54bf540598118339a04919e1f08c505a5ad57cd8
[linux-2.6.git] / arch / ia64 / mm / contig.c
1 /*
2  * This file is subject to the terms and conditions of the GNU General Public
3  * License.  See the file "COPYING" in the main directory of this archive
4  * for more details.
5  *
6  * Copyright (C) 1998-2003 Hewlett-Packard Co
7  *      David Mosberger-Tang <davidm@hpl.hp.com>
8  *      Stephane Eranian <eranian@hpl.hp.com>
9  * Copyright (C) 2000, Rohit Seth <rohit.seth@intel.com>
10  * Copyright (C) 1999 VA Linux Systems
11  * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
12  * Copyright (C) 2003 Silicon Graphics, Inc. All rights reserved.
13  *
14  * Routines used by ia64 machines with contiguous (or virtually contiguous)
15  * memory.
16  */
17 #include <linux/bootmem.h>
18 #include <linux/efi.h>
19 #include <linux/mm.h>
20 #include <linux/nmi.h>
21 #include <linux/swap.h>
22
23 #include <asm/meminit.h>
24 #include <asm/pgalloc.h>
25 #include <asm/pgtable.h>
26 #include <asm/sections.h>
27 #include <asm/mca.h>
28
29 #ifdef CONFIG_VIRTUAL_MEM_MAP
30 static unsigned long max_gap;
31 #endif
32
33 /**
34  * show_mem - give short summary of memory stats
35  *
36  * Shows a simple page count of reserved and used pages in the system.
37  * For discontig machines, it does this on a per-pgdat basis.
38  */
39 void show_mem(void)
40 {
41         int i, total_reserved = 0;
42         int total_shared = 0, total_cached = 0;
43         unsigned long total_present = 0;
44         pg_data_t *pgdat;
45
46         printk(KERN_INFO "Mem-info:\n");
47         show_free_areas();
48         printk(KERN_INFO "Node memory in pages:\n");
49         for_each_online_pgdat(pgdat) {
50                 unsigned long present;
51                 unsigned long flags;
52                 int shared = 0, cached = 0, reserved = 0;
53
54                 pgdat_resize_lock(pgdat, &flags);
55                 present = pgdat->node_present_pages;
56                 for(i = 0; i < pgdat->node_spanned_pages; i++) {
57                         struct page *page;
58                         if (unlikely(i % MAX_ORDER_NR_PAGES == 0))
59                                 touch_nmi_watchdog();
60                         if (pfn_valid(pgdat->node_start_pfn + i))
61                                 page = pfn_to_page(pgdat->node_start_pfn + i);
62                         else {
63 #ifdef CONFIG_VIRTUAL_MEM_MAP
64                                 if (max_gap < LARGE_GAP)
65                                         continue;
66 #endif
67                                 i = vmemmap_find_next_valid_pfn(pgdat->node_id,
68                                          i) - 1;
69                                 continue;
70                         }
71                         if (PageReserved(page))
72                                 reserved++;
73                         else if (PageSwapCache(page))
74                                 cached++;
75                         else if (page_count(page))
76                                 shared += page_count(page)-1;
77                 }
78                 pgdat_resize_unlock(pgdat, &flags);
79                 total_present += present;
80                 total_reserved += reserved;
81                 total_cached += cached;
82                 total_shared += shared;
83                 printk(KERN_INFO "Node %4d:  RAM: %11ld, rsvd: %8d, "
84                        "shrd: %10d, swpd: %10d\n", pgdat->node_id,
85                        present, reserved, shared, cached);
86         }
87         printk(KERN_INFO "%ld pages of RAM\n", total_present);
88         printk(KERN_INFO "%d reserved pages\n", total_reserved);
89         printk(KERN_INFO "%d pages shared\n", total_shared);
90         printk(KERN_INFO "%d pages swap cached\n", total_cached);
91         printk(KERN_INFO "Total of %ld pages in page table cache\n",
92                quicklist_total_size());
93         printk(KERN_INFO "%d free buffer pages\n", nr_free_buffer_pages());
94 }
95
96
97 /* physical address where the bootmem map is located */
98 unsigned long bootmap_start;
99
100 /**
101  * find_bootmap_location - callback to find a memory area for the bootmap
102  * @start: start of region
103  * @end: end of region
104  * @arg: unused callback data
105  *
106  * Find a place to put the bootmap and return its starting address in
107  * bootmap_start.  This address must be page-aligned.
108  */
109 static int __init
110 find_bootmap_location (u64 start, u64 end, void *arg)
111 {
112         u64 needed = *(unsigned long *)arg;
113         u64 range_start, range_end, free_start;
114         int i;
115
116 #if IGNORE_PFN0
117         if (start == PAGE_OFFSET) {
118                 start += PAGE_SIZE;
119                 if (start >= end)
120                         return 0;
121         }
122 #endif
123
124         free_start = PAGE_OFFSET;
125
126         for (i = 0; i < num_rsvd_regions; i++) {
127                 range_start = max(start, free_start);
128                 range_end   = min(end, rsvd_region[i].start & PAGE_MASK);
129
130                 free_start = PAGE_ALIGN(rsvd_region[i].end);
131
132                 if (range_end <= range_start)
133                         continue; /* skip over empty range */
134
135                 if (range_end - range_start >= needed) {
136                         bootmap_start = __pa(range_start);
137                         return -1;      /* done */
138                 }
139
140                 /* nothing more available in this segment */
141                 if (range_end == end)
142                         return 0;
143         }
144         return 0;
145 }
146
147 #ifdef CONFIG_SMP
148 static void *cpu_data;
149 /**
150  * per_cpu_init - setup per-cpu variables
151  *
152  * Allocate and setup per-cpu data areas.
153  */
154 void * __cpuinit
155 per_cpu_init (void)
156 {
157         static bool first_time = true;
158         void *cpu0_data = __cpu0_per_cpu;
159         unsigned int cpu;
160
161         if (!first_time)
162                 goto skip;
163         first_time = false;
164
165         /*
166          * get_free_pages() cannot be used before cpu_init() done.
167          * BSP allocates PERCPU_PAGE_SIZE bytes for all possible CPUs
168          * to avoid that AP calls get_zeroed_page().
169          */
170         for_each_possible_cpu(cpu) {
171                 void *src = cpu == 0 ? cpu0_data : __phys_per_cpu_start;
172
173                 memcpy(cpu_data, src, __per_cpu_end - __per_cpu_start);
174                 __per_cpu_offset[cpu] = (char *)cpu_data - __per_cpu_start;
175                 per_cpu(local_per_cpu_offset, cpu) = __per_cpu_offset[cpu];
176
177                 /*
178                  * percpu area for cpu0 is moved from the __init area
179                  * which is setup by head.S and used till this point.
180                  * Update ar.k3.  This move is ensures that percpu
181                  * area for cpu0 is on the correct node and its
182                  * virtual address isn't insanely far from other
183                  * percpu areas which is important for congruent
184                  * percpu allocator.
185                  */
186                 if (cpu == 0)
187                         ia64_set_kr(IA64_KR_PER_CPU_DATA, __pa(cpu_data) -
188                                     (unsigned long)__per_cpu_start);
189
190                 cpu_data += PERCPU_PAGE_SIZE;
191         }
192 skip:
193         return __per_cpu_start + __per_cpu_offset[smp_processor_id()];
194 }
195
196 static inline void
197 alloc_per_cpu_data(void)
198 {
199         cpu_data = __alloc_bootmem(PERCPU_PAGE_SIZE * num_possible_cpus(),
200                                    PERCPU_PAGE_SIZE, __pa(MAX_DMA_ADDRESS));
201 }
202
203 /**
204  * setup_per_cpu_areas - setup percpu areas
205  *
206  * Arch code has already allocated and initialized percpu areas.  All
207  * this function has to do is to teach the determined layout to the
208  * dynamic percpu allocator, which happens to be more complex than
209  * creating whole new ones using helpers.
210  */
211 void __init
212 setup_per_cpu_areas(void)
213 {
214         struct pcpu_alloc_info *ai;
215         struct pcpu_group_info *gi;
216         unsigned int cpu;
217         ssize_t static_size, reserved_size, dyn_size;
218         int rc;
219
220         ai = pcpu_alloc_alloc_info(1, num_possible_cpus());
221         if (!ai)
222                 panic("failed to allocate pcpu_alloc_info");
223         gi = &ai->groups[0];
224
225         /* units are assigned consecutively to possible cpus */
226         for_each_possible_cpu(cpu)
227                 gi->cpu_map[gi->nr_units++] = cpu;
228
229         /* set parameters */
230         static_size = __per_cpu_end - __per_cpu_start;
231         reserved_size = PERCPU_MODULE_RESERVE;
232         dyn_size = PERCPU_PAGE_SIZE - static_size - reserved_size;
233         if (dyn_size < 0)
234                 panic("percpu area overflow static=%zd reserved=%zd\n",
235                       static_size, reserved_size);
236
237         ai->static_size         = static_size;
238         ai->reserved_size       = reserved_size;
239         ai->dyn_size            = dyn_size;
240         ai->unit_size           = PERCPU_PAGE_SIZE;
241         ai->atom_size           = PAGE_SIZE;
242         ai->alloc_size          = PERCPU_PAGE_SIZE;
243
244         rc = pcpu_setup_first_chunk(ai, __per_cpu_start + __per_cpu_offset[0]);
245         if (rc)
246                 panic("failed to setup percpu area (err=%d)", rc);
247
248         pcpu_free_alloc_info(ai);
249 }
250 #else
251 #define alloc_per_cpu_data() do { } while (0)
252 #endif /* CONFIG_SMP */
253
254 /**
255  * find_memory - setup memory map
256  *
257  * Walk the EFI memory map and find usable memory for the system, taking
258  * into account reserved areas.
259  */
260 void __init
261 find_memory (void)
262 {
263         unsigned long bootmap_size;
264
265         reserve_memory();
266
267         /* first find highest page frame number */
268         min_low_pfn = ~0UL;
269         max_low_pfn = 0;
270         efi_memmap_walk(find_max_min_low_pfn, NULL);
271         max_pfn = max_low_pfn;
272         /* how many bytes to cover all the pages */
273         bootmap_size = bootmem_bootmap_pages(max_pfn) << PAGE_SHIFT;
274
275         /* look for a location to hold the bootmap */
276         bootmap_start = ~0UL;
277         efi_memmap_walk(find_bootmap_location, &bootmap_size);
278         if (bootmap_start == ~0UL)
279                 panic("Cannot find %ld bytes for bootmap\n", bootmap_size);
280
281         bootmap_size = init_bootmem_node(NODE_DATA(0),
282                         (bootmap_start >> PAGE_SHIFT), 0, max_pfn);
283
284         /* Free all available memory, then mark bootmem-map as being in use. */
285         efi_memmap_walk(filter_rsvd_memory, free_bootmem);
286         reserve_bootmem(bootmap_start, bootmap_size, BOOTMEM_DEFAULT);
287
288         find_initrd();
289
290         alloc_per_cpu_data();
291 }
292
293 static int count_pages(u64 start, u64 end, void *arg)
294 {
295         unsigned long *count = arg;
296
297         *count += (end - start) >> PAGE_SHIFT;
298         return 0;
299 }
300
301 /*
302  * Set up the page tables.
303  */
304
305 void __init
306 paging_init (void)
307 {
308         unsigned long max_dma;
309         unsigned long max_zone_pfns[MAX_NR_ZONES];
310
311         num_physpages = 0;
312         efi_memmap_walk(count_pages, &num_physpages);
313
314         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
315 #ifdef CONFIG_ZONE_DMA
316         max_dma = virt_to_phys((void *) MAX_DMA_ADDRESS) >> PAGE_SHIFT;
317         max_zone_pfns[ZONE_DMA] = max_dma;
318 #endif
319         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
320
321 #ifdef CONFIG_VIRTUAL_MEM_MAP
322         efi_memmap_walk(filter_memory, register_active_ranges);
323         efi_memmap_walk(find_largest_hole, (u64 *)&max_gap);
324         if (max_gap < LARGE_GAP) {
325                 vmem_map = (struct page *) 0;
326                 free_area_init_nodes(max_zone_pfns);
327         } else {
328                 unsigned long map_size;
329
330                 /* allocate virtual_mem_map */
331
332                 map_size = PAGE_ALIGN(ALIGN(max_low_pfn, MAX_ORDER_NR_PAGES) *
333                         sizeof(struct page));
334                 VMALLOC_END -= map_size;
335                 vmem_map = (struct page *) VMALLOC_END;
336                 efi_memmap_walk(create_mem_map_page_table, NULL);
337
338                 /*
339                  * alloc_node_mem_map makes an adjustment for mem_map
340                  * which isn't compatible with vmem_map.
341                  */
342                 NODE_DATA(0)->node_mem_map = vmem_map +
343                         find_min_pfn_with_active_regions();
344                 free_area_init_nodes(max_zone_pfns);
345
346                 printk("Virtual mem_map starts at 0x%p\n", mem_map);
347         }
348 #else /* !CONFIG_VIRTUAL_MEM_MAP */
349         add_active_range(0, 0, max_low_pfn);
350         free_area_init_nodes(max_zone_pfns);
351 #endif /* !CONFIG_VIRTUAL_MEM_MAP */
352         zero_page_memmap_ptr = virt_to_page(ia64_imva(empty_zero_page));
353 }