tte_vaddr = (unsigned long) KERNBASE;
499 phys_page = (prom_boot_mapping_phys_low >> 22UL) << 22UL; 542 phys_page = (prom_boot_mapping_phys_low >> 22UL) << 22UL;
500 tte_data = (phys_page | (_PAGE_VALID | _PAGE_SZ4MB | 543 tte_data = kern_large_tte(phys_page);
501 _PAGE_CP | _PAGE_CV | _PAGE_P |
502 _PAGE_L | _PAGE_W));
503 544
504 kern_locked_tte_data = tte_data; 545 kern_locked_tte_data = tte_data;
505 546
506 /* Now lock us into the TLBs via OBP. */ 547 /* Now lock us into the TLBs via Hypervisor or OBP. */
507 prom_dtlb_load(tlb_ent, tte_data, tte_vaddr); 548 if (tlb_type == hypervisor) {
508 prom_itlb_load(tlb_ent, tte_data, tte_vaddr); 549 hypervisor_tlb_lock(tte_vaddr, tte_data, HV_MMU_DMMU);
509 if (bigkernel) { 550 hypervisor_tlb_lock(tte_vaddr, tte_data, HV_MMU_IMMU);
510 tlb_ent -= 1; 551 if (bigkernel) {
511 prom_dtlb_load(tlb_ent, 552 tte_vaddr += 0x400000;
512 tte_data + 0x400000, 553 tte_data += 0x400000;
513 tte_vaddr + 0x400000); 554 hypervisor_tlb_lock(tte_vaddr, tte_data, HV_MMU_DMMU);
514 prom_itlb_load(tlb_ent, 555 hypervisor_tlb_lock(tte_vaddr, tte_data, HV_MMU_IMMU);
515 tte_data + 0x400000, 556 }
516 tte_vaddr + 0x400000); 557 } else {
558 prom_dtlb_load(tlb_ent, tte_data, tte_vaddr);
559 prom_itlb_load(tlb_ent, tte_data, tte_vaddr);
560 if (bigkernel) {
561 tlb_ent -= 1;
562 prom_dtlb_load(tlb_ent,
563 tte_data + 0x400000,
564 tte_vaddr + 0x400000);
565 prom_itlb_load(tlb_ent,
566 tte_data + 0x400000,
567 tte_vaddr + 0x400000);
568 }
569 sparc64_highest_unlocked_tlb_ent = tlb_ent - 1;
517 } 570 }
518 sparc64_highest_unlocked_tlb_ent = tlb_ent - 1;
519 if (tlb_type == cheetah_plus) { 571 if (tlb_type == cheetah_plus) {
520 sparc64_kern_pri_context = (CTX_CHEETAH_PLUS_CTX0 | 572 sparc64_kern_pri_context = (CTX_CHEETAH_PLUS_CTX0 |
521 CTX_CHEETAH_PLUS_NUC); 573 CTX_CHEETAH_PLUS_NUC);
@@ -533,372 +585,14 @@ static void __init inherit_prom_mappings(void)
533 prom_printf("Remapping the kernel... "); 585 prom_printf("Remapping the kernel... ");
534 remap_kernel(); 586 remap_kernel();
535 prom_printf("done.\n"); 587 prom_printf("done.\n");
536
537 prom_printf("Registering callbacks... ");
538 register_prom_callbacks();
539 prom_printf("done.\n");
540}
541
542/* The OBP specifications for sun4u mark 0xfffffffc00000000 and
543 * upwards as reserved for use by the firmware (I wonder if this
544 * will be the same on Cheetah...). We use this virtual address
545 * range for the VPTE table mappings of the nucleus so we need
546 * to zap them when we enter the PROM. -DaveM
547 */
548static void __flush_nucleus_vptes(void)
549{
550 unsigned long prom_reserved_base = 0xfffffffc00000000UL;
551 int i;
552
553 /* Only DTLB must be checked for VPTE entries. */
554 if (tlb_type == spitfire) {
555 for (i = 0; i < 63; i++) {
556 unsigned long tag;
557
558 /* Spitfire Errata #32 workaround */
559 /* NOTE: Always runs on spitfire, so no cheetah+
560 * page size encodings.
561 */
562 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
563 "flush %%g6"
564 : /* No outputs */
565 : "r" (0),
566 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
567
568 tag = spitfire_get_dtlb_tag(i);
569 if (((tag & ~(PAGE_MASK)) == 0) &&
570 ((tag & (PAGE_MASK)) >= prom_reserved_base)) {
571 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
572 "membar #Sync"
573 : /* no outputs */
574 : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
575 spitfire_put_dtlb_data(i, 0x0UL);
576 }
577 }
578 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
579 for (i = 0; i < 512; i++) {
580 unsigned long tag = cheetah_get_dtlb_tag(i, 2);
581
582 if ((tag & ~PAGE_MASK) == 0 &&
583 (tag & PAGE_MASK) >= prom_reserved_base) {
584 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
585 "membar #Sync"
586 : /* no outputs */
587 : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
588 cheetah_put_dtlb_data(i, 0x0UL, 2);
589 }
590
591 if (tlb_type != cheetah_plus)
592 continue;
593
594 tag = cheetah_get_dtlb_tag(i, 3);
595
596 if ((tag & ~PAGE_MASK) == 0 &&
597 (tag & PAGE_MASK) >= prom_reserved_base) {
598 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
599 "membar #Sync"
600 : /* no outputs */
601 : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
602 cheetah_put_dtlb_data(i, 0x0UL, 3);
603 }
604 }
605 } else {
606 /* Implement me :-) */
607 BUG();
608 }
609} 588}
610 589
611static int prom_ditlb_set;
612struct prom_tlb_entry {
613 int tlb_ent;
614 unsigned long tlb_tag;
615 unsigned long tlb_data;
616};
617struct prom_tlb_entry prom_itlb[16], prom_dtlb[16];
618
619void prom_world(int enter) 590void prom_world(int enter)
620{ 591{
621 unsigned long pstate;
622 int i;
623
624 if (!enter) 592 if (!enter)
625 set_fs((mm_segment_t) { get_thread_current_ds() }); 593 set_fs((mm_segment_t) { get_thread_current_ds() });
626 594
627 if (!prom_ditlb_set) 595 __asm__ __volatile__("flushw");
628 return;
629
630 /* Make sure the following runs atomically. */
631 __asm__ __volatile__("flushw\n\t"
632 "rdpr %%pstate, %0\n\t"
633 "wrpr %0, %1, %%pstate"
634 : "=r" (pstate)
635 : "i" (PSTATE_IE));
636
637 if (enter) {
638 /* Kick out nucleus VPTEs. */
639 __flush_nucleus_vptes();
640
641 /* Install PROM world. */
642 for (i = 0; i < 16; i++) {
643 if (prom_dtlb[i].tlb_ent != -1) {
644 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
645 "membar #Sync"
646 : : "r" (prom_dtlb[i].tlb_tag), "r" (TLB_TAG_ACCESS),
647 "i" (ASI_DMMU));
648 if (tlb_type == spitfire)
649 spitfire_put_dtlb_data(prom_dtlb[i].tlb_ent,
650 prom_dtlb[i].tlb_data);
651 else if (tlb_type == cheetah || tlb_type == cheetah_plus)
652 cheetah_put_ldtlb_data(prom_dtlb[i].tlb_ent,
653 prom_dtlb[i].tlb_data);
654 }
655 if (prom_itlb[i].tlb_ent != -1) {
656 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
657 "membar #Sync"
658 : : "r" (prom_itlb[i].tlb_tag),
659 "r" (TLB_TAG_ACCESS),
660 "i" (ASI_IMMU));
661 if (tlb_type == spitfire)
662 spitfire_put_itlb_data(prom_itlb[i].tlb_ent,
663 prom_itlb[i].tlb_data);
664 else if (tlb_type == cheetah || tlb_type == cheetah_plus)
665 cheetah_put_litlb_data(prom_itlb[i].tlb_ent,
666 prom_itlb[i].tlb_data);
667 }
668 }
669 } else {
670 for (i = 0; i < 16; i++) {
671 if (prom_dtlb[i].tlb_ent != -1) {
672 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
673 "membar #Sync"
674 : : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
675 if (tlb_type == spitfire)
676 spitfire_put_dtlb_data(prom_dtlb[i].tlb_ent, 0x0UL);
677 else
678 cheetah_put_ldtlb_data(prom_dtlb[i].tlb_ent, 0x0UL);
679 }
680 if (prom_itlb[i].tlb_ent != -1) {
681 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
682 "membar #Sync"
683 : : "r" (TLB_TAG_ACCESS),
684 "i" (ASI_IMMU));
685 if (tlb_type == spitfire)
686 spitfire_put_itlb_data(prom_itlb[i].tlb_ent, 0x0UL);
687 else
688 cheetah_put_litlb_data(prom_itlb[i].tlb_ent, 0x0UL);
689 }
690 }
691 }
692 __asm__ __volatile__("wrpr %0, 0, %%pstate"
693 : : "r" (pstate));
694}
695
696void inherit_locked_prom_mappings(int save_p)
697{
698 int i;
699 int dtlb_seen = 0;
700 int itlb_seen = 0;
701
702 /* Fucking losing PROM has more mappings in the TLB, but
703 * it (conveniently) fails to mention any of these in the
704 * translations property. The only ones that matter are
705 * the locked PROM tlb entries, so we impose the following
706 * irrecovable rule on the PROM, it is allowed 8 locked
707 * entries in the ITLB and 8 in the DTLB.
708 *
709 * Supposedly the upper 16GB of the address space is
710 * reserved for OBP, BUT I WISH THIS WAS DOCUMENTED
711 * SOMEWHERE!!!!!!!!!!!!!!!!! Furthermore the entire interface
712 * used between the client program and the firmware on sun5
713 * systems to coordinate mmu mappings is also COMPLETELY
714 * UNDOCUMENTED!!!!!! Thanks S(t)un!
715 */
716 if (save_p) {
717 for (i = 0; i < 16; i++) {
718 prom_itlb[i].tlb_ent = -1;
719 prom_dtlb[i].tlb_ent = -1;
720 }
721 }
722 if (tlb_type == spitfire) {
723 int high = sparc64_highest_unlocked_tlb_ent;
724 for (i = 0; i <= high; i++) {
725 unsigned long data;
726
727 /* Spitfire Errata #32 workaround */
728 /* NOTE: Always runs on spitfire, so no cheetah+
729 * page size encodings.
730 */
731 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
732 "flush %%g6"
733 : /* No outputs */
734 : "r" (0),
735 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
736
737 data = spitfire_get_dtlb_data(i);
738 if ((data & (_PAGE_L|_PAGE_VALID)) == (_PAGE_L|_PAGE_VALID)) {
739 unsigned long tag;
740
741 /* Spitfire Errata #32 workaround */
742 /* NOTE: Always runs on spitfire, so no
743 * cheetah+ page size encodings.
744 */
745 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
746 "flush %%g6"
747 : /* No outputs */
748 : "r" (0),
749 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
750
751 tag = spitfire_get_dtlb_tag(i);
752 if (save_p) {
753 prom_dtlb[dtlb_seen].tlb_ent = i;
754 prom_dtlb[dtlb_seen].tlb_tag = tag;
755 prom_dtlb[dtlb_seen].tlb_data = data;
756 }
757 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
758 "membar #Sync"
759 : : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
760 spitfire_put_dtlb_data(i, 0x0UL);
761
762 dtlb_seen++;
763 if (dtlb_seen > 15)
764 break;
765 }
766 }
767
768 for (i = 0; i < high; i++) {
769 unsigned long data;
770
771 /* Spitfire Errata #32 workaround */
772 /* NOTE: Always runs on spitfire, so no
773 * cheetah+ page size encodings.
774 */
775 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
776 "flush %%g6"
777 : /* No outputs */
778 : "r" (0),
779 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
780
781 data = spitfire_get_itlb_data(i);
782 if ((data & (_PAGE_L|_PAGE_VALID)) == (_PAGE_L|_PAGE_VALID)) {
783 unsigned long tag;
784
785 /* Spitfire Errata #32 workaround */
786 /* NOTE: Always runs on spitfire, so no
787 * cheetah+ page size encodings.
788 */
789 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
790 "flush %%g6"
791 : /* No outputs */
792 : "r" (0),
793 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
794
795 tag = spitfire_get_itlb_tag(i);
796 if (save_p) {
797 prom_itlb[itlb_seen].tlb_ent = i;
798 prom_itlb[itlb_seen].tlb_tag = tag;
799 prom_itlb[itlb_seen].tlb_data = data;
800 }
801 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
802 "membar #Sync"
803 : : "r" (TLB_TAG_ACCESS), "i" (ASI_IMMU));
804 spitfire_put_itlb_data(i, 0x0UL);
805
806 itlb_seen++;
807 if (itlb_seen > 15)
808 break;
809 }
810 }
811 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
812 int high = sparc64_highest_unlocked_tlb_ent;
813
814 for (i = 0; i <= high; i++) {
815 unsigned long data;
816
817 data = cheetah_get_ldtlb_data(i);
818 if ((data & (_PAGE_L|_PAGE_VALID)) == (_PAGE_L|_PAGE_VALID)) {
819 unsigned long tag;
820
821 tag = cheetah_get_ldtlb_tag(i);
822 if (save_p) {
823 prom_dtlb[dtlb_seen].tlb_ent = i;
824 prom_dtlb[dtlb_seen].tlb_tag = tag;
825 prom_dtlb[dtlb_seen].tlb_data = data;
826 }
827 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
828 "membar #Sync"
829 : : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
830 cheetah_put_ldtlb_data(i, 0x0UL);
831
832 dtlb_seen++;
833 if (dtlb_seen > 15)
834 break;
835 }
836 }
837
838 for (i = 0; i < high; i++) {
839 unsigned long data;
840
841 data = cheetah_get_litlb_data(i);
842 if ((data & (_PAGE_L|_PAGE_VALID)) == (_PAGE_L|_PAGE_VALID)) {
843 unsigned long tag;
844
845 tag = cheetah_get_litlb_tag(i);
846 if (save_p) {
847 prom_itlb[itlb_seen].tlb_ent = i;
848 prom_itlb[itlb_seen].tlb_tag = tag;
849 prom_itlb[itlb_seen].tlb_data = data;
850 }
851 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
852 "membar #Sync"
853 : : "r" (TLB_TAG_ACCESS), "i" (ASI_IMMU));
854 cheetah_put_litlb_data(i, 0x0UL);
855
856 itlb_seen++;
857 if (itlb_seen > 15)
858 break;
859 }
860 }
861 } else {
862 /* Implement me :-) */
863 BUG();
864 }
865 if (save_p)
866 prom_ditlb_set = 1;
867}
868
869/* Give PROM back his world, done during reboots... */
870void prom_reload_locked(void)
871{
872 int i;
873
874 for (i = 0; i < 16; i++) {
875 if (prom_dtlb[i].tlb_ent != -1) {
876 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
877 "membar #Sync"
878 : : "r" (prom_dtlb[i].tlb_tag), "r" (TLB_TAG_ACCESS),
879 "i" (ASI_DMMU));
880 if (tlb_type == spitfire)
881 spitfire_put_dtlb_data(prom_dtlb[i].tlb_ent,
882 prom_dtlb[i].tlb_data);
883 else if (tlb_type == cheetah || tlb_type == cheetah_plus)
884 cheetah_put_ldtlb_data(prom_dtlb[i].tlb_ent,
885 prom_dtlb[i].tlb_data);
886 }
887
888 if (prom_itlb[i].tlb_ent != -1) {
889 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
890 "membar #Sync"
891 : : "r" (prom_itlb[i].tlb_tag),
892 "r" (TLB_TAG_ACCESS),
893 "i" (ASI_IMMU));
894 if (tlb_type == spitfire)
895 spitfire_put_itlb_data(prom_itlb[i].tlb_ent,
896 prom_itlb[i].tlb_data);
897 else
898 cheetah_put_litlb_data(prom_itlb[i].tlb_ent,
899 prom_itlb[i].tlb_data);
900 }
901 }
902} 596}
903 597
904#ifdef DCACHE_ALIASING_POSSIBLE 598#ifdef DCACHE_ALIASING_POSSIBLE
@@ -914,7 +608,7 @@ void __flush_dcache_range(unsigned long start, unsigned long end)
914 if (++n >= 512) 608 if (++n >= 512)
915 break; 609 break;
916 } 610 }
917 } else { 611 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
918 start = __pa(start); 612 start = __pa(start);
919 end = __pa(end); 613 end = __pa(end);
920 for (va = start; va < end; va += 32) 614 for (va = start; va < end; va += 32)
@@ -927,63 +621,6 @@ void __flush_dcache_range(unsigned long start, unsigned long end)
927} 621}
928#endif /* DCACHE_ALIASING_POSSIBLE */ 622#endif /* DCACHE_ALIASING_POSSIBLE */
929 623
930/* If not locked, zap it. */
931void __flush_tlb_all(void)
932{
933 unsigned long pstate;
934 int i;
935
936 __asm__ __volatile__("flushw\n\t"
937 "rdpr %%pstate, %0\n\t"
938 "wrpr %0, %1, %%pstate"
939 : "=r" (pstate)
940 : "i" (PSTATE_IE));
941 if (tlb_type == spitfire) {
942 for (i = 0; i < 64; i++) {
943 /* Spitfire Errata #32 workaround */
944 /* NOTE: Always runs on spitfire, so no
945 * cheetah+ page size encodings.
946 */
947 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
948 "flush %%g6"
949 : /* No outputs */
950 : "r" (0),
951 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
952
953 if (!(spitfire_get_dtlb_data(i) & _PAGE_L)) {
954 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
955 "membar #Sync"
956 : /* no outputs */
957 : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
958 spitfire_put_dtlb_data(i, 0x0UL);
959 }
960
961 /* Spitfire Errata #32 workaround */
962 /* NOTE: Always runs on spitfire, so no
963 * cheetah+ page size encodings.
964 */
965 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
966 "flush %%g6"
967 : /* No outputs */
968 : "r" (0),
969 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
970
971 if (!(spitfire_get_itlb_data(i) & _PAGE_L)) {
972 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
973 "membar #Sync"
974 : /* no outputs */
975 : "r" (TLB_TAG_ACCESS), "i" (ASI_IMMU));
976 spitfire_put_itlb_data(i, 0x0UL);
977 }
978 }
979 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
980 cheetah_flush_dtlb_all();
981 cheetah_flush_itlb_all();
982 }
983 __asm__ __volatile__("wrpr %0, 0, %%pstate"
984 : : "r" (pstate));
985}
986
987/* Caller does TLB context flushing on local CPU if necessary. 624/* Caller does TLB context flushing on local CPU if necessary.
988 * The caller also ensures that CTX_VALID(mm->context) is false. 625 * The caller also ensures that CTX_VALID(mm->context) is false.
989 * 626 *
@@ -991,17 +628,21 @@ void __flush_tlb_all(void)
991 * let the user have CTX 0 (nucleus) or we ever use a CTX 628 * let the user have CTX 0 (nucleus) or we ever use a CTX
992 * version of zero (and thus NO_CONTEXT would not be caught 629 * version of zero (and thus NO_CONTEXT would not be caught
993 * by version mis-match tests in mmu_context.h). 630 * by version mis-match tests in mmu_context.h).
631 *
632 * Always invoked with interrupts disabled.
994 */ 633 */
995void get_new_mmu_context(struct mm_struct *mm) 634void get_new_mmu_context(struct mm_struct *mm)
996{ 635{
997 unsigned long ctx, new_ctx; 636 unsigned long ctx, new_ctx;
998 unsigned long orig_pgsz_bits; 637 unsigned long orig_pgsz_bits;
999 638 unsigned long flags;
639 int new_version;
1000 640
1001 spin_lock(&ctx_alloc_lock); 641 spin_lock_irqsave(&ctx_alloc_lock, flags);
1002 orig_pgsz_bits = (mm->context.sparc64_ctx_val & CTX_PGSZ_MASK); 642 orig_pgsz_bits = (mm->context.sparc64_ctx_val & CTX_PGSZ_MASK);
1003 ctx = (tlb_context_cache + 1) & CTX_NR_MASK; 643 ctx = (tlb_context_cache + 1) & CTX_NR_MASK;
1004 new_ctx = find_next_zero_bit(mmu_context_bmap, 1 << CTX_NR_BITS, ctx); 644 new_ctx = find_next_zero_bit(mmu_context_bmap, 1 << CTX_NR_BITS, ctx);
645 new_version = 0;
1005 if (new_ctx >= (1 << CTX_NR_BITS)) { 646 if (new_ctx >= (1 << CTX_NR_BITS)) {
1006 new_ctx = find_next_zero_bit(mmu_context_bmap, ctx, 1); 647 new_ctx = find_next_zero_bit(mmu_context_bmap, ctx, 1);
1007 if (new_ctx >= ctx) { 648 if (new_ctx >= ctx) {
@@ -1024,6 +665,7 @@ void get_new_mmu_context(struct mm_struct *mm)
1024 mmu_context_bmap[i + 2] = 0; 665 mmu_context_bmap[i + 2] = 0;
1025 mmu_context_bmap[i + 3] = 0; 666 mmu_context_bmap[i + 3] = 0;
1026 } 667 }
668 new_version = 1;
1027 goto out; 669 goto out;
1028 } 670 }
1029 } 671 }
@@ -1032,79 +674,10 @@ void get_new_mmu_context(struct mm_struct *mm)
1032out: 674out:
1033 tlb_context_cache = new_ctx; 675 tlb_context_cache = new_ctx;
1034 mm->context.sparc64_ctx_val = new_ctx | orig_pgsz_bits; 676 mm->context.sparc64_ctx_val = new_ctx | orig_pgsz_bits;
1035 spin_unlock(&ctx_alloc_lock); 677 spin_unlock_irqrestore(&ctx_alloc_lock, flags);
1036}
1037
1038#ifndef CONFIG_SMP
1039struct pgtable_cache_struct pgt_quicklists;
1040#endif
1041
1042/* OK, we have to color these pages. The page tables are accessed
1043 * by non-Dcache enabled mapping in the VPTE area by the dtlb_backend.S
1044 * code, as well as by PAGE_OFFSET range direct-mapped addresses by
1045 * other parts of the kernel. By coloring, we make sure that the tlbmiss
1046 * fast handlers do not get data from old/garbage dcache lines that
1047 * correspond to an old/stale virtual address (user/kernel) that
1048 * previously mapped the pagetable page while accessing vpte range
1049 * addresses. The idea is that if the vpte color and PAGE_OFFSET range
1050 * color is the same, then when the kernel initializes the pagetable
1051 * using the later address range, accesses with the first address
1052 * range will see the newly initialized data rather than the garbage.
1053 */
1054#ifdef DCACHE_ALIASING_POSSIBLE
1055#define DC_ALIAS_SHIFT 1
1056#else
1057#define DC_ALIAS_SHIFT 0
1058#endif
1059pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
1060{
1061 struct page *page;
1062 unsigned long color;
1063
1064 {
1065 pte_t *ptep = pte_alloc_one_fast(mm, address);
1066
1067 if (ptep)
1068 return ptep;
1069 }
1070 678
1071 color = VPTE_COLOR(address); 679 if (unlikely(new_version))
1072 page = alloc_pages(GFP_KERNEL|__GFP_REPEAT, DC_ALIAS_SHIFT); 680 smp_new_mmu_context_version();
1073 if (page) {
1074 unsigned long *to_free;
1075 unsigned long paddr;
1076 pte_t *pte;
1077
1078#ifdef DCACHE_ALIASING_POSSIBLE
1079 set_page_count(page, 1);
1080 ClearPageCompound(page);
1081
1082 set_page_count((page + 1), 1);
1083 ClearPageCompound(page + 1);
1084#endif
1085 paddr = (unsigned long) page_address(page);
1086 memset((char *)paddr, 0, (PAGE_SIZE << DC_ALIAS_SHIFT));
1087
1088 if (!color) {
1089 pte = (pte_t *) paddr;
1090 to_free = (unsigned long *) (paddr + PAGE_SIZE);
1091 } else {
1092 pte = (pte_t *) (paddr + PAGE_SIZE);
1093 to_free = (unsigned long *) paddr;
1094 }
1095
1096#ifdef DCACHE_ALIASING_POSSIBLE
1097 /* Now free the other one up, adjust cache size. */
1098 preempt_disable();
1099 *to_free = (unsigned long) pte_quicklist[color ^ 0x1];
1100 pte_quicklist[color ^ 0x1] = to_free;
1101 pgtable_cache_size++;
1102 preempt_enable();
1103#endif
1104
1105 return pte;
1106 }
1107 return NULL;
1108} 681}
1109 682
1110void sparc_ultra_dump_itlb(void) 683void sparc_ultra_dump_itlb(void)
@@ -1196,9 +769,78 @@ void sparc_ultra_dump_dtlb(void)
1196 769
1197extern unsigned long cmdline_memory_size; 770extern unsigned long cmdline_memory_size;
1198 771
1199unsigned long __init bootmem_init(unsigned long *pages_avail) 772/* Find a free area for the bootmem map, avoiding the kernel image
773 * and the initial ramdisk.
774 */
775static unsigned long __init choose_bootmap_pfn(unsigned long start_pfn,
776 unsigned long end_pfn)
777{
778 unsigned long avoid_start, avoid_end, bootmap_size;
779 int i;
780
781 bootmap_size = ((end_pfn - start_pfn) + 7) / 8;
782 bootmap_size = ALIGN(bootmap_size, sizeof(long));
783
784 avoid_start = avoid_end = 0;
785#ifdef CONFIG_BLK_DEV_INITRD
786 avoid_start = initrd_start;
787 avoid_end = PAGE_ALIGN(initrd_end);
788#endif
789
790#ifdef CONFIG_DEBUG_BOOTMEM
791 prom_printf("choose_bootmap_pfn: kern[%lx:%lx] avoid[%lx:%lx]\n",
792 kern_base, PAGE_ALIGN(kern_base + kern_size),
793 avoid_start, avoid_end);
794#endif
795 for (i = 0; i < pavail_ents; i++) {
796 unsigned long start, end;
797
798 start = pavail[i].phys_addr;
799 end = start + pavail[i].reg_size;
800
801 while (start < end) {
802 if (start >= kern_base &&
803 start < PAGE_ALIGN(kern_base + kern_size)) {
804 start = PAGE_ALIGN(kern_base + kern_size);
805 continue;
806 }
807 if (start >= avoid_start && start < avoid_end) {
808 start = avoid_end;
809 continue;
810 }
811
812 if ((end - start) < bootmap_size)
813 break;
814
815 if (start < kern_base &&
816 (start + bootmap_size) > kern_base) {
817 start = PAGE_ALIGN(kern_base + kern_size);
818 continue;
819 }
820
821 if (start < avoid_start &&
822 (start + bootmap_size) > avoid_start) {
823 start = avoid_end;
824 continue;
825 }
826
827 /* OK, it doesn't overlap anything, use it. */
828#ifdef CONFIG_DEBUG_BOOTMEM
829 prom_printf("choose_bootmap_pfn: Using %lx [%lx]\n",
830 start >> PAGE_SHIFT, start);
831#endif
832 return start >> PAGE_SHIFT;
833 }
834 }
835
836 prom_printf("Cannot find free area for bootmap, aborting.\n");
837 prom_halt();
838}
839
840static unsigned long __init bootmem_init(unsigned long *pages_avail,
841 unsigned long phys_base)
1200{ 842{
1201 unsigned long bootmap_size, start_pfn, end_pfn; 843 unsigned long bootmap_size, end_pfn;
1202 unsigned long end_of_phys_memory = 0UL; 844 unsigned long end_of_phys_memory = 0UL;
1203 unsigned long bootmap_pfn, bytes_avail, size; 845 unsigned long bootmap_pfn, bytes_avail, size;
1204 int i; 846 int i;
@@ -1236,14 +878,6 @@ unsigned long __init bootmem_init(unsigned long *pages_avail)
1236 878
1237 *pages_avail = bytes_avail >> PAGE_SHIFT; 879 *pages_avail = bytes_avail >> PAGE_SHIFT;
1238 880
1239 /* Start with page aligned address of last symbol in kernel
1240 * image. The kernel is hard mapped below PAGE_OFFSET in a
1241 * 4MB locked TLB translation.
1242 */
1243 start_pfn = PAGE_ALIGN(kern_base + kern_size) >> PAGE_SHIFT;
1244
1245 bootmap_pfn = start_pfn;
1246
1247 end_pfn = end_of_phys_memory >> PAGE_SHIFT; 881 end_pfn = end_of_phys_memory >> PAGE_SHIFT;
1248 882
1249#ifdef CONFIG_BLK_DEV_INITRD 883#ifdef CONFIG_BLK_DEV_INITRD
@@ -1260,23 +894,22 @@ unsigned long __init bootmem_init(unsigned long *pages_avail)
1260 "(0x%016lx > 0x%016lx)\ndisabling initrd\n", 894 "(0x%016lx > 0x%016lx)\ndisabling initrd\n",
1261 initrd_end, end_of_phys_memory); 895 initrd_end, end_of_phys_memory);
1262 initrd_start = 0; 896 initrd_start = 0;
1263 } 897 initrd_end = 0;
1264 if (initrd_start) {
1265 if (initrd_start >= (start_pfn << PAGE_SHIFT) &&
1266 initrd_start < (start_pfn << PAGE_SHIFT) + 2 * PAGE_SIZE)
1267 bootmap_pfn = PAGE_ALIGN (initrd_end) >> PAGE_SHIFT;
1268 } 898 }
1269 } 899 }
1270#endif 900#endif
1271 /* Initialize the boot-time allocator. */ 901 /* Initialize the boot-time allocator. */
1272 max_pfn = max_low_pfn = end_pfn; 902 max_pfn = max_low_pfn = end_pfn;
1273 min_low_pfn = pfn_base; 903 min_low_pfn = (phys_base >> PAGE_SHIFT);
904
905 bootmap_pfn = choose_bootmap_pfn(min_low_pfn, end_pfn);
1274 906
1275#ifdef CONFIG_DEBUG_BOOTMEM 907#ifdef CONFIG_DEBUG_BOOTMEM
1276 prom_printf("init_bootmem(min[%lx], bootmap[%lx], max[%lx])\n", 908 prom_printf("init_bootmem(min[%lx], bootmap[%lx], max[%lx])\n",
1277 min_low_pfn, bootmap_pfn, max_low_pfn); 909 min_low_pfn, bootmap_pfn, max_low_pfn);
1278#endif 910#endif
1279 bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap_pfn, pfn_base, end_pfn); 911 bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap_pfn,
912 min_low_pfn, end_pfn);
1280 913
1281 /* Now register the available physical memory with the 914 /* Now register the available physical memory with the
1282 * allocator. 915 * allocator.
@@ -1324,9 +957,26 @@ unsigned long __init bootmem_init(unsigned long *pages_avail)
1324 reserve_bootmem((bootmap_pfn << PAGE_SHIFT), size); 957 reserve_bootmem((bootmap_pfn << PAGE_SHIFT), size);
1325 *pages_avail -= PAGE_ALIGN(size) >> PAGE_SHIFT; 958 *pages_avail -= PAGE_ALIGN(size) >> PAGE_SHIFT;
1326 959
960 for (i = 0; i < pavail_ents; i++) {
961 unsigned long start_pfn, end_pfn;
962
963 start_pfn = pavail[i].phys_addr >> PAGE_SHIFT;
964 end_pfn = (start_pfn + (pavail[i].reg_size >> PAGE_SHIFT));
965#ifdef CONFIG_DEBUG_BOOTMEM
966 prom_printf("memory_present(0, %lx, %lx)\n",
967 start_pfn, end_pfn);
968#endif
969 memory_present(0, start_pfn, end_pfn);
970 }
971
972 sparse_init();
973
1327 return end_pfn; 974 return end_pfn;
1328} 975}
1329 976
977static struct linux_prom64_registers pall[MAX_BANKS] __initdata;
978static int pall_ents __initdata;
979
1330#ifdef CONFIG_DEBUG_PAGEALLOC 980#ifdef CONFIG_DEBUG_PAGEALLOC
1331static unsigned long kernel_map_range(unsigned long pstart, unsigned long pend, pgprot_t prot) 981static unsigned long kernel_map_range(unsigned long pstart, unsigned long pend, pgprot_t prot)
1332{ 982{
@@ -1382,14 +1032,44 @@ static unsigned long kernel_map_range(unsigned long pstart, unsigned long pend,
1382 return alloc_bytes; 1032 return alloc_bytes;
1383} 1033}
1384 1034
1385static struct linux_prom64_registers pall[MAX_BANKS] __initdata;
1386static int pall_ents __initdata;
1387
1388extern unsigned int kvmap_linear_patch[1]; 1035extern unsigned int kvmap_linear_patch[1];
1036#endif /* CONFIG_DEBUG_PAGEALLOC */
1037
1038static void __init mark_kpte_bitmap(unsigned long start, unsigned long end)
1039{
1040 const unsigned long shift_256MB = 28;
1041 const unsigned long mask_256MB = ((1UL << shift_256MB) - 1UL);
1042 const unsigned long size_256MB = (1UL << shift_256MB);
1043
1044 while (start < end) {
1045 long remains;
1046
1047 remains = end - start;
1048 if (remains < size_256MB)
1049 break;
1050
1051 if (start & mask_256MB) {
1052 start = (start + size_256MB) & ~mask_256MB;
1053 continue;
1054 }
1055
1056 while (remains >= size_256MB) {
1057 unsigned long index = start >> shift_256MB;
1058
1059 __set_bit(index, kpte_linear_bitmap);
1060
1061 start += size_256MB;
1062 remains -= size_256MB;
1063 }
1064 }
1065}
1389 1066
1390static void __init kernel_physical_mapping_init(void) 1067static void __init kernel_physical_mapping_init(void)
1391{ 1068{
1392 unsigned long i, mem_alloced = 0UL; 1069 unsigned long i;
1070#ifdef CONFIG_DEBUG_PAGEALLOC
1071 unsigned long mem_alloced = 0UL;
1072#endif
1393 1073
1394 read_obp_memory("reg", &pall[0], &pall_ents); 1074 read_obp_memory("reg", &pall[0], &pall_ents);
1395 1075
@@ -1398,10 +1078,16 @@ static void __init kernel_physical_mapping_init(void)
1398 1078
1399 phys_start = pall[i].phys_addr; 1079 phys_start = pall[i].phys_addr;
1400 phys_end = phys_start + pall[i].reg_size; 1080 phys_end = phys_start + pall[i].reg_size;
1081
1082 mark_kpte_bitmap(phys_start, phys_end);
1083
1084#ifdef CONFIG_DEBUG_PAGEALLOC
1401 mem_alloced += kernel_map_range(phys_start, phys_end, 1085 mem_alloced += kernel_map_range(phys_start, phys_end,
1402 PAGE_KERNEL); 1086 PAGE_KERNEL);
1087#endif
1403 } 1088 }
1404 1089
1090#ifdef CONFIG_DEBUG_PAGEALLOC
1405 printk("Allocated %ld bytes for kernel page tables.\n", 1091 printk("Allocated %ld bytes for kernel page tables.\n",
1406 mem_alloced); 1092 mem_alloced);
1407 1093
@@ -1409,8 +1095,10 @@ static void __init kernel_physical_mapping_init(void)
1409 flushi(&kvmap_linear_patch[0]); 1095 flushi(&kvmap_linear_patch[0]);
1410 1096
1411 __flush_tlb_all(); 1097 __flush_tlb_all();
1098#endif
1412} 1099}
1413 1100
1101#ifdef CONFIG_DEBUG_PAGEALLOC
1414void kernel_map_pages(struct page *page, int numpages, int enable) 1102void kernel_map_pages(struct page *page, int numpages, int enable)
1415{ 1103{
1416 unsigned long phys_start = page_to_pfn(page) << PAGE_SHIFT; 1104 unsigned long phys_start = page_to_pfn(page) << PAGE_SHIFT;
@@ -1419,6 +1107,9 @@ void kernel_map_pages(struct page *page, int numpages, int enable)
1419 kernel_map_range(phys_start, phys_end, 1107 kernel_map_range(phys_start, phys_end,
1420 (enable ? PAGE_KERNEL : __pgprot(0))); 1108 (enable ? PAGE_KERNEL : __pgprot(0)));
1421 1109
1110 flush_tsb_kernel_range(PAGE_OFFSET + phys_start,
1111 PAGE_OFFSET + phys_end);
1112
1422 /* we should perform an IPI and flush all tlbs, 1113 /* we should perform an IPI and flush all tlbs,
1423 * but that can deadlock->flush only current cpu. 1114 * but that can deadlock->flush only current cpu.
1424 */ 1115 */
@@ -1439,18 +1130,150 @@ unsigned long __init find_ecache_flush_span(unsigned long size)
1439 return ~0UL; 1130 return ~0UL;
1440} 1131}
1441 1132
1133static void __init tsb_phys_patch(void)
1134{
1135 struct tsb_ldquad_phys_patch_entry *pquad;
1136 struct tsb_phys_patch_entry *p;
1137
1138 pquad = &__tsb_ldquad_phys_patch;
1139 while (pquad < &__tsb_ldquad_phys_patch_end) {
1140 unsigned long addr = pquad->addr;
1141
1142 if (tlb_type == hypervisor)
1143 *(unsigned int *) addr = pquad->sun4v_insn;
1144 else
1145 *(unsigned int *) addr = pquad->sun4u_insn;
1146 wmb();
1147 __asm__ __volatile__("flush %0"
1148 : /* no outputs */
1149 : "r" (addr));
1150
1151 pquad++;
1152 }
1153
1154 p = &__tsb_phys_patch;
1155 while (p < &__tsb_phys_patch_end) {
1156 unsigned long addr = p->addr;
1157
1158 *(unsigned int *) addr = p->insn;
1159 wmb();
1160 __asm__ __volatile__("flush %0"
1161 : /* no outputs */
1162 : "r" (addr));
1163
1164 p++;
1165 }
1166}
1167
1168/* Don't mark as init, we give this to the Hypervisor. */
1169static struct hv_tsb_descr ktsb_descr[2];
1170extern struct tsb swapper_tsb[KERNEL_TSB_NENTRIES];
1171
1172static void __init sun4v_ktsb_init(void)
1173{
1174 unsigned long ktsb_pa;
1175
1176 /* First KTSB for PAGE_SIZE mappings. */
1177 ktsb_pa = kern_base + ((unsigned long)&swapper_tsb[0] - KERNBASE);
1178
1179 switch (PAGE_SIZE) {
1180 case 8 * 1024:
1181 default:
1182 ktsb_descr[0].pgsz_idx = HV_PGSZ_IDX_8K;
1183 ktsb_descr[0].pgsz_mask = HV_PGSZ_MASK_8K;
1184 break;
1185
1186 case 64 * 1024:
1187 ktsb_descr[0].pgsz_idx = HV_PGSZ_IDX_64K;
1188 ktsb_descr[0].pgsz_mask = HV_PGSZ_MASK_64K;
1189 break;
1190
1191 case 512 * 1024:
1192 ktsb_descr[0].pgsz_idx = HV_PGSZ_IDX_512K;
1193 ktsb_descr[0].pgsz_mask = HV_PGSZ_MASK_512K;
1194 break;
1195
1196 case 4 * 1024 * 1024:
1197 ktsb_descr[0].pgsz_idx = HV_PGSZ_IDX_4MB;
1198 ktsb_descr[0].pgsz_mask = HV_PGSZ_MASK_4MB;
1199 break;
1200 };
1201
1202 ktsb_descr[0].assoc = 1;
1203 ktsb_descr[0].num_ttes = KERNEL_TSB_NENTRIES;
1204 ktsb_descr[0].ctx_idx = 0;
1205 ktsb_descr[0].tsb_base = ktsb_pa;
1206 ktsb_descr[0].resv = 0;
1207
1208 /* Second KTSB for 4MB/256MB mappings. */
1209 ktsb_pa = (kern_base +
1210 ((unsigned long)&swapper_4m_tsb[0] - KERNBASE));
1211
1212 ktsb_descr[1].pgsz_idx = HV_PGSZ_IDX_4MB;
1213 ktsb_descr[1].pgsz_mask = (HV_PGSZ_MASK_4MB |
1214 HV_PGSZ_MASK_256MB);
1215 ktsb_descr[1].assoc = 1;
1216 ktsb_descr[1].num_ttes = KERNEL_TSB4M_NENTRIES;
1217 ktsb_descr[1].ctx_idx = 0;
1218 ktsb_descr[1].tsb_base = ktsb_pa;
1219 ktsb_descr[1].resv = 0;
1220}
1221
1222void __cpuinit sun4v_ktsb_register(void)
1223{
1224 register unsigned long func asm("%o5");
1225 register unsigned long arg0 asm("%o0");
1226 register unsigned long arg1 asm("%o1");
1227 unsigned long pa;
1228
1229 pa = kern_base + ((unsigned long)&ktsb_descr[0] - KERNBASE);
1230
1231 func = HV_FAST_MMU_TSB_CTX0;
1232 arg0 = 2;
1233 arg1 = pa;
1234 __asm__ __volatile__("ta %6"
1235 : "=&r" (func), "=&r" (arg0), "=&r" (arg1)
1236 : "0" (func), "1" (arg0), "2" (arg1),
1237 "i" (HV_FAST_TRAP));
1238}
1239
1442/* paging_init() sets up the page tables */ 1240/* paging_init() sets up the page tables */
1443 1241
1444extern void cheetah_ecache_flush_init(void); 1242extern void cheetah_ecache_flush_init(void);
1243extern void sun4v_patch_tlb_handlers(void);
1445 1244
1446static unsigned long last_valid_pfn; 1245static unsigned long last_valid_pfn;
1447pgd_t swapper_pg_dir[2048]; 1246pgd_t swapper_pg_dir[2048];
1448 1247
1248static void sun4u_pgprot_init(void);
1249static void sun4v_pgprot_init(void);
1250
1449void __init paging_init(void) 1251void __init paging_init(void)
1450{ 1252{
1451 unsigned long end_pfn, pages_avail, shift; 1253 unsigned long end_pfn, pages_avail, shift, phys_base;
1452 unsigned long real_end, i; 1254 unsigned long real_end, i;
1453 1255
1256 kern_base = (prom_boot_mapping_phys_low >> 22UL) << 22UL;
1257 kern_size = (unsigned long)&_end - (unsigned long)KERNBASE;
1258
1259 /* Invalidate both kernel TSBs. */
1260 memset(swapper_tsb, 0x40, sizeof(swapper_tsb));
1261 memset(swapper_4m_tsb, 0x40, sizeof(swapper_4m_tsb));
1262
1263 if (tlb_type == hypervisor)
1264 sun4v_pgprot_init();
1265 else
1266 sun4u_pgprot_init();
1267
1268 if (tlb_type == cheetah_plus ||
1269 tlb_type == hypervisor)
1270 tsb_phys_patch();
1271
1272 if (tlb_type == hypervisor) {
1273 sun4v_patch_tlb_handlers();
1274 sun4v_ktsb_init();
1275 }
1276
1454 /* Find available physical memory... */ 1277 /* Find available physical memory... */
1455 read_obp_memory("available", &pavail[0], &pavail_ents); 1278 read_obp_memory("available", &pavail[0], &pavail_ents);
1456 1279
@@ -1458,11 +1281,6 @@ void __init paging_init(void)
1458 for (i = 0; i < pavail_ents; i++) 1281 for (i = 0; i < pavail_ents; i++)
1459 phys_base = min(phys_base, pavail[i].phys_addr); 1282 phys_base = min(phys_base, pavail[i].phys_addr);
1460 1283
1461 pfn_base = phys_base >> PAGE_SHIFT;
1462
1463 kern_base = (prom_boot_mapping_phys_low >> 22UL) << 22UL;
1464 kern_size = (unsigned long)&_end - (unsigned long)KERNBASE;
1465
1466 set_bit(0, mmu_context_bmap); 1284 set_bit(0, mmu_context_bmap);
1467 1285
1468 shift = kern_base + PAGE_OFFSET - ((unsigned long)KERNBASE); 1286 shift = kern_base + PAGE_OFFSET - ((unsigned long)KERNBASE);
@@ -1486,47 +1304,38 @@ void __init paging_init(void)
1486 pud_set(pud_offset(&swapper_pg_dir[0], 0), 1304 pud_set(pud_offset(&swapper_pg_dir[0], 0),
1487 swapper_low_pmd_dir + (shift / sizeof(pgd_t))); 1305 swapper_low_pmd_dir + (shift / sizeof(pgd_t)));
1488 1306
1489 swapper_pgd_zero = pgd_val(swapper_pg_dir[0]);
1490
1491 inherit_prom_mappings(); 1307 inherit_prom_mappings();
1492 1308
1493 /* Ok, we can use our TLB miss and window trap handlers safely. 1309 /* Ok, we can use our TLB miss and window trap handlers safely. */
1494 * We need to do a quick peek here to see if we are on StarFire 1310 setup_tba();
1495 * or not, so setup_tba can setup the IRQ globals correctly (it
1496 * needs to get the hard smp processor id correctly).
1497 */
1498 {
1499 extern void setup_tba(int);
1500 setup_tba(this_is_starfire);
1501 }
1502
1503 inherit_locked_prom_mappings(1);
1504 1311
1505 __flush_tlb_all(); 1312 __flush_tlb_all();
1506 1313
1314 if (tlb_type == hypervisor)
1315 sun4v_ktsb_register();
1316
1507 /* Setup bootmem... */ 1317 /* Setup bootmem... */
1508 pages_avail = 0; 1318 pages_avail = 0;
1509 last_valid_pfn = end_pfn = bootmem_init(&pages_avail); 1319 last_valid_pfn = end_pfn = bootmem_init(&pages_avail, phys_base);
1320
1321 max_mapnr = last_valid_pfn;
1510 1322
1511#ifdef CONFIG_DEBUG_PAGEALLOC
1512 kernel_physical_mapping_init(); 1323 kernel_physical_mapping_init();
1513#endif
1514 1324
1515 { 1325 {
1516 unsigned long zones_size[MAX_NR_ZONES]; 1326 unsigned long zones_size[MAX_NR_ZONES];
1517 unsigned long zholes_size[MAX_NR_ZONES]; 1327 unsigned long zholes_size[MAX_NR_ZONES];
1518 unsigned long npages;
1519 int znum; 1328 int znum;
1520 1329
1521 for (znum = 0; znum < MAX_NR_ZONES; znum++) 1330 for (znum = 0; znum < MAX_NR_ZONES; znum++)
1522 zones_size[znum] = zholes_size[znum] = 0; 1331 zones_size[znum] = zholes_size[znum] = 0;
1523 1332
1524 npages = end_pfn - pfn_base; 1333 zones_size[ZONE_DMA] = end_pfn;
1525 zones_size[ZONE_DMA] = npages; 1334 zholes_size[ZONE_DMA] = end_pfn - pages_avail;
1526 zholes_size[ZONE_DMA] = npages - pages_avail;
1527 1335
1528 free_area_init_node(0, &contig_page_data, zones_size, 1336 free_area_init_node(0, &contig_page_data, zones_size,
1529 phys_base >> PAGE_SHIFT, zholes_size); 1337 __pa(PAGE_OFFSET) >> PAGE_SHIFT,
1338 zholes_size);
1530 } 1339 }
1531 1340
1532 device_scan(); 1341 device_scan();
@@ -1596,7 +1405,6 @@ void __init mem_init(void)
1596 1405
1597 taint_real_pages(); 1406 taint_real_pages();
1598 1407
1599 max_mapnr = last_valid_pfn - pfn_base;
1600 high_memory = __va(last_valid_pfn << PAGE_SHIFT); 1408 high_memory = __va(last_valid_pfn << PAGE_SHIFT);
1601 1409
1602#ifdef CONFIG_DEBUG_BOOTMEM 1410#ifdef CONFIG_DEBUG_BOOTMEM
@@ -1676,3 +1484,342 @@ void free_initrd_mem(unsigned long start, unsigned long end)
1676 } 1484 }
1677} 1485}
1678#endif 1486#endif
1487
1488#define _PAGE_CACHE_4U (_PAGE_CP_4U | _PAGE_CV_4U)
1489#define _PAGE_CACHE_4V (_PAGE_CP_4V | _PAGE_CV_4V)
1490#define __DIRTY_BITS_4U (_PAGE_MODIFIED_4U | _PAGE_WRITE_4U | _PAGE_W_4U)
1491#define __DIRTY_BITS_4V (_PAGE_MODIFIED_4V | _PAGE_WRITE_4V | _PAGE_W_4V)
1492#define __ACCESS_BITS_4U (_PAGE_ACCESSED_4U | _PAGE_READ_4U | _PAGE_R)
1493#define __ACCESS_BITS_4V (_PAGE_ACCESSED_4V | _PAGE_READ_4V | _PAGE_R)
1494
1495pgprot_t PAGE_KERNEL __read_mostly;
1496EXPORT_SYMBOL(PAGE_KERNEL);
1497
1498pgprot_t PAGE_KERNEL_LOCKED __read_mostly;
1499pgprot_t PAGE_COPY __read_mostly;
1500
1501pgprot_t PAGE_SHARED __read_mostly;
1502EXPORT_SYMBOL(PAGE_SHARED);
1503
1504pgprot_t PAGE_EXEC __read_mostly;
1505unsigned long pg_iobits __read_mostly;
1506
1507unsigned long _PAGE_IE __read_mostly;
1508
1509unsigned long _PAGE_E __read_mostly;
1510EXPORT_SYMBOL(_PAGE_E);
1511
1512unsigned long _PAGE_CACHE __read_mostly;
1513EXPORT_SYMBOL(_PAGE_CACHE);
1514
1515static void prot_init_common(unsigned long page_none,
1516 unsigned long page_shared,
1517 unsigned long page_copy,
1518 unsigned long page_readonly,
1519 unsigned long page_exec_bit)
1520{
1521 PAGE_COPY = __pgprot(page_copy);
1522 PAGE_SHARED = __pgprot(page_shared);
1523
1524 protection_map[0x0] = __pgprot(page_none);
1525 protection_map[0x1] = __pgprot(page_readonly & ~page_exec_bit);
1526 protection_map[0x2] = __pgprot(page_copy & ~page_exec_bit);
1527 protection_map[0x3] = __pgprot(page_copy & ~page_exec_bit);
1528 protection_map[0x4] = __pgprot(page_readonly);
1529 protection_map[0x5] = __pgprot(page_readonly);
1530 protection_map[0x6] = __pgprot(page_copy);
1531 protection_map[0x7] = __pgprot(page_copy);
1532 protection_map[0x8] = __pgprot(page_none);
1533 protection_map[0x9] = __pgprot(page_readonly & ~page_exec_bit);
1534 protection_map[0xa] = __pgprot(page_shared & ~page_exec_bit);
1535 protection_map[0xb] = __pgprot(page_shared & ~page_exec_bit);
1536 protection_map[0xc] = __pgprot(page_readonly);
1537 protection_map[0xd] = __pgprot(page_readonly);
1538 protection_map[0xe] = __pgprot(page_shared);
1539 protection_map[0xf] = __pgprot(page_shared);
1540}
1541
1542static void __init sun4u_pgprot_init(void)
1543{
1544 unsigned long page_none, page_shared, page_copy, page_readonly;
1545 unsigned long page_exec_bit;
1546
1547 PAGE_KERNEL = __pgprot (_PAGE_PRESENT_4U | _PAGE_VALID |
1548 _PAGE_CACHE_4U | _PAGE_P_4U |
1549 __ACCESS_BITS_4U | __DIRTY_BITS_4U |
1550 _PAGE_EXEC_4U);
1551 PAGE_KERNEL_LOCKED = __pgprot (_PAGE_PRESENT_4U | _PAGE_VALID |
1552 _PAGE_CACHE_4U | _PAGE_P_4U |
1553 __ACCESS_BITS_4U | __DIRTY_BITS_4U |
1554 _PAGE_EXEC_4U | _PAGE_L_4U);
1555 PAGE_EXEC = __pgprot(_PAGE_EXEC_4U);
1556
1557 _PAGE_IE = _PAGE_IE_4U;
1558 _PAGE_E = _PAGE_E_4U;
1559 _PAGE_CACHE = _PAGE_CACHE_4U;
1560
1561 pg_iobits = (_PAGE_VALID | _PAGE_PRESENT_4U | __DIRTY_BITS_4U |
1562 __ACCESS_BITS_4U | _PAGE_E_4U);
1563
1564 kern_linear_pte_xor[0] = (_PAGE_VALID | _PAGE_SZ4MB_4U) ^
1565 0xfffff80000000000;
1566 kern_linear_pte_xor[0] |= (_PAGE_CP_4U | _PAGE_CV_4U |
1567 _PAGE_P_4U | _PAGE_W_4U);
1568
1569 /* XXX Should use 256MB on Panther. XXX */
1570 kern_linear_pte_xor[1] = kern_linear_pte_xor[0];
1571
1572 _PAGE_SZBITS = _PAGE_SZBITS_4U;
1573 _PAGE_ALL_SZ_BITS = (_PAGE_SZ4MB_4U | _PAGE_SZ512K_4U |
1574 _PAGE_SZ64K_4U | _PAGE_SZ8K_4U |
1575 _PAGE_SZ32MB_4U | _PAGE_SZ256MB_4U);
1576
1577
1578 page_none = _PAGE_PRESENT_4U | _PAGE_ACCESSED_4U | _PAGE_CACHE_4U;
1579 page_shared = (_PAGE_VALID | _PAGE_PRESENT_4U | _PAGE_CACHE_4U |
1580 __ACCESS_BITS_4U | _PAGE_WRITE_4U | _PAGE_EXEC_4U);
1581 page_copy = (_PAGE_VALID | _PAGE_PRESENT_4U | _PAGE_CACHE_4U |
1582 __ACCESS_BITS_4U | _PAGE_EXEC_4U);
1583 page_readonly = (_PAGE_VALID | _PAGE_PRESENT_4U | _PAGE_CACHE_4U |
1584 __ACCESS_BITS_4U | _PAGE_EXEC_4U);
1585
1586 page_exec_bit = _PAGE_EXEC_4U;
1587
1588 prot_init_common(page_none, page_shared, page_copy, page_readonly,
1589 page_exec_bit);
1590}
1591
1592static void __init sun4v_pgprot_init(void)
1593{
1594 unsigned long page_none, page_shared, page_copy, page_readonly;
1595 unsigned long page_exec_bit;
1596
1597 PAGE_KERNEL = __pgprot (_PAGE_PRESENT_4V | _PAGE_VALID |
1598 _PAGE_CACHE_4V | _PAGE_P_4V |
1599 __ACCESS_BITS_4V | __DIRTY_BITS_4V |
1600 _PAGE_EXEC_4V);
1601 PAGE_KERNEL_LOCKED = PAGE_KERNEL;
1602 PAGE_EXEC = __pgprot(_PAGE_EXEC_4V);
1603
1604 _PAGE_IE = _PAGE_IE_4V;
1605 _PAGE_E = _PAGE_E_4V;
1606 _PAGE_CACHE = _PAGE_CACHE_4V;
1607
1608 kern_linear_pte_xor[0] = (_PAGE_VALID | _PAGE_SZ4MB_4V) ^
1609 0xfffff80000000000;
1610 kern_linear_pte_xor[0] |= (_PAGE_CP_4V | _PAGE_CV_4V |
1611 _PAGE_P_4V | _PAGE_W_4V);
1612
1613 kern_linear_pte_xor[1] = (_PAGE_VALID | _PAGE_SZ256MB_4V) ^
1614 0xfffff80000000000;
1615 kern_linear_pte_xor[1] |= (_PAGE_CP_4V | _PAGE_CV_4V |
1616 _PAGE_P_4V | _PAGE_W_4V);
1617
1618 pg_iobits = (_PAGE_VALID | _PAGE_PRESENT_4V | __DIRTY_BITS_4V |
1619 __ACCESS_BITS_4V | _PAGE_E_4V);
1620
1621 _PAGE_SZBITS = _PAGE_SZBITS_4V;
1622 _PAGE_ALL_SZ_BITS = (_PAGE_SZ16GB_4V | _PAGE_SZ2GB_4V |
1623 _PAGE_SZ256MB_4V | _PAGE_SZ32MB_4V |
1624 _PAGE_SZ4MB_4V | _PAGE_SZ512K_4V |
1625 _PAGE_SZ64K_4V | _PAGE_SZ8K_4V);
1626
1627 page_none = _PAGE_PRESENT_4V | _PAGE_ACCESSED_4V | _PAGE_CACHE_4V;
1628 page_shared = (_PAGE_VALID | _PAGE_PRESENT_4V | _PAGE_CACHE_4V |
1629 __ACCESS_BITS_4V | _PAGE_WRITE_4V | _PAGE_EXEC_4V);
1630 page_copy = (_PAGE_VALID | _PAGE_PRESENT_4V | _PAGE_CACHE_4V |
1631 __ACCESS_BITS_4V | _PAGE_EXEC_4V);
1632 page_readonly = (_PAGE_VALID | _PAGE_PRESENT_4V | _PAGE_CACHE_4V |
1633 __ACCESS_BITS_4V | _PAGE_EXEC_4V);
1634
1635 page_exec_bit = _PAGE_EXEC_4V;
1636
1637 prot_init_common(page_none, page_shared, page_copy, page_readonly,
1638 page_exec_bit);
1639}
1640
1641unsigned long pte_sz_bits(unsigned long sz)
1642{
1643 if (tlb_type == hypervisor) {
1644 switch (sz) {
1645 case 8 * 1024:
1646 default:
1647 return _PAGE_SZ8K_4V;
1648 case 64 * 1024:
1649 return _PAGE_SZ64K_4V;
1650 case 512 * 1024:
1651 return _PAGE_SZ512K_4V;
1652 case 4 * 1024 * 1024:
1653 return _PAGE_SZ4MB_4V;
1654 };
1655 } else {
1656 switch (sz) {
1657 case 8 * 1024:
1658 default:
1659 return _PAGE_SZ8K_4U;
1660 case 64 * 1024:
1661 return _PAGE_SZ64K_4U;
1662 case 512 * 1024:
1663 return _PAGE_SZ512K_4U;
1664 case 4 * 1024 * 1024:
1665 return _PAGE_SZ4MB_4U;
1666 };
1667 }
1668}
1669
1670pte_t mk_pte_io(unsigned long page, pgprot_t prot, int space, unsigned long page_size)
1671{
1672 pte_t pte;
1673
1674 pte_val(pte) = page | pgprot_val(pgprot_noncached(prot));
1675 pte_val(pte) |= (((unsigned long)space) << 32);
1676 pte_val(pte) |= pte_sz_bits(page_size);
1677
1678 return pte;
1679}
1680
1681static unsigned long kern_large_tte(unsigned long paddr)
1682{
1683 unsigned long val;
1684
1685 val = (_PAGE_VALID | _PAGE_SZ4MB_4U |
1686 _PAGE_CP_4U | _PAGE_CV_4U | _PAGE_P_4U |
1687 _PAGE_EXEC_4U | _PAGE_L_4U | _PAGE_W_4U);
1688 if (tlb_type == hypervisor)
1689 val = (_PAGE_VALID | _PAGE_SZ4MB_4V |
1690 _PAGE_CP_4V | _PAGE_CV_4V | _PAGE_P_4V |
1691 _PAGE_EXEC_4V | _PAGE_W_4V);
1692
1693 return val | paddr;
1694}
1695
1696/*
1697 * Translate PROM's mapping we capture at boot time into physical address.
1698 * The second parameter is only set from prom_callback() invocations.
1699 */
1700unsigned long prom_virt_to_phys(unsigned long promva, int *error)
1701{
1702 unsigned long mask;
1703 int i;
1704
1705 mask = _PAGE_PADDR_4U;
1706 if (tlb_type == hypervisor)
1707 mask = _PAGE_PADDR_4V;
1708
1709 for (i = 0; i < prom_trans_ents; i++) {
1710 struct linux_prom_translation *p = &prom_trans[i];
1711
1712 if (promva >= p->virt &&
1713 promva < (p->virt + p->size)) {
1714 unsigned long base = p->data & mask;
1715
1716 if (error)
1717 *error = 0;
1718 return base + (promva & (8192 - 1));
1719 }
1720 }
1721 if (error)
1722 *error = 1;
1723 return 0UL;
1724}
1725
1726/* XXX We should kill off this ugly thing at so me point. XXX */
1727unsigned long sun4u_get_pte(unsigned long addr)
1728{
1729 pgd_t *pgdp;
1730 pud_t *pudp;
1731 pmd_t *pmdp;
1732 pte_t *ptep;
1733 unsigned long mask = _PAGE_PADDR_4U;
1734
1735 if (tlb_type == hypervisor)
1736 mask = _PAGE_PADDR_4V;
1737
1738 if (addr >= PAGE_OFFSET)
1739 return addr & mask;
1740
1741 if ((addr >= LOW_OBP_ADDRESS) && (addr < HI_OBP_ADDRESS))
1742 return prom_virt_to_phys(addr, NULL);
1743
1744 pgdp = pgd_offset_k(addr);
1745 pudp = pud_offset(pgdp, addr);
1746 pmdp = pmd_offset(pudp, addr);
1747 ptep = pte_offset_kernel(pmdp, addr);
1748
1749 return pte_val(*ptep) & mask;
1750}
1751
1752/* If not locked, zap it. */
1753void __flush_tlb_all(void)
1754{
1755 unsigned long pstate;
1756 int i;
1757
1758 __asm__ __volatile__("flushw\n\t"
1759 "rdpr %%pstate, %0\n\t"
1760 "wrpr %0, %1, %%pstate"
1761 : "=r" (pstate)
1762 : "i" (PSTATE_IE));
1763 if (tlb_type == spitfire) {
1764 for (i = 0; i < 64; i++) {
1765 /* Spitfire Errata #32 workaround */
1766 /* NOTE: Always runs on spitfire, so no
1767 * cheetah+ page size encodings.
1768 */
1769 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
1770 "flush %%g6"
1771 : /* No outputs */
1772 : "r" (0),
1773 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
1774
1775 if (!(spitfire_get_dtlb_data(i) & _PAGE_L_4U)) {
1776 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
1777 "membar #Sync"
1778 : /* no outputs */
1779 : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
1780 spitfire_put_dtlb_data(i, 0x0UL);
1781 }
1782
1783 /* Spitfire Errata #32 workaround */
1784 /* NOTE: Always runs on spitfire, so no
1785 * cheetah+ page size encodings.
1786 */
1787 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
1788 "flush %%g6"
1789 : /* No outputs */
1790 : "r" (0),
1791 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
1792
1793 if (!(spitfire_get_itlb_data(i) & _PAGE_L_4U)) {
1794 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
1795 "membar #Sync"
1796 : /* no outputs */
1797 : "r" (TLB_TAG_ACCESS), "i" (ASI_IMMU));
1798 spitfire_put_itlb_data(i, 0x0UL);
1799 }
1800 }
1801 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
1802 cheetah_flush_dtlb_all();
1803 cheetah_flush_itlb_all();
1804 }
1805 __asm__ __volatile__("wrpr %0, 0, %%pstate"
1806 : : "r" (pstate));
1807}
1808
1809#ifdef CONFIG_MEMORY_HOTPLUG
1810
1811void online_page(struct page *page)
1812{
1813 ClearPageReserved(page);
1814 set_page_count(page, 0);
1815 free_cold_page(page);
1816 totalram_pages++;
1817 num_physpages++;
1818}
1819
1820int remove_memory(u64 start, u64 size)
1821{
1822 return -EINVAL;
1823}
1824
1825#endif /* CONFIG_MEMORY_HOTPLUG */
generated by cgit v1.2.2 (git 2.25.0) at 2025-10-24 15:54:29 -0400