aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2010-03-03 11:15:05 -0500
committerLinus Torvalds <torvalds@linux-foundation.org>2010-03-03 11:15:05 -0500
commita626b46e17d0762d664ce471d40bc506b6e721ab (patch)
tree445f6ac655ea9247d2e27529f23ba02d0991fec0
parentc1dcb4bb1e3e16e9baee578d9bb040e5fba1063e (diff)
parentdce46a04d55d6358d2d4ab44a4946a19f9425fe2 (diff)
Merge branch 'x86-bootmem-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip
* 'x86-bootmem-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (30 commits) early_res: Need to save the allocation name in drop_range_partial() sparsemem: Fix compilation on PowerPC early_res: Add free_early_partial() x86: Fix non-bootmem compilation on PowerPC core: Move early_res from arch/x86 to kernel/ x86: Add find_fw_memmap_area Move round_up/down to kernel.h x86: Make 32bit support NO_BOOTMEM early_res: Enhance check_and_double_early_res x86: Move back find_e820_area to e820.c x86: Add find_early_area_size x86: Separate early_res related code from e820.c x86: Move bios page reserve early to head32/64.c sparsemem: Put mem map for one node together. sparsemem: Put usemap for one node together x86: Make 64 bit use early_res instead of bootmem before slab x86: Only call dma32_reserve_bootmem 64bit !CONFIG_NUMA x86: Make early_node_mem get mem > 4 GB if possible x86: Dynamically increase early_res array size x86: Introduce max_early_res and early_res_count ...
Diffstat
-rw-r--r--arch/x86/Kconfig15
-rw-r--r--arch/x86/include/asm/e820.h5
-rw-r--r--arch/x86/include/asm/pci.h2
-rw-r--r--arch/x86/include/asm/pci_64.h2
-rw-r--r--arch/x86/include/asm/proto.h10
-rw-r--r--arch/x86/kernel/cpu/mtrr/cleanup.c208
-rw-r--r--arch/x86/kernel/e820.c349
-rw-r--r--arch/x86/kernel/head32.c10
-rw-r--r--arch/x86/kernel/mmconf-fam10h_64.c7
-rw-r--r--arch/x86/kernel/pci-dma.c13
-rw-r--r--arch/x86/kernel/setup.c10
-rw-r--r--arch/x86/kernel/setup_percpu.c6
-rw-r--r--arch/x86/mm/init_32.c7
-rw-r--r--arch/x86/mm/init_64.c9
-rw-r--r--arch/x86/mm/numa_32.c3
-rw-r--r--arch/x86/mm/numa_64.c97
-rw-r--r--arch/x86/pci/Makefile3
-rw-r--r--arch/x86/pci/amd_bus.c127
-rw-r--r--arch/x86/pci/bus_numa.c25
-rw-r--r--arch/x86/pci/bus_numa.h9
-rw-r--r--arch/x86/pci/i386.c4
-rw-r--r--drivers/pci/hotplug/ibmphp_res.c14
-rw-r--r--include/linux/bootmem.h7
-rw-r--r--include/linux/early_res.h23
-rw-r--r--include/linux/kernel.h10
-rw-r--r--include/linux/mm.h12
-rw-r--r--include/linux/mmzone.h2
-rw-r--r--include/linux/range.h30
-rw-r--r--kernel/Makefile3
-rw-r--r--kernel/early_res.c578
-rw-r--r--kernel/range.c163
-rw-r--r--mm/Kconfig4
-rw-r--r--mm/bootmem.c195
-rw-r--r--mm/page_alloc.c61
-rw-r--r--mm/sparse-vmemmap.c76
-rw-r--r--mm/sparse.c196
36 files changed, 1568 insertions, 727 deletions
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 0896008f7509..57ccdcec1469 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -184,6 +184,9 @@ config ARCH_SUPPORTS_OPTIMIZED_INLINING
184config ARCH_SUPPORTS_DEBUG_PAGEALLOC 184config ARCH_SUPPORTS_DEBUG_PAGEALLOC
185 def_bool y 185 def_bool y
186 186
187config HAVE_EARLY_RES
188 def_bool y
189
187config HAVE_INTEL_TXT 190config HAVE_INTEL_TXT
188 def_bool y 191 def_bool y
189 depends on EXPERIMENTAL && DMAR && ACPI 192 depends on EXPERIMENTAL && DMAR && ACPI
@@ -569,6 +572,18 @@ config PARAVIRT_DEBUG
569 Enable to debug paravirt_ops internals. Specifically, BUG if 572 Enable to debug paravirt_ops internals. Specifically, BUG if
570 a paravirt_op is missing when it is called. 573 a paravirt_op is missing when it is called.
571 574
575config NO_BOOTMEM
576 default y
577 bool "Disable Bootmem code"
578 ---help---
579 Use early_res directly instead of bootmem before slab is ready.
580 - allocator (buddy) [generic]
581 - early allocator (bootmem) [generic]
582 - very early allocator (reserve_early*()) [x86]
583 - very very early allocator (early brk model) [x86]
584 So reduce one layer between early allocator to final allocator
585
586
572config MEMTEST 587config MEMTEST
573 bool "Memtest" 588 bool "Memtest"
574 ---help--- 589 ---help---
diff --git a/arch/x86/include/asm/e820.h b/arch/x86/include/asm/e820.h
index 761249e396fe..0e22296790d3 100644
--- a/arch/x86/include/asm/e820.h
+++ b/arch/x86/include/asm/e820.h
@@ -111,11 +111,8 @@ extern unsigned long end_user_pfn;
111 111
112extern u64 find_e820_area(u64 start, u64 end, u64 size, u64 align); 112extern u64 find_e820_area(u64 start, u64 end, u64 size, u64 align);
113extern u64 find_e820_area_size(u64 start, u64 *sizep, u64 align); 113extern u64 find_e820_area_size(u64 start, u64 *sizep, u64 align);
114extern void reserve_early(u64 start, u64 end, char *name);
115extern void reserve_early_overlap_ok(u64 start, u64 end, char *name);
116extern void free_early(u64 start, u64 end);
117extern void early_res_to_bootmem(u64 start, u64 end);
118extern u64 early_reserve_e820(u64 startt, u64 sizet, u64 align); 114extern u64 early_reserve_e820(u64 startt, u64 sizet, u64 align);
115#include <linux/early_res.h>
119 116
120extern unsigned long e820_end_of_ram_pfn(void); 117extern unsigned long e820_end_of_ram_pfn(void);
121extern unsigned long e820_end_of_low_ram_pfn(void); 118extern unsigned long e820_end_of_low_ram_pfn(void);
diff --git a/arch/x86/include/asm/pci.h b/arch/x86/include/asm/pci.h
index ada8c201d513..b4a00dd4eed5 100644
--- a/arch/x86/include/asm/pci.h
+++ b/arch/x86/include/asm/pci.h
@@ -124,6 +124,8 @@ extern void pci_iommu_alloc(void);
124#include "pci_64.h" 124#include "pci_64.h"
125#endif 125#endif
126 126
127void dma32_reserve_bootmem(void);
128
127/* implement the pci_ DMA API in terms of the generic device dma_ one */ 129/* implement the pci_ DMA API in terms of the generic device dma_ one */
128#include <asm-generic/pci-dma-compat.h> 130#include <asm-generic/pci-dma-compat.h>
129 131
diff --git a/arch/x86/include/asm/pci_64.h b/arch/x86/include/asm/pci_64.h
index ae5e40f67daf..fe15cfb21b9b 100644
--- a/arch/x86/include/asm/pci_64.h
+++ b/arch/x86/include/asm/pci_64.h
@@ -22,8 +22,6 @@ extern int (*pci_config_read)(int seg, int bus, int dev, int fn,
22extern int (*pci_config_write)(int seg, int bus, int dev, int fn, 22extern int (*pci_config_write)(int seg, int bus, int dev, int fn,
23 int reg, int len, u32 value); 23 int reg, int len, u32 value);
24 24
25extern void dma32_reserve_bootmem(void);
26
27#endif /* __KERNEL__ */ 25#endif /* __KERNEL__ */
28 26
29#endif /* _ASM_X86_PCI_64_H */ 27#endif /* _ASM_X86_PCI_64_H */
diff --git a/arch/x86/include/asm/proto.h b/arch/x86/include/asm/proto.h
index 4009f6534f52..6f414ed88620 100644
--- a/arch/x86/include/asm/proto.h
+++ b/arch/x86/include/asm/proto.h
@@ -23,14 +23,4 @@ extern int reboot_force;
23 23
24long do_arch_prctl(struct task_struct *task, int code, unsigned long addr); 24long do_arch_prctl(struct task_struct *task, int code, unsigned long addr);
25 25
26/*
27 * This looks more complex than it should be. But we need to
28 * get the type for the ~ right in round_down (it needs to be
29 * as wide as the result!), and we want to evaluate the macro
30 * arguments just once each.
31 */
32#define __round_mask(x,y) ((__typeof__(x))((y)-1))
33#define round_up(x,y) ((((x)-1) | __round_mask(x,y))+1)
34#define round_down(x,y) ((x) & ~__round_mask(x,y))
35
36#endif /* _ASM_X86_PROTO_H */ 26#endif /* _ASM_X86_PROTO_H */
diff --git a/arch/x86/kernel/cpu/mtrr/cleanup.c b/arch/x86/kernel/cpu/mtrr/cleanup.c
index 09b1698e0466..06130b52f012 100644
--- a/arch/x86/kernel/cpu/mtrr/cleanup.c
+++ b/arch/x86/kernel/cpu/mtrr/cleanup.c
@@ -22,10 +22,10 @@
22#include <linux/pci.h> 22#include <linux/pci.h>
23#include <linux/smp.h> 23#include <linux/smp.h>
24#include <linux/cpu.h> 24#include <linux/cpu.h>
25#include <linux/sort.h>
26#include <linux/mutex.h> 25#include <linux/mutex.h>
27#include <linux/uaccess.h> 26#include <linux/uaccess.h>
28#include <linux/kvm_para.h> 27#include <linux/kvm_para.h>
28#include <linux/range.h>
29 29
30#include <asm/processor.h> 30#include <asm/processor.h>
31#include <asm/e820.h> 31#include <asm/e820.h>
@@ -34,11 +34,6 @@
34 34
35#include "mtrr.h" 35#include "mtrr.h"
36 36
37struct res_range {
38 unsigned long start;
39 unsigned long end;
40};
41
42struct var_mtrr_range_state { 37struct var_mtrr_range_state {
43 unsigned long base_pfn; 38 unsigned long base_pfn;
44 unsigned long size_pfn; 39 unsigned long size_pfn;
@@ -56,7 +51,7 @@ struct var_mtrr_state {
56/* Should be related to MTRR_VAR_RANGES nums */ 51/* Should be related to MTRR_VAR_RANGES nums */
57#define RANGE_NUM 256 52#define RANGE_NUM 256
58 53
59static struct res_range __initdata range[RANGE_NUM]; 54static struct range __initdata range[RANGE_NUM];
60static int __initdata nr_range; 55static int __initdata nr_range;
61 56
62static struct var_mtrr_range_state __initdata range_state[RANGE_NUM]; 57static struct var_mtrr_range_state __initdata range_state[RANGE_NUM];
@@ -64,152 +59,11 @@ static struct var_mtrr_range_state __initdata range_state[RANGE_NUM];
64static int __initdata debug_print; 59static int __initdata debug_print;
65#define Dprintk(x...) do { if (debug_print) printk(KERN_DEBUG x); } while (0) 60#define Dprintk(x...) do { if (debug_print) printk(KERN_DEBUG x); } while (0)
66 61
67
68static int __init
69add_range(struct res_range *range, int nr_range,
70 unsigned long start, unsigned long end)
71{
72 /* Out of slots: */
73 if (nr_range >= RANGE_NUM)
74 return nr_range;
75
76 range[nr_range].start = start;
77 range[nr_range].end = end;
78
79 nr_range++;
80
81 return nr_range;
82}
83
84static int __init
85add_range_with_merge(struct res_range *range, int nr_range,
86 unsigned long start, unsigned long end)
87{
88 int i;
89
90 /* Try to merge it with old one: */
91 for (i = 0; i < nr_range; i++) {
92 unsigned long final_start, final_end;
93 unsigned long common_start, common_end;
94
95 if (!range[i].end)
96 continue;
97
98 common_start = max(range[i].start, start);
99 common_end = min(range[i].end, end);
100 if (common_start > common_end + 1)
101 continue;
102
103 final_start = min(range[i].start, start);
104 final_end = max(range[i].end, end);
105
106 range[i].start = final_start;
107 range[i].end = final_end;
108 return nr_range;
109 }
110
111 /* Need to add it: */
112 return add_range(range, nr_range, start, end);
113}
114
115static void __init
116subtract_range(struct res_range *range, unsigned long start, unsigned long end)
117{
118 int i, j;
119
120 for (j = 0; j < RANGE_NUM; j++) {
121 if (!range[j].end)
122 continue;
123
124 if (start <= range[j].start && end >= range[j].end) {
125 range[j].start = 0;
126 range[j].end = 0;
127 continue;
128 }
129
130 if (start <= range[j].start && end < range[j].end &&
131 range[j].start < end + 1) {
132 range[j].start = end + 1;
133 continue;
134 }
135
136
137 if (start > range[j].start && end >= range[j].end &&
138 range[j].end > start - 1) {
139 range[j].end = start - 1;
140 continue;
141 }
142
143 if (start > range[j].start && end < range[j].end) {
144 /* Find the new spare: */
145 for (i = 0; i < RANGE_NUM; i++) {
146 if (range[i].end == 0)
147 break;
148 }
149 if (i < RANGE_NUM) {
150 range[i].end = range[j].end;
151 range[i].start = end + 1;
152 } else {
153 printk(KERN_ERR "run of slot in ranges\n");
154 }
155 range[j].end = start - 1;
156 continue;
157 }
158 }
159}
160
161static int __init cmp_range(const void *x1, const void *x2)
162{
163 const struct res_range *r1 = x1;
164 const struct res_range *r2 = x2;
165 long start1, start2;
166
167 start1 = r1->start;
168 start2 = r2->start;
169
170 return start1 - start2;
171}
172
173static int __init clean_sort_range(struct res_range *range, int az)
174{
175 int i, j, k = az - 1, nr_range = 0;
176
177 for (i = 0; i < k; i++) {
178 if (range[i].end)
179 continue;
180 for (j = k; j > i; j--) {
181 if (range[j].end) {
182 k = j;
183 break;
184 }
185 }
186 if (j == i)
187 break;
188 range[i].start = range[k].start;
189 range[i].end = range[k].end;
190 range[k].start = 0;
191 range[k].end = 0;
192 k--;
193 }
194 /* count it */
195 for (i = 0; i < az; i++) {
196 if (!range[i].end) {
197 nr_range = i;
198 break;
199 }
200 }
201
202 /* sort them */
203 sort(range, nr_range, sizeof(struct res_range), cmp_range, NULL);
204
205 return nr_range;
206}
207
208#define BIOS_BUG_MSG KERN_WARNING \ 62#define BIOS_BUG_MSG KERN_WARNING \
209 "WARNING: BIOS bug: VAR MTRR %d contains strange UC entry under 1M, check with your system vendor!\n" 63 "WARNING: BIOS bug: VAR MTRR %d contains strange UC entry under 1M, check with your system vendor!\n"
210 64
211static int __init 65static int __init
212x86_get_mtrr_mem_range(struct res_range *range, int nr_range, 66x86_get_mtrr_mem_range(struct range *range, int nr_range,
213 unsigned long extra_remove_base, 67 unsigned long extra_remove_base,
214 unsigned long extra_remove_size) 68 unsigned long extra_remove_size)
215{ 69{
@@ -223,14 +77,14 @@ x86_get_mtrr_mem_range(struct res_range *range, int nr_range,
223 continue; 77 continue;
224 base = range_state[i].base_pfn; 78 base = range_state[i].base_pfn;
225 size = range_state[i].size_pfn; 79 size = range_state[i].size_pfn;
226 nr_range = add_range_with_merge(range, nr_range, base, 80 nr_range = add_range_with_merge(range, RANGE_NUM, nr_range,
227 base + size - 1); 81 base, base + size);
228 } 82 }
229 if (debug_print) { 83 if (debug_print) {
230 printk(KERN_DEBUG "After WB checking\n"); 84 printk(KERN_DEBUG "After WB checking\n");
231 for (i = 0; i < nr_range; i++) 85 for (i = 0; i < nr_range; i++)
232 printk(KERN_DEBUG "MTRR MAP PFN: %016lx - %016lx\n", 86 printk(KERN_DEBUG "MTRR MAP PFN: %016llx - %016llx\n",
233 range[i].start, range[i].end + 1); 87 range[i].start, range[i].end);
234 } 88 }
235 89
236 /* Take out UC ranges: */ 90 /* Take out UC ranges: */
@@ -252,19 +106,19 @@ x86_get_mtrr_mem_range(struct res_range *range, int nr_range,
252 size -= (1<<(20-PAGE_SHIFT)) - base; 106 size -= (1<<(20-PAGE_SHIFT)) - base;
253 base = 1<<(20-PAGE_SHIFT); 107 base = 1<<(20-PAGE_SHIFT);
254 } 108 }
255 subtract_range(range, base, base + size - 1); 109 subtract_range(range, RANGE_NUM, base, base + size);
256 } 110 }
257 if (extra_remove_size) 111 if (extra_remove_size)
258 subtract_range(range, extra_remove_base, 112 subtract_range(range, RANGE_NUM, extra_remove_base,
259 extra_remove_base + extra_remove_size - 1); 113 extra_remove_base + extra_remove_size);
260 114
261 if (debug_print) { 115 if (debug_print) {
262 printk(KERN_DEBUG "After UC checking\n"); 116 printk(KERN_DEBUG "After UC checking\n");
263 for (i = 0; i < RANGE_NUM; i++) { 117 for (i = 0; i < RANGE_NUM; i++) {
264 if (!range[i].end) 118 if (!range[i].end)
265 continue; 119 continue;
266 printk(KERN_DEBUG "MTRR MAP PFN: %016lx - %016lx\n", 120 printk(KERN_DEBUG "MTRR MAP PFN: %016llx - %016llx\n",
267 range[i].start, range[i].end + 1); 121 range[i].start, range[i].end);
268 } 122 }
269 } 123 }
270 124
@@ -273,26 +127,22 @@ x86_get_mtrr_mem_range(struct res_range *range, int nr_range,
273 if (debug_print) { 127 if (debug_print) {
274 printk(KERN_DEBUG "After sorting\n"); 128 printk(KERN_DEBUG "After sorting\n");
275 for (i = 0; i < nr_range; i++) 129 for (i = 0; i < nr_range; i++)
276 printk(KERN_DEBUG "MTRR MAP PFN: %016lx - %016lx\n", 130 printk(KERN_DEBUG "MTRR MAP PFN: %016llx - %016llx\n",
277 range[i].start, range[i].end + 1); 131 range[i].start, range[i].end);
278 } 132 }
279 133
280 /* clear those is not used */
281 for (i = nr_range; i < RANGE_NUM; i++)
282 memset(&range[i], 0, sizeof(range[i]));
283
284 return nr_range; 134 return nr_range;
285} 135}
286 136
287#ifdef CONFIG_MTRR_SANITIZER 137#ifdef CONFIG_MTRR_SANITIZER
288 138
289static unsigned long __init sum_ranges(struct res_range *range, int nr_range) 139static unsigned long __init sum_ranges(struct range *range, int nr_range)
290{ 140{
291 unsigned long sum = 0; 141 unsigned long sum = 0;
292 int i; 142 int i;
293 143
294 for (i = 0; i < nr_range; i++) 144 for (i = 0; i < nr_range; i++)
295 sum += range[i].end + 1 - range[i].start; 145 sum += range[i].end - range[i].start;
296 146
297 return sum; 147 return sum;
298} 148}
@@ -621,7 +471,7 @@ static int __init parse_mtrr_spare_reg(char *arg)
621early_param("mtrr_spare_reg_nr", parse_mtrr_spare_reg); 471early_param("mtrr_spare_reg_nr", parse_mtrr_spare_reg);
622 472
623static int __init 473static int __init
624x86_setup_var_mtrrs(struct res_range *range, int nr_range, 474x86_setup_var_mtrrs(struct range *range, int nr_range,
625 u64 chunk_size, u64 gran_size) 475 u64 chunk_size, u64 gran_size)
626{ 476{
627 struct var_mtrr_state var_state; 477 struct var_mtrr_state var_state;
@@ -639,7 +489,7 @@ x86_setup_var_mtrrs(struct res_range *range, int nr_range,
639 /* Write the range: */ 489 /* Write the range: */
640 for (i = 0; i < nr_range; i++) { 490 for (i = 0; i < nr_range; i++) {
641 set_var_mtrr_range(&var_state, range[i].start, 491 set_var_mtrr_range(&var_state, range[i].start,
642 range[i].end - range[i].start + 1); 492 range[i].end - range[i].start);
643 } 493 }
644 494
645 /* Write the last range: */ 495 /* Write the last range: */
@@ -742,7 +592,7 @@ mtrr_calc_range_state(u64 chunk_size, u64 gran_size,
742 unsigned long x_remove_base, 592 unsigned long x_remove_base,
743 unsigned long x_remove_size, int i) 593 unsigned long x_remove_size, int i)
744{ 594{
745 static struct res_range range_new[RANGE_NUM]; 595 static struct range range_new[RANGE_NUM];
746 unsigned long range_sums_new; 596 unsigned long range_sums_new;
747 static int nr_range_new; 597 static int nr_range_new;
748 int num_reg; 598 int num_reg;
@@ -869,10 +719,10 @@ int __init mtrr_cleanup(unsigned address_bits)
869 * [0, 1M) should always be covered by var mtrr with WB 719 * [0, 1M) should always be covered by var mtrr with WB
870 * and fixed mtrrs should take effect before var mtrr for it: 720 * and fixed mtrrs should take effect before var mtrr for it:
871 */ 721 */
872 nr_range = add_range_with_merge(range, nr_range, 0, 722 nr_range = add_range_with_merge(range, RANGE_NUM, nr_range, 0,
873 (1ULL<<(20 - PAGE_SHIFT)) - 1); 723 1ULL<<(20 - PAGE_SHIFT));
874 /* Sort the ranges: */ 724 /* Sort the ranges: */
875 sort(range, nr_range, sizeof(struct res_range), cmp_range, NULL); 725 sort_range(range, nr_range);
876 726
877 range_sums = sum_ranges(range, nr_range); 727 range_sums = sum_ranges(range, nr_range);
878 printk(KERN_INFO "total RAM covered: %ldM\n", 728 printk(KERN_INFO "total RAM covered: %ldM\n",
@@ -1089,9 +939,9 @@ int __init mtrr_trim_uncached_memory(unsigned long end_pfn)
1089 nr_range = 0; 939 nr_range = 0;
1090 if (mtrr_tom2) { 940 if (mtrr_tom2) {
1091 range[nr_range].start = (1ULL<<(32 - PAGE_SHIFT)); 941 range[nr_range].start = (1ULL<<(32 - PAGE_SHIFT));
1092 range[nr_range].end = (mtrr_tom2 >> PAGE_SHIFT) - 1; 942 range[nr_range].end = mtrr_tom2 >> PAGE_SHIFT;
1093 if (highest_pfn < range[nr_range].end + 1) 943 if (highest_pfn < range[nr_range].end)
1094 highest_pfn = range[nr_range].end + 1; 944 highest_pfn = range[nr_range].end;
1095 nr_range++; 945 nr_range++;
1096 } 946 }
1097 nr_range = x86_get_mtrr_mem_range(range, nr_range, 0, 0); 947 nr_range = x86_get_mtrr_mem_range(range, nr_range, 0, 0);
@@ -1103,15 +953,15 @@ int __init mtrr_trim_uncached_memory(unsigned long end_pfn)
1103 953
1104 /* Check the holes: */ 954 /* Check the holes: */
1105 for (i = 0; i < nr_range - 1; i++) { 955 for (i = 0; i < nr_range - 1; i++) {
1106 if (range[i].end + 1 < range[i+1].start) 956 if (range[i].end < range[i+1].start)
1107 total_trim_size += real_trim_memory(range[i].end + 1, 957 total_trim_size += real_trim_memory(range[i].end,
1108 range[i+1].start); 958 range[i+1].start);
1109 } 959 }
1110 960
1111 /* Check the top: */ 961 /* Check the top: */
1112 i = nr_range - 1; 962 i = nr_range - 1;
1113 if (range[i].end + 1 < end_pfn) 963 if (range[i].end < end_pfn)
1114 total_trim_size += real_trim_memory(range[i].end + 1, 964 total_trim_size += real_trim_memory(range[i].end,
1115 end_pfn); 965 end_pfn);
1116 966
1117 if (total_trim_size) { 967 if (total_trim_size) {
diff --git a/arch/x86/kernel/e820.c b/arch/x86/kernel/e820.c
index a966b753e496..740b440fbd73 100644
--- a/arch/x86/kernel/e820.c
+++ b/arch/x86/kernel/e820.c
@@ -12,21 +12,13 @@
12#include <linux/types.h> 12#include <linux/types.h>
13#include <linux/init.h> 13#include <linux/init.h>
14#include <linux/bootmem.h> 14#include <linux/bootmem.h>
15#include <linux/ioport.h>
16#include <linux/string.h>
17#include <linux/kexec.h>
18#include <linux/module.h>
19#include <linux/mm.h>
20#include <linux/pfn.h> 15#include <linux/pfn.h>
21#include <linux/suspend.h> 16#include <linux/suspend.h>
22#include <linux/firmware-map.h> 17#include <linux/firmware-map.h>
23 18
24#include <asm/pgtable.h>
25#include <asm/page.h>
26#include <asm/e820.h> 19#include <asm/e820.h>
27#include <asm/proto.h> 20#include <asm/proto.h>
28#include <asm/setup.h> 21#include <asm/setup.h>
29#include <asm/trampoline.h>
30 22
31/* 23/*
32 * The e820 map is the map that gets modified e.g. with command line parameters 24 * The e820 map is the map that gets modified e.g. with command line parameters
@@ -730,319 +722,44 @@ core_initcall(e820_mark_nvs_memory);
730#endif 722#endif
731 723
732/* 724/*
733 * Early reserved memory areas. 725 * Find a free area with specified alignment in a specific range.
734 */
735#define MAX_EARLY_RES 32
736
737struct early_res {
738 u64 start, end;
739 char name[16];
740 char overlap_ok;
741};
742static struct early_res early_res[MAX_EARLY_RES] __initdata = {
743 { 0, PAGE_SIZE, "BIOS data page", 1 }, /* BIOS data page */
744#if defined(CONFIG_X86_32) && defined(CONFIG_X86_TRAMPOLINE)
745 /*
746 * But first pinch a few for the stack/trampoline stuff
747 * FIXME: Don't need the extra page at 4K, but need to fix
748 * trampoline before removing it. (see the GDT stuff)
749 */
750 { PAGE_SIZE, PAGE_SIZE + PAGE_SIZE, "EX TRAMPOLINE", 1 },
751#endif
752
753 {}
754};
755
756static int __init find_overlapped_early(u64 start, u64 end)
757{
758 int i;
759 struct early_res *r;
760
761 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
762 r = &early_res[i];
763 if (end > r->start && start < r->end)
764 break;
765 }
766
767 return i;
768}
769
770/*
771 * Drop the i-th range from the early reservation map,
772 * by copying any higher ranges down one over it, and
773 * clearing what had been the last slot.
774 */
775static void __init drop_range(int i)
776{
777 int j;
778
779 for (j = i + 1; j < MAX_EARLY_RES && early_res[j].end; j++)
780 ;
781
782 memmove(&early_res[i], &early_res[i + 1],
783 (j - 1 - i) * sizeof(struct early_res));
784
785 early_res[j - 1].end = 0;
786}
787
788/*
789 * Split any existing ranges that:
790 * 1) are marked 'overlap_ok', and
791 * 2) overlap with the stated range [start, end)
792 * into whatever portion (if any) of the existing range is entirely
793 * below or entirely above the stated range. Drop the portion
794 * of the existing range that overlaps with the stated range,
795 * which will allow the caller of this routine to then add that
796 * stated range without conflicting with any existing range.
797 */ 726 */
798static void __init drop_overlaps_that_are_ok(u64 start, u64 end) 727u64 __init find_e820_area(u64 start, u64 end, u64 size, u64 align)
799{ 728{
800 int i; 729 int i;
801 struct early_res *r;
802 u64 lower_start, lower_end;
803 u64 upper_start, upper_end;
804 char name[16];
805 730
806 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) { 731 for (i = 0; i < e820.nr_map; i++) {
807 r = &early_res[i]; 732 struct e820entry *ei = &e820.map[i];
733 u64 addr;
734 u64 ei_start, ei_last;
808 735
809 /* Continue past non-overlapping ranges */ 736 if (ei->type != E820_RAM)
810 if (end <= r->start || start >= r->end)
811 continue; 737 continue;
812 738
813 /* 739 ei_last = ei->addr + ei->size;
814 * Leave non-ok overlaps as is; let caller 740 ei_start = ei->addr;
815 * panic "Overlapping early reservations" 741 addr = find_early_area(ei_start, ei_last, start, end,
816 * when it hits this overlap. 742 size, align);
817 */
818 if (!r->overlap_ok)
819 return;
820
821 /*
822 * We have an ok overlap. We will drop it from the early
823 * reservation map, and add back in any non-overlapping
824 * portions (lower or upper) as separate, overlap_ok,
825 * non-overlapping ranges.
826 */
827
828 /* 1. Note any non-overlapping (lower or upper) ranges. */
829 strncpy(name, r->name, sizeof(name) - 1);
830
831 lower_start = lower_end = 0;
832 upper_start = upper_end = 0;
833 if (r->start < start) {
834 lower_start = r->start;
835 lower_end = start;
836 }
837 if (r->end > end) {
838 upper_start = end;
839 upper_end = r->end;
840 }
841
842 /* 2. Drop the original ok overlapping range */
843 drop_range(i);
844
845 i--; /* resume for-loop on copied down entry */
846
847 /* 3. Add back in any non-overlapping ranges. */
848 if (lower_end)
849 reserve_early_overlap_ok(lower_start, lower_end, name);
850 if (upper_end)
851 reserve_early_overlap_ok(upper_start, upper_end, name);
852 }
853}
854
855static void __init __reserve_early(u64 start, u64 end, char *name,
856 int overlap_ok)
857{
858 int i;
859 struct early_res *r;
860
861 i = find_overlapped_early(start, end);
862 if (i >= MAX_EARLY_RES)
863 panic("Too many early reservations");
864 r = &early_res[i];
865 if (r->end)
866 panic("Overlapping early reservations "
867 "%llx-%llx %s to %llx-%llx %s\n",
868 start, end - 1, name?name:"", r->start,
869 r->end - 1, r->name);
870 r->start = start;
871 r->end = end;
872 r->overlap_ok = overlap_ok;
873 if (name)
874 strncpy(r->name, name, sizeof(r->name) - 1);
875}
876
877/*
878 * A few early reservtations come here.
879 *
880 * The 'overlap_ok' in the name of this routine does -not- mean it
881 * is ok for these reservations to overlap an earlier reservation.
882 * Rather it means that it is ok for subsequent reservations to
883 * overlap this one.
884 *
885 * Use this entry point to reserve early ranges when you are doing
886 * so out of "Paranoia", reserving perhaps more memory than you need,
887 * just in case, and don't mind a subsequent overlapping reservation
888 * that is known to be needed.
889 *
890 * The drop_overlaps_that_are_ok() call here isn't really needed.
891 * It would be needed if we had two colliding 'overlap_ok'
892 * reservations, so that the second such would not panic on the
893 * overlap with the first. We don't have any such as of this
894 * writing, but might as well tolerate such if it happens in
895 * the future.
896 */
897void __init reserve_early_overlap_ok(u64 start, u64 end, char *name)
898{
899 drop_overlaps_that_are_ok(start, end);
900 __reserve_early(start, end, name, 1);
901}
902
903/*
904 * Most early reservations come here.
905 *
906 * We first have drop_overlaps_that_are_ok() drop any pre-existing
907 * 'overlap_ok' ranges, so that we can then reserve this memory
908 * range without risk of panic'ing on an overlapping overlap_ok
909 * early reservation.
910 */
911void __init reserve_early(u64 start, u64 end, char *name)
912{
913 if (start >= end)
914 return;
915
916 drop_overlaps_that_are_ok(start, end);
917 __reserve_early(start, end, name, 0);
918}
919
920void __init free_early(u64 start, u64 end)
921{
922 struct early_res *r;
923 int i;
924
925 i = find_overlapped_early(start, end);
926 r = &early_res[i];
927 if (i >= MAX_EARLY_RES || r->end != end || r->start != start)
928 panic("free_early on not reserved area: %llx-%llx!",
929 start, end - 1);
930
931 drop_range(i);
932}
933
934void __init early_res_to_bootmem(u64 start, u64 end)
935{
936 int i, count;
937 u64 final_start, final_end;
938
939 count = 0;
940 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++)
941 count++;
942
943 printk(KERN_INFO "(%d early reservations) ==> bootmem [%010llx - %010llx]\n",
944 count, start, end);
945 for (i = 0; i < count; i++) {
946 struct early_res *r = &early_res[i];
947 printk(KERN_INFO " #%d [%010llx - %010llx] %16s", i,
948 r->start, r->end, r->name);
949 final_start = max(start, r->start);
950 final_end = min(end, r->end);
951 if (final_start >= final_end) {
952 printk(KERN_CONT "\n");
953 continue;
954 }
955 printk(KERN_CONT " ==> [%010llx - %010llx]\n",
956 final_start, final_end);
957 reserve_bootmem_generic(final_start, final_end - final_start,
958 BOOTMEM_DEFAULT);
959 }
960}
961 743
962/* Check for already reserved areas */ 744 if (addr != -1ULL)
963static inline int __init bad_addr(u64 *addrp, u64 size, u64 align) 745 return addr;
964{
965 int i;
966 u64 addr = *addrp;
967 int changed = 0;
968 struct early_res *r;
969again:
970 i = find_overlapped_early(addr, addr + size);
971 r = &early_res[i];
972 if (i < MAX_EARLY_RES && r->end) {
973 *addrp = addr = round_up(r->end, align);
974 changed = 1;
975 goto again;
976 } 746 }
977 return changed; 747 return -1ULL;
978} 748}
979 749
980/* Check for already reserved areas */ 750u64 __init find_fw_memmap_area(u64 start, u64 end, u64 size, u64 align)
981static inline int __init bad_addr_size(u64 *addrp, u64 *sizep, u64 align)
982{ 751{
983 int i; 752 return find_e820_area(start, end, size, align);
984 u64 addr = *addrp, last;
985 u64 size = *sizep;
986 int changed = 0;
987again:
988 last = addr + size;
989 for (i = 0; i < MAX_EARLY_RES && early_res[i].end; i++) {
990 struct early_res *r = &early_res[i];
991 if (last > r->start && addr < r->start) {
992 size = r->start - addr;
993 changed = 1;
994 goto again;
995 }
996 if (last > r->end && addr < r->end) {
997 addr = round_up(r->end, align);
998 size = last - addr;
999 changed = 1;
1000 goto again;
1001 }
1002 if (last <= r->end && addr >= r->start) {
1003 (*sizep)++;
1004 return 0;
1005 }
1006 }
1007 if (changed) {
1008 *addrp = addr;
1009 *sizep = size;
1010 }
1011 return changed;
1012} 753}
1013 754
1014/* 755u64 __init get_max_mapped(void)
1015 * Find a free area with specified alignment in a specific range.
1016 */
1017u64 __init find_e820_area(u64 start, u64 end, u64 size, u64 align)
1018{ 756{
1019 int i; 757 u64 end = max_pfn_mapped;
1020 758
1021 for (i = 0; i < e820.nr_map; i++) { 759 end <<= PAGE_SHIFT;
1022 struct e820entry *ei = &e820.map[i];
1023 u64 addr, last;
1024 u64 ei_last;
1025 760
1026 if (ei->type != E820_RAM) 761 return end;
1027 continue;
1028 addr = round_up(ei->addr, align);
1029 ei_last = ei->addr + ei->size;
1030 if (addr < start)
1031 addr = round_up(start, align);
1032 if (addr >= ei_last)
1033 continue;
1034 while (bad_addr(&addr, size, align) && addr+size <= ei_last)
1035 ;
1036 last = addr + size;
1037 if (last > ei_last)
1038 continue;
1039 if (last > end)
1040 continue;
1041 return addr;
1042 }
1043 return -1ULL;
1044} 762}
1045
1046/* 763/*
1047 * Find next free range after *start 764 * Find next free range after *start
1048 */ 765 */
@@ -1052,25 +769,19 @@ u64 __init find_e820_area_size(u64 start, u64 *sizep, u64 align)
1052 769
1053 for (i = 0; i < e820.nr_map; i++) { 770 for (i = 0; i < e820.nr_map; i++) {
1054 struct e820entry *ei = &e820.map[i]; 771 struct e820entry *ei = &e820.map[i];
1055 u64 addr, last; 772 u64 addr;
1056 u64 ei_last; 773 u64 ei_start, ei_last;
1057 774
1058 if (ei->type != E820_RAM) 775 if (ei->type != E820_RAM)
1059 continue; 776 continue;
1060 addr = round_up(ei->addr, align); 777
1061 ei_last = ei->addr + ei->size; 778 ei_last = ei->addr + ei->size;
1062 if (addr < start) 779 ei_start = ei->addr;
1063 addr = round_up(start, align); 780 addr = find_early_area_size(ei_start, ei_last, start,
1064 if (addr >= ei_last) 781 sizep, align);
1065 continue; 782
1066 *sizep = ei_last - addr; 783 if (addr != -1ULL)
1067 while (bad_addr_size(&addr, sizep, align) && 784 return addr;
1068 addr + *sizep <= ei_last)
1069 ;
1070 last = addr + *sizep;
1071 if (last > ei_last)
1072 continue;
1073 return addr;
1074 } 785 }
1075 786
1076 return -1ULL; 787 return -1ULL;
@@ -1429,6 +1140,8 @@ void __init e820_reserve_resources_late(void)
1429 end = MAX_RESOURCE_SIZE; 1140 end = MAX_RESOURCE_SIZE;
1430 if (start >= end) 1141 if (start >= end)
1431 continue; 1142 continue;
1143 printk(KERN_DEBUG "reserve RAM buffer: %016llx - %016llx ",
1144 start, end);
1432 reserve_region_with_split(&iomem_resource, start, end, 1145 reserve_region_with_split(&iomem_resource, start, end,
1433 "RAM buffer"); 1146 "RAM buffer");
1434 } 1147 }
diff --git a/arch/x86/kernel/head32.c b/arch/x86/kernel/head32.c
index 5051b94c9069..adedeef1dedc 100644
--- a/arch/x86/kernel/head32.c
+++ b/arch/x86/kernel/head32.c
@@ -29,6 +29,16 @@ static void __init i386_default_early_setup(void)
29 29
30void __init i386_start_kernel(void) 30void __init i386_start_kernel(void)
31{ 31{
32#ifdef CONFIG_X86_TRAMPOLINE
33 /*
34 * But first pinch a few for the stack/trampoline stuff
35 * FIXME: Don't need the extra page at 4K, but need to fix
36 * trampoline before removing it. (see the GDT stuff)
37 */
38 reserve_early_overlap_ok(PAGE_SIZE, PAGE_SIZE + PAGE_SIZE,
39 "EX TRAMPOLINE");
40#endif
41
32 reserve_early(__pa_symbol(&_text), __pa_symbol(&__bss_stop), "TEXT DATA BSS"); 42 reserve_early(__pa_symbol(&_text), __pa_symbol(&__bss_stop), "TEXT DATA BSS");
33 43
34#ifdef CONFIG_BLK_DEV_INITRD 44#ifdef CONFIG_BLK_DEV_INITRD
diff --git a/arch/x86/kernel/mmconf-fam10h_64.c b/arch/x86/kernel/mmconf-fam10h_64.c
index 712d15fdc416..71825806cd44 100644
--- a/arch/x86/kernel/mmconf-fam10h_64.c
+++ b/arch/x86/kernel/mmconf-fam10h_64.c
@@ -7,6 +7,8 @@
7#include <linux/string.h> 7#include <linux/string.h>
8#include <linux/pci.h> 8#include <linux/pci.h>
9#include <linux/dmi.h> 9#include <linux/dmi.h>
10#include <linux/range.h>
11
10#include <asm/pci-direct.h> 12#include <asm/pci-direct.h>
11#include <linux/sort.h> 13#include <linux/sort.h>
12#include <asm/io.h> 14#include <asm/io.h>
@@ -30,11 +32,6 @@ static struct pci_hostbridge_probe pci_probes[] __cpuinitdata = {
30 { 0xff, 0, PCI_VENDOR_ID_AMD, 0x1200 }, 32 { 0xff, 0, PCI_VENDOR_ID_AMD, 0x1200 },
31}; 33};
32 34
33struct range {
34 u64 start;
35 u64 end;
36};
37
38static int __cpuinit cmp_range(const void *x1, const void *x2) 35static int __cpuinit cmp_range(const void *x1, const void *x2)
39{ 36{
40 const struct range *r1 = x1; 37 const struct range *r1 = x1;
diff --git a/arch/x86/kernel/pci-dma.c b/arch/x86/kernel/pci-dma.c
index 75e14e21f61a..1aa966c565f9 100644
--- a/arch/x86/kernel/pci-dma.c
+++ b/arch/x86/kernel/pci-dma.c
@@ -65,7 +65,7 @@ int dma_set_mask(struct device *dev, u64 mask)
65} 65}
66EXPORT_SYMBOL(dma_set_mask); 66EXPORT_SYMBOL(dma_set_mask);
67 67
68#ifdef CONFIG_X86_64 68#if defined(CONFIG_X86_64) && !defined(CONFIG_NUMA)
69static __initdata void *dma32_bootmem_ptr; 69static __initdata void *dma32_bootmem_ptr;
70static unsigned long dma32_bootmem_size __initdata = (128ULL<<20); 70static unsigned long dma32_bootmem_size __initdata = (128ULL<<20);
71 71
@@ -116,14 +116,21 @@ static void __init dma32_free_bootmem(void)
116 dma32_bootmem_ptr = NULL; 116 dma32_bootmem_ptr = NULL;
117 dma32_bootmem_size = 0; 117 dma32_bootmem_size = 0;
118} 118}
119#else
120void __init dma32_reserve_bootmem(void)
121{
122}
123static void __init dma32_free_bootmem(void)
124{
125}
126
119#endif 127#endif
120 128
121void __init pci_iommu_alloc(void) 129void __init pci_iommu_alloc(void)
122{ 130{
123#ifdef CONFIG_X86_64
124 /* free the range so iommu could get some range less than 4G */ 131 /* free the range so iommu could get some range less than 4G */
125 dma32_free_bootmem(); 132 dma32_free_bootmem();
126#endif 133
127 if (pci_swiotlb_detect()) 134 if (pci_swiotlb_detect())
128 goto out; 135 goto out;
129 136
diff --git a/arch/x86/kernel/setup.c b/arch/x86/kernel/setup.c
index cb42109a55b4..5d7ba1a449bd 100644
--- a/arch/x86/kernel/setup.c
+++ b/arch/x86/kernel/setup.c
@@ -969,15 +969,11 @@ void __init setup_arch(char **cmdline_p)
969#endif 969#endif
970 970
971 initmem_init(0, max_pfn, acpi, k8); 971 initmem_init(0, max_pfn, acpi, k8);
972#ifndef CONFIG_NO_BOOTMEM
973 early_res_to_bootmem(0, max_low_pfn<<PAGE_SHIFT);
974#endif
972 975
973#ifdef CONFIG_X86_64
974 /*
975 * dma32_reserve_bootmem() allocates bootmem which may conflict
976 * with the crashkernel command line, so do that after
977 * reserve_crashkernel()
978 */
979 dma32_reserve_bootmem(); 976 dma32_reserve_bootmem();
980#endif
981 977
982 reserve_ibft_region(); 978 reserve_ibft_region();
983 979
diff --git a/arch/x86/kernel/setup_percpu.c b/arch/x86/kernel/setup_percpu.c
index 35abcb8b00e9..ef6370b00e70 100644
--- a/arch/x86/kernel/setup_percpu.c
+++ b/arch/x86/kernel/setup_percpu.c
@@ -137,7 +137,13 @@ static void * __init pcpu_fc_alloc(unsigned int cpu, size_t size, size_t align)
137 137
138static void __init pcpu_fc_free(void *ptr, size_t size) 138static void __init pcpu_fc_free(void *ptr, size_t size)
139{ 139{
140#ifdef CONFIG_NO_BOOTMEM
141 u64 start = __pa(ptr);
142 u64 end = start + size;
143 free_early_partial(start, end);
144#else
140 free_bootmem(__pa(ptr), size); 145 free_bootmem(__pa(ptr), size);
146#endif
141} 147}
142 148
143static int __init pcpu_cpu_distance(unsigned int from, unsigned int to) 149static int __init pcpu_cpu_distance(unsigned int from, unsigned int to)
diff --git a/arch/x86/mm/init_32.c b/arch/x86/mm/init_32.c
index 2226f2c70ea3..5cb3f0f54f47 100644
--- a/arch/x86/mm/init_32.c
+++ b/arch/x86/mm/init_32.c
@@ -750,6 +750,7 @@ static void __init zone_sizes_init(void)
750 free_area_init_nodes(max_zone_pfns); 750 free_area_init_nodes(max_zone_pfns);
751} 751}
752 752
753#ifndef CONFIG_NO_BOOTMEM
753static unsigned long __init setup_node_bootmem(int nodeid, 754static unsigned long __init setup_node_bootmem(int nodeid,
754 unsigned long start_pfn, 755 unsigned long start_pfn,
755 unsigned long end_pfn, 756 unsigned long end_pfn,
@@ -766,13 +767,14 @@ static unsigned long __init setup_node_bootmem(int nodeid,
766 printk(KERN_INFO " node %d bootmap %08lx - %08lx\n", 767 printk(KERN_INFO " node %d bootmap %08lx - %08lx\n",
767 nodeid, bootmap, bootmap + bootmap_size); 768 nodeid, bootmap, bootmap + bootmap_size);
768 free_bootmem_with_active_regions(nodeid, end_pfn); 769 free_bootmem_with_active_regions(nodeid, end_pfn);
769 early_res_to_bootmem(start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
770 770
771 return bootmap + bootmap_size; 771 return bootmap + bootmap_size;
772} 772}
773#endif
773 774
774void __init setup_bootmem_allocator(void) 775void __init setup_bootmem_allocator(void)
775{ 776{
777#ifndef CONFIG_NO_BOOTMEM
776 int nodeid; 778 int nodeid;
777 unsigned long bootmap_size, bootmap; 779 unsigned long bootmap_size, bootmap;
778 /* 780 /*
@@ -784,11 +786,13 @@ void __init setup_bootmem_allocator(void)
784 if (bootmap == -1L) 786 if (bootmap == -1L)
785 panic("Cannot find bootmem map of size %ld\n", bootmap_size); 787 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
786 reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP"); 788 reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
789#endif
787 790
788 printk(KERN_INFO " mapped low ram: 0 - %08lx\n", 791 printk(KERN_INFO " mapped low ram: 0 - %08lx\n",
789 max_pfn_mapped<<PAGE_SHIFT); 792 max_pfn_mapped<<PAGE_SHIFT);
790 printk(KERN_INFO " low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT); 793 printk(KERN_INFO " low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
791 794
795#ifndef CONFIG_NO_BOOTMEM
792 for_each_online_node(nodeid) { 796 for_each_online_node(nodeid) {
793 unsigned long start_pfn, end_pfn; 797 unsigned long start_pfn, end_pfn;
794 798
@@ -806,6 +810,7 @@ void __init setup_bootmem_allocator(void)
806 bootmap = setup_node_bootmem(nodeid, start_pfn, end_pfn, 810 bootmap = setup_node_bootmem(nodeid, start_pfn, end_pfn,
807 bootmap); 811 bootmap);
808 } 812 }
813#endif
809 814
810 after_bootmem = 1; 815 after_bootmem = 1;
811} 816}
diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
index 69ddfbd91135..e9b040e1cde5 100644
--- a/arch/x86/mm/init_64.c
+++ b/arch/x86/mm/init_64.c
@@ -572,6 +572,7 @@ kernel_physical_mapping_init(unsigned long start,
572void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn, 572void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn,
573 int acpi, int k8) 573 int acpi, int k8)
574{ 574{
575#ifndef CONFIG_NO_BOOTMEM
575 unsigned long bootmap_size, bootmap; 576 unsigned long bootmap_size, bootmap;
576 577
577 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT; 578 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
@@ -579,13 +580,15 @@ void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn,
579 PAGE_SIZE); 580 PAGE_SIZE);
580 if (bootmap == -1L) 581 if (bootmap == -1L)
581 panic("Cannot find bootmem map of size %ld\n", bootmap_size); 582 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
583 reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
582 /* don't touch min_low_pfn */ 584 /* don't touch min_low_pfn */
583 bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT, 585 bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT,
584 0, end_pfn); 586 0, end_pfn);
585 e820_register_active_regions(0, start_pfn, end_pfn); 587 e820_register_active_regions(0, start_pfn, end_pfn);
586 free_bootmem_with_active_regions(0, end_pfn); 588 free_bootmem_with_active_regions(0, end_pfn);
587 early_res_to_bootmem(0, end_pfn<<PAGE_SHIFT); 589#else
588 reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT); 590 e820_register_active_regions(0, start_pfn, end_pfn);
591#endif
589} 592}
590#endif 593#endif
591 594
@@ -974,7 +977,7 @@ vmemmap_populate(struct page *start_page, unsigned long size, int node)
974 if (pmd_none(*pmd)) { 977 if (pmd_none(*pmd)) {
975 pte_t entry; 978 pte_t entry;
976 979
977 p = vmemmap_alloc_block(PMD_SIZE, node); 980 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
978 if (!p) 981 if (!p)
979 return -ENOMEM; 982 return -ENOMEM;
980 983
diff --git a/arch/x86/mm/numa_32.c b/arch/x86/mm/numa_32.c
index b20760ca7244..809baaaf48b1 100644
--- a/arch/x86/mm/numa_32.c
+++ b/arch/x86/mm/numa_32.c
@@ -418,7 +418,10 @@ void __init initmem_init(unsigned long start_pfn, unsigned long end_pfn,
418 418
419 for_each_online_node(nid) { 419 for_each_online_node(nid) {
420 memset(NODE_DATA(nid), 0, sizeof(struct pglist_data)); 420 memset(NODE_DATA(nid), 0, sizeof(struct pglist_data));
421 NODE_DATA(nid)->node_id = nid;
422#ifndef CONFIG_NO_BOOTMEM
421 NODE_DATA(nid)->bdata = &bootmem_node_data[nid]; 423 NODE_DATA(nid)->bdata = &bootmem_node_data[nid];
424#endif
422 } 425 }
423 426
424 setup_bootmem_allocator(); 427 setup_bootmem_allocator();
diff --git a/arch/x86/mm/numa_64.c b/arch/x86/mm/numa_64.c
index 3307ea8bd43a..8948f47fde05 100644
--- a/arch/x86/mm/numa_64.c
+++ b/arch/x86/mm/numa_64.c
@@ -163,30 +163,48 @@ static void * __init early_node_mem(int nodeid, unsigned long start,
163 unsigned long end, unsigned long size, 163 unsigned long end, unsigned long size,
164 unsigned long align) 164 unsigned long align)
165{ 165{
166 unsigned long mem = find_e820_area(start, end, size, align); 166 unsigned long mem;
167 void *ptr;
168 167
168 /*
169 * put it on high as possible
170 * something will go with NODE_DATA
171 */
172 if (start < (MAX_DMA_PFN<<PAGE_SHIFT))
173 start = MAX_DMA_PFN<<PAGE_SHIFT;
174 if (start < (MAX_DMA32_PFN<<PAGE_SHIFT) &&
175 end > (MAX_DMA32_PFN<<PAGE_SHIFT))
176 start = MAX_DMA32_PFN<<PAGE_SHIFT;
177 mem = find_e820_area(start, end, size, align);
178 if (mem != -1L)
179 return __va(mem);
180
181 /* extend the search scope */
182 end = max_pfn_mapped << PAGE_SHIFT;
183 if (end > (MAX_DMA32_PFN<<PAGE_SHIFT))
184 start = MAX_DMA32_PFN<<PAGE_SHIFT;
185 else
186 start = MAX_DMA_PFN<<PAGE_SHIFT;
187 mem = find_e820_area(start, end, size, align);
169 if (mem != -1L) 188 if (mem != -1L)
170 return __va(mem); 189 return __va(mem);
171 190
172 ptr = __alloc_bootmem_nopanic(size, align, __pa(MAX_DMA_ADDRESS)); 191 printk(KERN_ERR "Cannot find %lu bytes in node %d\n",
173 if (ptr == NULL) {
174 printk(KERN_ERR "Cannot find %lu bytes in node %d\n",
175 size, nodeid); 192 size, nodeid);
176 return NULL; 193
177 } 194 return NULL;
178 return ptr;
179} 195}
180 196
181/* Initialize bootmem allocator for a node */ 197/* Initialize bootmem allocator for a node */
182void __init 198void __init
183setup_node_bootmem(int nodeid, unsigned long start, unsigned long end) 199setup_node_bootmem(int nodeid, unsigned long start, unsigned long end)
184{ 200{
185 unsigned long start_pfn, last_pfn, bootmap_pages, bootmap_size; 201 unsigned long start_pfn, last_pfn, nodedata_phys;
186 const int pgdat_size = roundup(sizeof(pg_data_t), PAGE_SIZE); 202 const int pgdat_size = roundup(sizeof(pg_data_t), PAGE_SIZE);
187 unsigned long bootmap_start, nodedata_phys;
188 void *bootmap;
189 int nid; 203 int nid;
204#ifndef CONFIG_NO_BOOTMEM
205 unsigned long bootmap_start, bootmap_pages, bootmap_size;
206 void *bootmap;
207#endif
190 208
191 if (!end) 209 if (!end)
192 return; 210 return;
@@ -200,7 +218,7 @@ setup_node_bootmem(int nodeid, unsigned long start, unsigned long end)
200 218
201 start = roundup(start, ZONE_ALIGN); 219 start = roundup(start, ZONE_ALIGN);
202 220
203 printk(KERN_INFO "Bootmem setup node %d %016lx-%016lx\n", nodeid, 221 printk(KERN_INFO "Initmem setup node %d %016lx-%016lx\n", nodeid,
204 start, end); 222 start, end);
205 223
206 start_pfn = start >> PAGE_SHIFT; 224 start_pfn = start >> PAGE_SHIFT;
@@ -211,14 +229,21 @@ setup_node_bootmem(int nodeid, unsigned long start, unsigned long end)
211 if (node_data[nodeid] == NULL) 229 if (node_data[nodeid] == NULL)
212 return; 230 return;
213 nodedata_phys = __pa(node_data[nodeid]); 231 nodedata_phys = __pa(node_data[nodeid]);
232 reserve_early(nodedata_phys, nodedata_phys + pgdat_size, "NODE_DATA");
214 printk(KERN_INFO " NODE_DATA [%016lx - %016lx]\n", nodedata_phys, 233 printk(KERN_INFO " NODE_DATA [%016lx - %016lx]\n", nodedata_phys,
215 nodedata_phys + pgdat_size - 1); 234 nodedata_phys + pgdat_size - 1);
235 nid = phys_to_nid(nodedata_phys);
236 if (nid != nodeid)
237 printk(KERN_INFO " NODE_DATA(%d) on node %d\n", nodeid, nid);
216 238
217 memset(NODE_DATA(nodeid), 0, sizeof(pg_data_t)); 239 memset(NODE_DATA(nodeid), 0, sizeof(pg_data_t));
218 NODE_DATA(nodeid)->bdata = &bootmem_node_data[nodeid]; 240 NODE_DATA(nodeid)->node_id = nodeid;
219 NODE_DATA(nodeid)->node_start_pfn = start_pfn; 241 NODE_DATA(nodeid)->node_start_pfn = start_pfn;
220 NODE_DATA(nodeid)->node_spanned_pages = last_pfn - start_pfn; 242 NODE_DATA(nodeid)->node_spanned_pages = last_pfn - start_pfn;
221 243
244#ifndef CONFIG_NO_BOOTMEM
245 NODE_DATA(nodeid)->bdata = &bootmem_node_data[nodeid];
246
222 /* 247 /*
223 * Find a place for the bootmem map 248 * Find a place for the bootmem map
224 * nodedata_phys could be on other nodes by alloc_bootmem, 249 * nodedata_phys could be on other nodes by alloc_bootmem,
@@ -227,11 +252,7 @@ setup_node_bootmem(int nodeid, unsigned long start, unsigned long end)
227 * of alloc_bootmem, that could clash with reserved range 252 * of alloc_bootmem, that could clash with reserved range
228 */ 253 */
229 bootmap_pages = bootmem_bootmap_pages(last_pfn - start_pfn); 254 bootmap_pages = bootmem_bootmap_pages(last_pfn - start_pfn);
230 nid = phys_to_nid(nodedata_phys); 255 bootmap_start = roundup(nodedata_phys + pgdat_size, PAGE_SIZE);
231 if (nid == nodeid)
232 bootmap_start = roundup(nodedata_phys + pgdat_size, PAGE_SIZE);
233 else
234 bootmap_start = roundup(start, PAGE_SIZE);
235 /* 256 /*
236 * SMP_CACHE_BYTES could be enough, but init_bootmem_node like 257 * SMP_CACHE_BYTES could be enough, but init_bootmem_node like
237 * to use that to align to PAGE_SIZE 258 * to use that to align to PAGE_SIZE
@@ -239,18 +260,13 @@ setup_node_bootmem(int nodeid, unsigned long start, unsigned long end)
239 bootmap = early_node_mem(nodeid, bootmap_start, end, 260 bootmap = early_node_mem(nodeid, bootmap_start, end,
240 bootmap_pages<<PAGE_SHIFT, PAGE_SIZE); 261 bootmap_pages<<PAGE_SHIFT, PAGE_SIZE);
241 if (bootmap == NULL) { 262 if (bootmap == NULL) {
242 if (nodedata_phys < start || nodedata_phys >= end) { 263 free_early(nodedata_phys, nodedata_phys + pgdat_size);
243 /*
244 * only need to free it if it is from other node
245 * bootmem
246 */
247 if (nid != nodeid)
248 free_bootmem(nodedata_phys, pgdat_size);
249 }
250 node_data[nodeid] = NULL; 264 node_data[nodeid] = NULL;
251 return; 265 return;
252 } 266 }
253 bootmap_start = __pa(bootmap); 267 bootmap_start = __pa(bootmap);
268 reserve_early(bootmap_start, bootmap_start+(bootmap_pages<<PAGE_SHIFT),
269 "BOOTMAP");
254 270
255 bootmap_size = init_bootmem_node(NODE_DATA(nodeid), 271 bootmap_size = init_bootmem_node(NODE_DATA(nodeid),
256 bootmap_start >> PAGE_SHIFT, 272 bootmap_start >> PAGE_SHIFT,
@@ -259,31 +275,12 @@ setup_node_bootmem(int nodeid, unsigned long start, unsigned long end)
259 printk(KERN_INFO " bootmap [%016lx - %016lx] pages %lx\n", 275 printk(KERN_INFO " bootmap [%016lx - %016lx] pages %lx\n",
260 bootmap_start, bootmap_start + bootmap_size - 1, 276 bootmap_start, bootmap_start + bootmap_size - 1,
261 bootmap_pages); 277 bootmap_pages);
262
263 free_bootmem_with_active_regions(nodeid, end);
264
265 /*
266 * convert early reserve to bootmem reserve earlier
267 * otherwise early_node_mem could use early reserved mem
268 * on previous node
269 */
270 early_res_to_bootmem(start, end);
271
272 /*
273 * in some case early_node_mem could use alloc_bootmem
274 * to get range on other node, don't reserve that again
275 */
276 if (nid != nodeid)
277 printk(KERN_INFO " NODE_DATA(%d) on node %d\n", nodeid, nid);
278 else
279 reserve_bootmem_node(NODE_DATA(nodeid), nodedata_phys,
280 pgdat_size, BOOTMEM_DEFAULT);
281 nid = phys_to_nid(bootmap_start); 278 nid = phys_to_nid(bootmap_start);
282 if (nid != nodeid) 279 if (nid != nodeid)
283 printk(KERN_INFO " bootmap(%d) on node %d\n", nodeid, nid); 280 printk(KERN_INFO " bootmap(%d) on node %d\n", nodeid, nid);
284 else 281
285 reserve_bootmem_node(NODE_DATA(nodeid), bootmap_start, 282 free_bootmem_with_active_regions(nodeid, end);
286 bootmap_pages<<PAGE_SHIFT, BOOTMEM_DEFAULT); 283#endif
287 284
288 node_set_online(nodeid); 285 node_set_online(nodeid);
289} 286}
@@ -709,6 +706,10 @@ unsigned long __init numa_free_all_bootmem(void)
709 for_each_online_node(i) 706 for_each_online_node(i)
710 pages += free_all_bootmem_node(NODE_DATA(i)); 707 pages += free_all_bootmem_node(NODE_DATA(i));
711 708
709#ifdef CONFIG_NO_BOOTMEM
710 pages += free_all_memory_core_early(MAX_NUMNODES);
711#endif
712
712 return pages; 713 return pages;
713} 714}
714 715
diff --git a/arch/x86/pci/Makefile b/arch/x86/pci/Makefile
index 39fba37f702f..0b7d3e9593e1 100644
--- a/arch/x86/pci/Makefile
+++ b/arch/x86/pci/Makefile
@@ -14,8 +14,7 @@ obj-$(CONFIG_X86_VISWS) += visws.o
14obj-$(CONFIG_X86_NUMAQ) += numaq_32.o 14obj-$(CONFIG_X86_NUMAQ) += numaq_32.o
15 15
16obj-y += common.o early.o 16obj-y += common.o early.o
17obj-y += amd_bus.o 17obj-y += amd_bus.o bus_numa.o
18obj-$(CONFIG_X86_64) += bus_numa.o
19 18
20ifeq ($(CONFIG_PCI_DEBUG),y) 19ifeq ($(CONFIG_PCI_DEBUG),y)
21EXTRA_CFLAGS += -DDEBUG 20EXTRA_CFLAGS += -DDEBUG
diff --git a/arch/x86/pci/amd_bus.c b/arch/x86/pci/amd_bus.c
index 95ecbd495955..fc1e8fe07e5c 100644
--- a/arch/x86/pci/amd_bus.c
+++ b/arch/x86/pci/amd_bus.c
@@ -2,11 +2,11 @@
2#include <linux/pci.h> 2#include <linux/pci.h>
3#include <linux/topology.h> 3#include <linux/topology.h>
4#include <linux/cpu.h> 4#include <linux/cpu.h>
5#include <linux/range.h>
6
5#include <asm/pci_x86.h> 7#include <asm/pci_x86.h>
6 8
7#ifdef CONFIG_X86_64
8#include <asm/pci-direct.h> 9#include <asm/pci-direct.h>
9#endif
10 10
11#include "bus_numa.h" 11#include "bus_numa.h"
12 12
@@ -15,60 +15,6 @@
15 * also get peer root bus resource for io,mmio 15 * also get peer root bus resource for io,mmio
16 */ 16 */
17 17
18#ifdef CONFIG_X86_64
19
20#define RANGE_NUM 16
21
22struct res_range {
23 size_t start;
24 size_t end;
25};
26
27static void __init update_range(struct res_range *range, size_t start,
28 size_t end)
29{
30 int i;
31 int j;
32
33 for (j = 0; j < RANGE_NUM; j++) {
34 if (!range[j].end)
35 continue;
36
37 if (start <= range[j].start && end >= range[j].end) {
38 range[j].start = 0;
39 range[j].end = 0;
40 continue;
41 }
42
43 if (start <= range[j].start && end < range[j].end && range[j].start < end + 1) {
44 range[j].start = end + 1;
45 continue;
46 }
47
48
49 if (start > range[j].start && end >= range[j].end && range[j].end > start - 1) {
50 range[j].end = start - 1;
51 continue;
52 }
53
54 if (start > range[j].start && end < range[j].end) {
55 /* find the new spare */
56 for (i = 0; i < RANGE_NUM; i++) {
57 if (range[i].end == 0)
58 break;
59 }
60 if (i < RANGE_NUM) {
61 range[i].end = range[j].end;
62 range[i].start = end + 1;
63 } else {
64 printk(KERN_ERR "run of slot in ranges\n");
65 }
66 range[j].end = start - 1;
67 continue;
68 }
69 }
70}
71
72struct pci_hostbridge_probe { 18struct pci_hostbridge_probe {
73 u32 bus; 19 u32 bus;
74 u32 slot; 20 u32 slot;
@@ -111,6 +57,8 @@ static void __init get_pci_mmcfg_amd_fam10h_range(void)
111 fam10h_mmconf_end = base + (1ULL<<(segn_busn_bits + 20)) - 1; 57 fam10h_mmconf_end = base + (1ULL<<(segn_busn_bits + 20)) - 1;
112} 58}
113 59
60#define RANGE_NUM 16
61
114/** 62/**
115 * early_fill_mp_bus_to_node() 63 * early_fill_mp_bus_to_node()
116 * called before pcibios_scan_root and pci_scan_bus 64 * called before pcibios_scan_root and pci_scan_bus
@@ -130,16 +78,17 @@ static int __init early_fill_mp_bus_info(void)
130 struct pci_root_info *info; 78 struct pci_root_info *info;
131 u32 reg; 79 u32 reg;
132 struct resource *res; 80 struct resource *res;
133 size_t start; 81 u64 start;
134 size_t end; 82 u64 end;
135 struct res_range range[RANGE_NUM]; 83 struct range range[RANGE_NUM];
136 u64 val; 84 u64 val;
137 u32 address; 85 u32 address;
86 bool found;
138 87
139 if (!early_pci_allowed()) 88 if (!early_pci_allowed())
140 return -1; 89 return -1;
141 90
142 found_all_numa_early = 0; 91 found = false;
143 for (i = 0; i < ARRAY_SIZE(pci_probes); i++) { 92 for (i = 0; i < ARRAY_SIZE(pci_probes); i++) {
144 u32 id; 93 u32 id;
145 u16 device; 94 u16 device;
@@ -153,12 +102,12 @@ static int __init early_fill_mp_bus_info(void)
153 device = (id>>16) & 0xffff; 102 device = (id>>16) & 0xffff;
154 if (pci_probes[i].vendor == vendor && 103 if (pci_probes[i].vendor == vendor &&
155 pci_probes[i].device == device) { 104 pci_probes[i].device == device) {
156 found_all_numa_early = 1; 105 found = true;
157 break; 106 break;
158 } 107 }
159 } 108 }
160 109
161 if (!found_all_numa_early) 110 if (!found)
162 return 0; 111 return 0;
163 112
164 pci_root_num = 0; 113 pci_root_num = 0;
@@ -196,7 +145,7 @@ static int __init early_fill_mp_bus_info(void)
196 def_link = (reg >> 8) & 0x03; 145 def_link = (reg >> 8) & 0x03;
197 146
198 memset(range, 0, sizeof(range)); 147 memset(range, 0, sizeof(range));
199 range[0].end = 0xffff; 148 add_range(range, RANGE_NUM, 0, 0, 0xffff + 1);
200 /* io port resource */ 149 /* io port resource */
201 for (i = 0; i < 4; i++) { 150 for (i = 0; i < 4; i++) {
202 reg = read_pci_config(bus, slot, 1, 0xc0 + (i << 3)); 151 reg = read_pci_config(bus, slot, 1, 0xc0 + (i << 3));
@@ -220,13 +169,13 @@ static int __init early_fill_mp_bus_info(void)
220 169
221 info = &pci_root_info[j]; 170 info = &pci_root_info[j];
222 printk(KERN_DEBUG "node %d link %d: io port [%llx, %llx]\n", 171 printk(KERN_DEBUG "node %d link %d: io port [%llx, %llx]\n",
223 node, link, (u64)start, (u64)end); 172 node, link, start, end);
224 173
225 /* kernel only handle 16 bit only */ 174 /* kernel only handle 16 bit only */
226 if (end > 0xffff) 175 if (end > 0xffff)
227 end = 0xffff; 176 end = 0xffff;
228 update_res(info, start, end, IORESOURCE_IO, 1); 177 update_res(info, start, end, IORESOURCE_IO, 1);
229 update_range(range, start, end); 178 subtract_range(range, RANGE_NUM, start, end + 1);
230 } 179 }
231 /* add left over io port range to def node/link, [0, 0xffff] */ 180 /* add left over io port range to def node/link, [0, 0xffff] */
232 /* find the position */ 181 /* find the position */
@@ -241,29 +190,32 @@ static int __init early_fill_mp_bus_info(void)
241 if (!range[i].end) 190 if (!range[i].end)
242 continue; 191 continue;
243 192
244 update_res(info, range[i].start, range[i].end, 193 update_res(info, range[i].start, range[i].end - 1,
245 IORESOURCE_IO, 1); 194 IORESOURCE_IO, 1);
246 } 195 }
247 } 196 }
248 197
249 memset(range, 0, sizeof(range)); 198 memset(range, 0, sizeof(range));
250 /* 0xfd00000000-0xffffffffff for HT */ 199 /* 0xfd00000000-0xffffffffff for HT */
251 range[0].end = (0xfdULL<<32) - 1; 200 end = cap_resource((0xfdULL<<32) - 1);
201 end++;
202 add_range(range, RANGE_NUM, 0, 0, end);
252 203
253 /* need to take out [0, TOM) for RAM*/ 204 /* need to take out [0, TOM) for RAM*/
254 address = MSR_K8_TOP_MEM1; 205 address = MSR_K8_TOP_MEM1;
255 rdmsrl(address, val); 206 rdmsrl(address, val);
256 end = (val & 0xffffff800000ULL); 207 end = (val & 0xffffff800000ULL);
257 printk(KERN_INFO "TOM: %016lx aka %ldM\n", end, end>>20); 208 printk(KERN_INFO "TOM: %016llx aka %lldM\n", end, end>>20);
258 if (end < (1ULL<<32)) 209 if (end < (1ULL<<32))
259 update_range(range, 0, end - 1); 210 subtract_range(range, RANGE_NUM, 0, end);
260 211
261 /* get mmconfig */ 212 /* get mmconfig */
262 get_pci_mmcfg_amd_fam10h_range(); 213 get_pci_mmcfg_amd_fam10h_range();
263 /* need to take out mmconf range */ 214 /* need to take out mmconf range */
264 if (fam10h_mmconf_end) { 215 if (fam10h_mmconf_end) {
265 printk(KERN_DEBUG "Fam 10h mmconf [%llx, %llx]\n", fam10h_mmconf_start, fam10h_mmconf_end); 216 printk(KERN_DEBUG "Fam 10h mmconf [%llx, %llx]\n", fam10h_mmconf_start, fam10h_mmconf_end);
266 update_range(range, fam10h_mmconf_start, fam10h_mmconf_end); 217 subtract_range(range, RANGE_NUM, fam10h_mmconf_start,
218 fam10h_mmconf_end + 1);
267 } 219 }
268 220
269 /* mmio resource */ 221 /* mmio resource */
@@ -293,7 +245,7 @@ static int __init early_fill_mp_bus_info(void)
293 info = &pci_root_info[j]; 245 info = &pci_root_info[j];
294 246
295 printk(KERN_DEBUG "node %d link %d: mmio [%llx, %llx]", 247 printk(KERN_DEBUG "node %d link %d: mmio [%llx, %llx]",
296 node, link, (u64)start, (u64)end); 248 node, link, start, end);
297 /* 249 /*
298 * some sick allocation would have range overlap with fam10h 250 * some sick allocation would have range overlap with fam10h
299 * mmconf range, so need to update start and end. 251 * mmconf range, so need to update start and end.
@@ -318,14 +270,15 @@ static int __init early_fill_mp_bus_info(void)
318 /* we got a hole */ 270 /* we got a hole */
319 endx = fam10h_mmconf_start - 1; 271 endx = fam10h_mmconf_start - 1;
320 update_res(info, start, endx, IORESOURCE_MEM, 0); 272 update_res(info, start, endx, IORESOURCE_MEM, 0);
321 update_range(range, start, endx); 273 subtract_range(range, RANGE_NUM, start,
322 printk(KERN_CONT " ==> [%llx, %llx]", (u64)start, endx); 274 endx + 1);
275 printk(KERN_CONT " ==> [%llx, %llx]", start, endx);
323 start = fam10h_mmconf_end + 1; 276 start = fam10h_mmconf_end + 1;
324 changed = 1; 277 changed = 1;
325 } 278 }
326 if (changed) { 279 if (changed) {
327 if (start <= end) { 280 if (start <= end) {
328 printk(KERN_CONT " %s [%llx, %llx]", endx?"and":"==>", (u64)start, (u64)end); 281 printk(KERN_CONT " %s [%llx, %llx]", endx ? "and" : "==>", start, end);
329 } else { 282 } else {
330 printk(KERN_CONT "%s\n", endx?"":" ==> none"); 283 printk(KERN_CONT "%s\n", endx?"":" ==> none");
331 continue; 284 continue;
@@ -333,8 +286,9 @@ static int __init early_fill_mp_bus_info(void)
333 } 286 }
334 } 287 }
335 288
336 update_res(info, start, end, IORESOURCE_MEM, 1); 289 update_res(info, cap_resource(start), cap_resource(end),
337 update_range(range, start, end); 290 IORESOURCE_MEM, 1);
291 subtract_range(range, RANGE_NUM, start, end + 1);
338 printk(KERN_CONT "\n"); 292 printk(KERN_CONT "\n");
339 } 293 }
340 294
@@ -348,8 +302,8 @@ static int __init early_fill_mp_bus_info(void)
348 address = MSR_K8_TOP_MEM2; 302 address = MSR_K8_TOP_MEM2;
349 rdmsrl(address, val); 303 rdmsrl(address, val);
350 end = (val & 0xffffff800000ULL); 304 end = (val & 0xffffff800000ULL);
351 printk(KERN_INFO "TOM2: %016lx aka %ldM\n", end, end>>20); 305 printk(KERN_INFO "TOM2: %016llx aka %lldM\n", end, end>>20);
352 update_range(range, 1ULL<<32, end - 1); 306 subtract_range(range, RANGE_NUM, 1ULL<<32, end);
353 } 307 }
354 308
355 /* 309 /*
@@ -368,7 +322,8 @@ static int __init early_fill_mp_bus_info(void)
368 if (!range[i].end) 322 if (!range[i].end)
369 continue; 323 continue;
370 324
371 update_res(info, range[i].start, range[i].end, 325 update_res(info, cap_resource(range[i].start),
326 cap_resource(range[i].end - 1),
372 IORESOURCE_MEM, 1); 327 IORESOURCE_MEM, 1);
373 } 328 }
374 } 329 }
@@ -384,24 +339,14 @@ static int __init early_fill_mp_bus_info(void)
384 info->bus_min, info->bus_max, info->node, info->link); 339 info->bus_min, info->bus_max, info->node, info->link);
385 for (j = 0; j < res_num; j++) { 340 for (j = 0; j < res_num; j++) {
386 res = &info->res[j]; 341 res = &info->res[j];
387 printk(KERN_DEBUG "bus: %02x index %x %s: [%llx, %llx]\n", 342 printk(KERN_DEBUG "bus: %02x index %x %pR\n",
388 busnum, j, 343 busnum, j, res);
389 (res->flags & IORESOURCE_IO)?"io port":"mmio",
390 res->start, res->end);
391 } 344 }
392 } 345 }
393 346
394 return 0; 347 return 0;
395} 348}
396 349
397#else /* !CONFIG_X86_64 */
398
399static int __init early_fill_mp_bus_info(void) { return 0; }
400
401#endif /* !CONFIG_X86_64 */
402
403/* common 32/64 bit code */
404
405#define ENABLE_CF8_EXT_CFG (1ULL << 46) 350#define ENABLE_CF8_EXT_CFG (1ULL << 46)
406 351
407static void enable_pci_io_ecs(void *unused) 352static void enable_pci_io_ecs(void *unused)
diff --git a/arch/x86/pci/bus_numa.c b/arch/x86/pci/bus_numa.c
index 12d54ff3654d..64a122883896 100644
--- a/arch/x86/pci/bus_numa.c
+++ b/arch/x86/pci/bus_numa.c
@@ -1,11 +1,11 @@
1#include <linux/init.h> 1#include <linux/init.h>
2#include <linux/pci.h> 2#include <linux/pci.h>
3#include <linux/range.h>
3 4
4#include "bus_numa.h" 5#include "bus_numa.h"
5 6
6int pci_root_num; 7int pci_root_num;
7struct pci_root_info pci_root_info[PCI_ROOT_NR]; 8struct pci_root_info pci_root_info[PCI_ROOT_NR];
8int found_all_numa_early;
9 9
10void x86_pci_root_bus_res_quirks(struct pci_bus *b) 10void x86_pci_root_bus_res_quirks(struct pci_bus *b)
11{ 11{
@@ -21,10 +21,6 @@ void x86_pci_root_bus_res_quirks(struct pci_bus *b)
21 if (!pci_root_num) 21 if (!pci_root_num)
22 return; 22 return;
23 23
24 /* for amd, if only one root bus, don't need to do anything */
25 if (pci_root_num < 2 && found_all_numa_early)
26 return;
27
28 for (i = 0; i < pci_root_num; i++) { 24 for (i = 0; i < pci_root_num; i++) {
29 if (pci_root_info[i].bus_min == b->number) 25 if (pci_root_info[i].bus_min == b->number)
30 break; 26 break;
@@ -52,8 +48,8 @@ void x86_pci_root_bus_res_quirks(struct pci_bus *b)
52 } 48 }
53} 49}
54 50
55void __devinit update_res(struct pci_root_info *info, size_t start, 51void __devinit update_res(struct pci_root_info *info, resource_size_t start,
56 size_t end, unsigned long flags, int merge) 52 resource_size_t end, unsigned long flags, int merge)
57{ 53{
58 int i; 54 int i;
59 struct resource *res; 55 struct resource *res;
@@ -61,25 +57,28 @@ void __devinit update_res(struct pci_root_info *info, size_t start,
61 if (start > end) 57 if (start > end)
62 return; 58 return;
63 59
60 if (start == MAX_RESOURCE)
61 return;
62
64 if (!merge) 63 if (!merge)
65 goto addit; 64 goto addit;
66 65
67 /* try to merge it with old one */ 66 /* try to merge it with old one */
68 for (i = 0; i < info->res_num; i++) { 67 for (i = 0; i < info->res_num; i++) {
69 size_t final_start, final_end; 68 resource_size_t final_start, final_end;
70 size_t common_start, common_end; 69 resource_size_t common_start, common_end;
71 70
72 res = &info->res[i]; 71 res = &info->res[i];
73 if (res->flags != flags) 72 if (res->flags != flags)
74 continue; 73 continue;
75 74
76 common_start = max((size_t)res->start, start); 75 common_start = max(res->start, start);
77 common_end = min((size_t)res->end, end); 76 common_end = min(res->end, end);
78 if (common_start > common_end + 1) 77 if (common_start > common_end + 1)
79 continue; 78 continue;
80 79
81 final_start = min((size_t)res->start, start); 80 final_start = min(res->start, start);
82 final_end = max((size_t)res->end, end); 81 final_end = max(res->end, end);
83 82
84 res->start = final_start; 83 res->start = final_start;
85 res->end = final_end; 84 res->end = final_end;
diff --git a/arch/x86/pci/bus_numa.h b/arch/x86/pci/bus_numa.h
index 731b64ee8d84..804a4b40c31a 100644
--- a/arch/x86/pci/bus_numa.h
+++ b/arch/x86/pci/bus_numa.h
@@ -1,5 +1,5 @@
1#ifdef CONFIG_X86_64 1#ifndef __BUS_NUMA_H
2 2#define __BUS_NUMA_H
3/* 3/*
4 * sub bus (transparent) will use entres from 3 to store extra from 4 * sub bus (transparent) will use entres from 3 to store extra from
5 * root, so need to make sure we have enough slot there. 5 * root, so need to make sure we have enough slot there.
@@ -19,8 +19,7 @@ struct pci_root_info {
19#define PCI_ROOT_NR 4 19#define PCI_ROOT_NR 4
20extern int pci_root_num; 20extern int pci_root_num;
21extern struct pci_root_info pci_root_info[PCI_ROOT_NR]; 21extern struct pci_root_info pci_root_info[PCI_ROOT_NR];
22extern int found_all_numa_early;
23 22
24extern void update_res(struct pci_root_info *info, size_t start, 23extern void update_res(struct pci_root_info *info, resource_size_t start,
25 size_t end, unsigned long flags, int merge); 24 resource_size_t end, unsigned long flags, int merge);
26#endif 25#endif
diff --git a/arch/x86/pci/i386.c b/arch/x86/pci/i386.c
index 5a8fbf8d4cac..dece3eb9c906 100644
--- a/arch/x86/pci/i386.c
+++ b/arch/x86/pci/i386.c
@@ -255,10 +255,6 @@ void __init pcibios_resource_survey(void)
255 */ 255 */
256fs_initcall(pcibios_assign_resources); 256fs_initcall(pcibios_assign_resources);
257 257
258void __weak x86_pci_root_bus_res_quirks(struct pci_bus *b)
259{
260}
261
262/* 258/*
263 * If we set up a device for bus mastering, we need to check the latency 259 * If we set up a device for bus mastering, we need to check the latency
264 * timer as certain crappy BIOSes forget to set it properly. 260 * timer as certain crappy BIOSes forget to set it properly.
diff --git a/drivers/pci/hotplug/ibmphp_res.c b/drivers/pci/hotplug/ibmphp_res.c
index ec73294d1fa6..e2dc289f767c 100644
--- a/drivers/pci/hotplug/ibmphp_res.c
+++ b/drivers/pci/hotplug/ibmphp_res.c
@@ -40,7 +40,7 @@ static void update_resources (struct bus_node *bus_cur, int type, int rangeno);
40static int once_over (void); 40static int once_over (void);
41static int remove_ranges (struct bus_node *, struct bus_node *); 41static int remove_ranges (struct bus_node *, struct bus_node *);
42static int update_bridge_ranges (struct bus_node **); 42static int update_bridge_ranges (struct bus_node **);
43static int add_range (int type, struct range_node *, struct bus_node *); 43static int add_bus_range (int type, struct range_node *, struct bus_node *);
44static void fix_resources (struct bus_node *); 44static void fix_resources (struct bus_node *);
45static struct bus_node *find_bus_wprev (u8, struct bus_node **, u8); 45static struct bus_node *find_bus_wprev (u8, struct bus_node **, u8);
46 46
@@ -133,7 +133,7 @@ static int __init alloc_bus_range (struct bus_node **new_bus, struct range_node
133 newrange->rangeno = 1; 133 newrange->rangeno = 1;
134 else { 134 else {
135 /* need to insert our range */ 135 /* need to insert our range */
136 add_range (flag, newrange, newbus); 136 add_bus_range (flag, newrange, newbus);
137 debug ("%d resource Primary Bus inserted on bus %x [%x - %x]\n", flag, newbus->busno, newrange->start, newrange->end); 137 debug ("%d resource Primary Bus inserted on bus %x [%x - %x]\n", flag, newbus->busno, newrange->start, newrange->end);
138 } 138 }
139 139
@@ -384,7 +384,7 @@ int __init ibmphp_rsrc_init (void)
384 * Input: type of the resource, range to add, current bus 384 * Input: type of the resource, range to add, current bus
385 * Output: 0 or -1, bus and range ptrs 385 * Output: 0 or -1, bus and range ptrs
386 ********************************************************************************/ 386 ********************************************************************************/
387static int add_range (int type, struct range_node *range, struct bus_node *bus_cur) 387static int add_bus_range (int type, struct range_node *range, struct bus_node *bus_cur)
388{ 388{
389 struct range_node *range_cur = NULL; 389 struct range_node *range_cur = NULL;
390 struct range_node *range_prev; 390 struct range_node *range_prev;
@@ -455,7 +455,7 @@ static int add_range (int type, struct range_node *range, struct bus_node *bus_c
455 455
456/******************************************************************************* 456/*******************************************************************************
457 * This routine goes through the list of resources of type 'type' and updates 457 * This routine goes through the list of resources of type 'type' and updates
458 * the range numbers that they correspond to. It was called from add_range fnc 458 * the range numbers that they correspond to. It was called from add_bus_range fnc
459 * 459 *
460 * Input: bus, type of the resource, the rangeno starting from which to update 460 * Input: bus, type of the resource, the rangeno starting from which to update
461 ******************************************************************************/ 461 ******************************************************************************/
@@ -1999,7 +1999,7 @@ static int __init update_bridge_ranges (struct bus_node **bus)
1999 1999
2000 if (bus_sec->noIORanges > 0) { 2000 if (bus_sec->noIORanges > 0) {
2001 if (!range_exists_already (range, bus_sec, IO)) { 2001 if (!range_exists_already (range, bus_sec, IO)) {
2002 add_range (IO, range, bus_sec); 2002 add_bus_range (IO, range, bus_sec);
2003 ++bus_sec->noIORanges; 2003 ++bus_sec->noIORanges;
2004 } else { 2004 } else {
2005 kfree (range); 2005 kfree (range);
@@ -2048,7 +2048,7 @@ static int __init update_bridge_ranges (struct bus_node **bus)
2048 2048
2049 if (bus_sec->noMemRanges > 0) { 2049 if (bus_sec->noMemRanges > 0) {
2050 if (!range_exists_already (range, bus_sec, MEM)) { 2050 if (!range_exists_already (range, bus_sec, MEM)) {
2051 add_range (MEM, range, bus_sec); 2051 add_bus_range (MEM, range, bus_sec);
2052 ++bus_sec->noMemRanges; 2052 ++bus_sec->noMemRanges;
2053 } else { 2053 } else {
2054 kfree (range); 2054 kfree (range);
@@ -2102,7 +2102,7 @@ static int __init update_bridge_ranges (struct bus_node **bus)
2102 2102
2103 if (bus_sec->noPFMemRanges > 0) { 2103 if (bus_sec->noPFMemRanges > 0) {
2104 if (!range_exists_already (range, bus_sec, PFMEM)) { 2104 if (!range_exists_already (range, bus_sec, PFMEM)) {
2105 add_range (PFMEM, range, bus_sec); 2105 add_bus_range (PFMEM, range, bus_sec);
2106 ++bus_sec->noPFMemRanges; 2106 ++bus_sec->noPFMemRanges;
2107 } else { 2107 } else {
2108 kfree (range); 2108 kfree (range);
diff --git a/include/linux/bootmem.h b/include/linux/bootmem.h
index b10ec49ee2dd..266ab9291232 100644
--- a/include/linux/bootmem.h
+++ b/include/linux/bootmem.h
@@ -23,6 +23,7 @@ extern unsigned long max_pfn;
23extern unsigned long saved_max_pfn; 23extern unsigned long saved_max_pfn;
24#endif 24#endif
25 25
26#ifndef CONFIG_NO_BOOTMEM
26/* 27/*
27 * node_bootmem_map is a map pointer - the bits represent all physical 28 * node_bootmem_map is a map pointer - the bits represent all physical
28 * memory pages (including holes) on the node. 29 * memory pages (including holes) on the node.
@@ -37,6 +38,7 @@ typedef struct bootmem_data {
37} bootmem_data_t; 38} bootmem_data_t;
38 39
39extern bootmem_data_t bootmem_node_data[]; 40extern bootmem_data_t bootmem_node_data[];
41#endif
40 42
41extern unsigned long bootmem_bootmap_pages(unsigned long); 43extern unsigned long bootmem_bootmap_pages(unsigned long);
42 44
@@ -46,6 +48,7 @@ extern unsigned long init_bootmem_node(pg_data_t *pgdat,
46 unsigned long endpfn); 48 unsigned long endpfn);
47extern unsigned long init_bootmem(unsigned long addr, unsigned long memend); 49extern unsigned long init_bootmem(unsigned long addr, unsigned long memend);
48 50
51unsigned long free_all_memory_core_early(int nodeid);
49extern unsigned long free_all_bootmem_node(pg_data_t *pgdat); 52extern unsigned long free_all_bootmem_node(pg_data_t *pgdat);
50extern unsigned long free_all_bootmem(void); 53extern unsigned long free_all_bootmem(void);
51 54
@@ -84,6 +87,10 @@ extern void *__alloc_bootmem_node(pg_data_t *pgdat,
84 unsigned long size, 87 unsigned long size,
85 unsigned long align, 88 unsigned long align,
86 unsigned long goal); 89 unsigned long goal);
90void *__alloc_bootmem_node_high(pg_data_t *pgdat,
91 unsigned long size,
92 unsigned long align,
93 unsigned long goal);
87extern void *__alloc_bootmem_node_nopanic(pg_data_t *pgdat, 94extern void *__alloc_bootmem_node_nopanic(pg_data_t *pgdat,
88 unsigned long size, 95 unsigned long size,
89 unsigned long align, 96 unsigned long align,
diff --git a/include/linux/early_res.h b/include/linux/early_res.h
new file mode 100644
index 000000000000..29c09f57a13c
--- /dev/null
+++ b/include/linux/early_res.h
@@ -0,0 +1,23 @@
1#ifndef _LINUX_EARLY_RES_H
2#define _LINUX_EARLY_RES_H
3#ifdef __KERNEL__
4
5extern void reserve_early(u64 start, u64 end, char *name);
6extern void reserve_early_overlap_ok(u64 start, u64 end, char *name);
7extern void free_early(u64 start, u64 end);
8void free_early_partial(u64 start, u64 end);
9extern void early_res_to_bootmem(u64 start, u64 end);
10
11void reserve_early_without_check(u64 start, u64 end, char *name);
12u64 find_early_area(u64 ei_start, u64 ei_last, u64 start, u64 end,
13 u64 size, u64 align);
14u64 find_early_area_size(u64 ei_start, u64 ei_last, u64 start,
15 u64 *sizep, u64 align);
16u64 find_fw_memmap_area(u64 start, u64 end, u64 size, u64 align);
17u64 get_max_mapped(void);
18#include <linux/range.h>
19int get_free_all_memory_range(struct range **rangep, int nodeid);
20
21#endif /* __KERNEL__ */
22
23#endif /* _LINUX_EARLY_RES_H */
diff --git a/include/linux/kernel.h b/include/linux/kernel.h
index 1221d2331a6d..7f0707463360 100644
--- a/include/linux/kernel.h
+++ b/include/linux/kernel.h
@@ -44,6 +44,16 @@ extern const char linux_proc_banner[];
44 44
45#define ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]) + __must_be_array(arr)) 45#define ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]) + __must_be_array(arr))
46 46
47/*
48 * This looks more complex than it should be. But we need to
49 * get the type for the ~ right in round_down (it needs to be
50 * as wide as the result!), and we want to evaluate the macro
51 * arguments just once each.
52 */
53#define __round_mask(x, y) ((__typeof__(x))((y)-1))
54#define round_up(x, y) ((((x)-1) | __round_mask(x, y))+1)
55#define round_down(x, y) ((x) & ~__round_mask(x, y))
56
47#define FIELD_SIZEOF(t, f) (sizeof(((t*)0)->f)) 57#define FIELD_SIZEOF(t, f) (sizeof(((t*)0)->f))
48#define DIV_ROUND_UP(n,d) (((n) + (d) - 1) / (d)) 58#define DIV_ROUND_UP(n,d) (((n) + (d) - 1) / (d))
49#define roundup(x, y) ((((x) + ((y) - 1)) / (y)) * (y)) 59#define roundup(x, y) ((((x) + ((y) - 1)) / (y)) * (y))
diff --git a/include/linux/mm.h b/include/linux/mm.h
index 2e724c877ec1..90957f14195c 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -12,6 +12,7 @@
12#include <linux/prio_tree.h> 12#include <linux/prio_tree.h>
13#include <linux/debug_locks.h> 13#include <linux/debug_locks.h>
14#include <linux/mm_types.h> 14#include <linux/mm_types.h>
15#include <linux/range.h>
15 16
16struct mempolicy; 17struct mempolicy;
17struct anon_vma; 18struct anon_vma;
@@ -1049,6 +1050,10 @@ extern void get_pfn_range_for_nid(unsigned int nid,
1049extern unsigned long find_min_pfn_with_active_regions(void); 1050extern unsigned long find_min_pfn_with_active_regions(void);
1050extern void free_bootmem_with_active_regions(int nid, 1051extern void free_bootmem_with_active_regions(int nid,
1051 unsigned long max_low_pfn); 1052 unsigned long max_low_pfn);
1053int add_from_early_node_map(struct range *range, int az,
1054 int nr_range, int nid);
1055void *__alloc_memory_core_early(int nodeid, u64 size, u64 align,
1056 u64 goal, u64 limit);
1052typedef int (*work_fn_t)(unsigned long, unsigned long, void *); 1057typedef int (*work_fn_t)(unsigned long, unsigned long, void *);
1053extern void work_with_active_regions(int nid, work_fn_t work_fn, void *data); 1058extern void work_with_active_regions(int nid, work_fn_t work_fn, void *data);
1054extern void sparse_memory_present_with_active_regions(int nid); 1059extern void sparse_memory_present_with_active_regions(int nid);
@@ -1317,12 +1322,19 @@ extern int randomize_va_space;
1317const char * arch_vma_name(struct vm_area_struct *vma); 1322const char * arch_vma_name(struct vm_area_struct *vma);
1318void print_vma_addr(char *prefix, unsigned long rip); 1323void print_vma_addr(char *prefix, unsigned long rip);
1319 1324
1325void sparse_mem_maps_populate_node(struct page **map_map,
1326 unsigned long pnum_begin,
1327 unsigned long pnum_end,
1328 unsigned long map_count,
1329 int nodeid);
1330
1320struct page *sparse_mem_map_populate(unsigned long pnum, int nid); 1331struct page *sparse_mem_map_populate(unsigned long pnum, int nid);
1321pgd_t *vmemmap_pgd_populate(unsigned long addr, int node); 1332pgd_t *vmemmap_pgd_populate(unsigned long addr, int node);
1322pud_t *vmemmap_pud_populate(pgd_t *pgd, unsigned long addr, int node); 1333pud_t *vmemmap_pud_populate(pgd_t *pgd, unsigned long addr, int node);
1323pmd_t *vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node); 1334pmd_t *vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node);
1324pte_t *vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node); 1335pte_t *vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node);
1325void *vmemmap_alloc_block(unsigned long size, int node); 1336void *vmemmap_alloc_block(unsigned long size, int node);
1337void *vmemmap_alloc_block_buf(unsigned long size, int node);
1326void vmemmap_verify(pte_t *, int, unsigned long, unsigned long); 1338void vmemmap_verify(pte_t *, int, unsigned long, unsigned long);
1327int vmemmap_populate_basepages(struct page *start_page, 1339int vmemmap_populate_basepages(struct page *start_page,
1328 unsigned long pages, int node); 1340 unsigned long pages, int node);
diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
index 41acd4bf7664..a01a103341bd 100644
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h
@@ -612,7 +612,9 @@ typedef struct pglist_data {
612 struct page_cgroup *node_page_cgroup; 612 struct page_cgroup *node_page_cgroup;
613#endif 613#endif
614#endif 614#endif
615#ifndef CONFIG_NO_BOOTMEM
615 struct bootmem_data *bdata; 616 struct bootmem_data *bdata;
617#endif
616#ifdef CONFIG_MEMORY_HOTPLUG 618#ifdef CONFIG_MEMORY_HOTPLUG
617 /* 619 /*
618 * Must be held any time you expect node_start_pfn, node_present_pages 620 * Must be held any time you expect node_start_pfn, node_present_pages
diff --git a/include/linux/range.h b/include/linux/range.h
new file mode 100644
index 000000000000..bd184a5db791
--- /dev/null
+++ b/include/linux/range.h
@@ -0,0 +1,30 @@
1#ifndef _LINUX_RANGE_H
2#define _LINUX_RANGE_H
3
4struct range {
5 u64 start;
6 u64 end;
7};
8
9int add_range(struct range *range, int az, int nr_range,
10 u64 start, u64 end);
11
12
13int add_range_with_merge(struct range *range, int az, int nr_range,
14 u64 start, u64 end);
15
16void subtract_range(struct range *range, int az, u64 start, u64 end);
17
18int clean_sort_range(struct range *range, int az);
19
20void sort_range(struct range *range, int nr_range);
21
22#define MAX_RESOURCE ((resource_size_t)~0)
23static inline resource_size_t cap_resource(u64 val)
24{
25 if (val > MAX_RESOURCE)
26 return MAX_RESOURCE;
27
28 return val;
29}
30#endif
diff --git a/kernel/Makefile b/kernel/Makefile
index 6aebdeb2aa34..7b974699f8c2 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -10,7 +10,8 @@ obj-y = sched.o fork.o exec_domain.o panic.o printk.o \
10 kthread.o wait.o kfifo.o sys_ni.o posix-cpu-timers.o mutex.o \ 10 kthread.o wait.o kfifo.o sys_ni.o posix-cpu-timers.o mutex.o \
11 hrtimer.o rwsem.o nsproxy.o srcu.o semaphore.o \ 11 hrtimer.o rwsem.o nsproxy.o srcu.o semaphore.o \
12 notifier.o ksysfs.o pm_qos_params.o sched_clock.o cred.o \ 12 notifier.o ksysfs.o pm_qos_params.o sched_clock.o cred.o \
13 async.o 13 async.o range.o
14obj-$(CONFIG_HAVE_EARLY_RES) += early_res.o
14obj-y += groups.o 15obj-y += groups.o
15 16
16ifdef CONFIG_FUNCTION_TRACER 17ifdef CONFIG_FUNCTION_TRACER
diff --git a/kernel/early_res.c b/kernel/early_res.c
new file mode 100644
index 000000000000..3cb2c661bb78
--- /dev/null
+++ b/kernel/early_res.c
@@ -0,0 +1,578 @@
1/*
2 * early_res, could be used to replace bootmem
3 */
4#include <linux/kernel.h>
5#include <linux/types.h>
6#include <linux/init.h>
7#include <linux/bootmem.h>
8#include <linux/mm.h>
9#include <linux/early_res.h>
10
11/*
12 * Early reserved memory areas.
13 */
14/*
15 * need to make sure this one is bigger enough before
16 * find_fw_memmap_area could be used
17 */
18#define MAX_EARLY_RES_X 32
19
20struct early_res {
21 u64 start, end;
22 char name[15];
23 char overlap_ok;
24};
25static struct early_res early_res_x[MAX_EARLY_RES_X] __initdata;
26
27static int max_early_res __initdata = MAX_EARLY_RES_X;
28static struct early_res *early_res __initdata = &early_res_x[0];
29static int early_res_count __initdata;
30
31static int __init find_overlapped_early(u64 start, u64 end)
32{
33 int i;
34 struct early_res *r;
35
36 for (i = 0; i < max_early_res && early_res[i].end; i++) {
37 r = &early_res[i];
38 if (end > r->start && start < r->end)
39 break;
40 }
41
42 return i;
43}
44
45/*
46 * Drop the i-th range from the early reservation map,
47 * by copying any higher ranges down one over it, and
48 * clearing what had been the last slot.
49 */
50static void __init drop_range(int i)
51{
52 int j;
53
54 for (j = i + 1; j < max_early_res && early_res[j].end; j++)
55 ;
56
57 memmove(&early_res[i], &early_res[i + 1],
58 (j - 1 - i) * sizeof(struct early_res));
59
60 early_res[j - 1].end = 0;
61 early_res_count--;
62}
63
64static void __init drop_range_partial(int i, u64 start, u64 end)
65{
66 u64 common_start, common_end;
67 u64 old_start, old_end;
68
69 old_start = early_res[i].start;
70 old_end = early_res[i].end;
71 common_start = max(old_start, start);
72 common_end = min(old_end, end);
73
74 /* no overlap ? */
75 if (common_start >= common_end)
76 return;
77
78 if (old_start < common_start) {
79 /* make head segment */
80 early_res[i].end = common_start;
81 if (old_end > common_end) {
82 char name[15];
83
84 /*
85 * Save a local copy of the name, since the
86 * early_res array could get resized inside
87 * reserve_early_without_check() ->
88 * __check_and_double_early_res(), which would
89 * make the current name pointer invalid.
90 */
91 strncpy(name, early_res[i].name,
92 sizeof(early_res[i].name) - 1);
93 /* add another for left over on tail */
94 reserve_early_without_check(common_end, old_end, name);
95 }
96 return;
97 } else {
98 if (old_end > common_end) {
99 /* reuse the entry for tail left */
100 early_res[i].start = common_end;
101 return;
102 }
103 /* all covered */
104 drop_range(i);
105 }
106}
107
108/*
109 * Split any existing ranges that:
110 * 1) are marked 'overlap_ok', and
111 * 2) overlap with the stated range [start, end)
112 * into whatever portion (if any) of the existing range is entirely
113 * below or entirely above the stated range. Drop the portion
114 * of the existing range that overlaps with the stated range,
115 * which will allow the caller of this routine to then add that
116 * stated range without conflicting with any existing range.
117 */
118static void __init drop_overlaps_that_are_ok(u64 start, u64 end)
119{
120 int i;
121 struct early_res *r;
122 u64 lower_start, lower_end;
123 u64 upper_start, upper_end;
124 char name[15];
125
126 for (i = 0; i < max_early_res && early_res[i].end; i++) {
127 r = &early_res[i];
128
129 /* Continue past non-overlapping ranges */
130 if (end <= r->start || start >= r->end)
131 continue;
132
133 /*
134 * Leave non-ok overlaps as is; let caller
135 * panic "Overlapping early reservations"
136 * when it hits this overlap.
137 */
138 if (!r->overlap_ok)
139 return;
140
141 /*
142 * We have an ok overlap. We will drop it from the early
143 * reservation map, and add back in any non-overlapping
144 * portions (lower or upper) as separate, overlap_ok,
145 * non-overlapping ranges.
146 */
147
148 /* 1. Note any non-overlapping (lower or upper) ranges. */
149 strncpy(name, r->name, sizeof(name) - 1);
150
151 lower_start = lower_end = 0;
152 upper_start = upper_end = 0;
153 if (r->start < start) {
154 lower_start = r->start;
155 lower_end = start;
156 }
157 if (r->end > end) {
158 upper_start = end;
159 upper_end = r->end;
160 }
161
162 /* 2. Drop the original ok overlapping range */
163 drop_range(i);
164
165 i--; /* resume for-loop on copied down entry */
166
167 /* 3. Add back in any non-overlapping ranges. */
168 if (lower_end)
169 reserve_early_overlap_ok(lower_start, lower_end, name);
170 if (upper_end)
171 reserve_early_overlap_ok(upper_start, upper_end, name);
172 }
173}
174
175static void __init __reserve_early(u64 start, u64 end, char *name,
176 int overlap_ok)
177{
178 int i;
179 struct early_res *r;
180
181 i = find_overlapped_early(start, end);
182 if (i >= max_early_res)
183 panic("Too many early reservations");
184 r = &early_res[i];
185 if (r->end)
186 panic("Overlapping early reservations "
187 "%llx-%llx %s to %llx-%llx %s\n",
188 start, end - 1, name ? name : "", r->start,
189 r->end - 1, r->name);
190 r->start = start;
191 r->end = end;
192 r->overlap_ok = overlap_ok;
193 if (name)
194 strncpy(r->name, name, sizeof(r->name) - 1);
195 early_res_count++;
196}
197
198/*
199 * A few early reservtations come here.
200 *
201 * The 'overlap_ok' in the name of this routine does -not- mean it
202 * is ok for these reservations to overlap an earlier reservation.
203 * Rather it means that it is ok for subsequent reservations to
204 * overlap this one.
205 *
206 * Use this entry point to reserve early ranges when you are doing
207 * so out of "Paranoia", reserving perhaps more memory than you need,
208 * just in case, and don't mind a subsequent overlapping reservation
209 * that is known to be needed.
210 *
211 * The drop_overlaps_that_are_ok() call here isn't really needed.
212 * It would be needed if we had two colliding 'overlap_ok'
213 * reservations, so that the second such would not panic on the
214 * overlap with the first. We don't have any such as of this
215 * writing, but might as well tolerate such if it happens in
216 * the future.
217 */
218void __init reserve_early_overlap_ok(u64 start, u64 end, char *name)
219{
220 drop_overlaps_that_are_ok(start, end);
221 __reserve_early(start, end, name, 1);
222}
223
224static void __init __check_and_double_early_res(u64 ex_start, u64 ex_end)
225{
226 u64 start, end, size, mem;
227 struct early_res *new;
228
229 /* do we have enough slots left ? */
230 if ((max_early_res - early_res_count) > max(max_early_res/8, 2))
231 return;
232
233 /* double it */
234 mem = -1ULL;
235 size = sizeof(struct early_res) * max_early_res * 2;
236 if (early_res == early_res_x)
237 start = 0;
238 else
239 start = early_res[0].end;
240 end = ex_start;
241 if (start + size < end)
242 mem = find_fw_memmap_area(start, end, size,
243 sizeof(struct early_res));
244 if (mem == -1ULL) {
245 start = ex_end;
246 end = get_max_mapped();
247 if (start + size < end)
248 mem = find_fw_memmap_area(start, end, size,
249 sizeof(struct early_res));
250 }
251 if (mem == -1ULL)
252 panic("can not find more space for early_res array");
253
254 new = __va(mem);
255 /* save the first one for own */
256 new[0].start = mem;
257 new[0].end = mem + size;
258 new[0].overlap_ok = 0;
259 /* copy old to new */
260 if (early_res == early_res_x) {
261 memcpy(&new[1], &early_res[0],
262 sizeof(struct early_res) * max_early_res);
263 memset(&new[max_early_res+1], 0,
264 sizeof(struct early_res) * (max_early_res - 1));
265 early_res_count++;
266 } else {
267 memcpy(&new[1], &early_res[1],
268 sizeof(struct early_res) * (max_early_res - 1));
269 memset(&new[max_early_res], 0,
270 sizeof(struct early_res) * max_early_res);
271 }
272 memset(&early_res[0], 0, sizeof(struct early_res) * max_early_res);
273 early_res = new;
274 max_early_res *= 2;
275 printk(KERN_DEBUG "early_res array is doubled to %d at [%llx - %llx]\n",
276 max_early_res, mem, mem + size - 1);
277}
278
279/*
280 * Most early reservations come here.
281 *
282 * We first have drop_overlaps_that_are_ok() drop any pre-existing
283 * 'overlap_ok' ranges, so that we can then reserve this memory
284 * range without risk of panic'ing on an overlapping overlap_ok
285 * early reservation.
286 */
287void __init reserve_early(u64 start, u64 end, char *name)
288{
289 if (start >= end)
290 return;
291
292 __check_and_double_early_res(start, end);
293
294 drop_overlaps_that_are_ok(start, end);
295 __reserve_early(start, end, name, 0);
296}
297
298void __init reserve_early_without_check(u64 start, u64 end, char *name)
299{
300 struct early_res *r;
301
302 if (start >= end)
303 return;
304
305 __check_and_double_early_res(start, end);
306
307 r = &early_res[early_res_count];
308
309 r->start = start;
310 r->end = end;
311 r->overlap_ok = 0;
312 if (name)
313 strncpy(r->name, name, sizeof(r->name) - 1);
314 early_res_count++;
315}
316
317void __init free_early(u64 start, u64 end)
318{
319 struct early_res *r;
320 int i;
321
322 i = find_overlapped_early(start, end);
323 r = &early_res[i];
324 if (i >= max_early_res || r->end != end || r->start != start)
325 panic("free_early on not reserved area: %llx-%llx!",
326 start, end - 1);
327
328 drop_range(i);
329}
330
331void __init free_early_partial(u64 start, u64 end)
332{
333 struct early_res *r;
334 int i;
335
336try_next:
337 i = find_overlapped_early(start, end);
338 if (i >= max_early_res)
339 return;
340
341 r = &early_res[i];
342 /* hole ? */
343 if (r->end >= end && r->start <= start) {
344 drop_range_partial(i, start, end);
345 return;
346 }
347
348 drop_range_partial(i, start, end);
349 goto try_next;
350}
351
352#ifdef CONFIG_NO_BOOTMEM
353static void __init subtract_early_res(struct range *range, int az)
354{
355 int i, count;
356 u64 final_start, final_end;
357 int idx = 0;
358
359 count = 0;
360 for (i = 0; i < max_early_res && early_res[i].end; i++)
361 count++;
362
363 /* need to skip first one ?*/
364 if (early_res != early_res_x)
365 idx = 1;
366
367#define DEBUG_PRINT_EARLY_RES 1
368
369#if DEBUG_PRINT_EARLY_RES
370 printk(KERN_INFO "Subtract (%d early reservations)\n", count);
371#endif
372 for (i = idx; i < count; i++) {
373 struct early_res *r = &early_res[i];
374#if DEBUG_PRINT_EARLY_RES
375 printk(KERN_INFO " #%d [%010llx - %010llx] %15s\n", i,
376 r->start, r->end, r->name);
377#endif
378 final_start = PFN_DOWN(r->start);
379 final_end = PFN_UP(r->end);
380 if (final_start >= final_end)
381 continue;
382 subtract_range(range, az, final_start, final_end);
383 }
384
385}
386
387int __init get_free_all_memory_range(struct range **rangep, int nodeid)
388{
389 int i, count;
390 u64 start = 0, end;
391 u64 size;
392 u64 mem;
393 struct range *range;
394 int nr_range;
395
396 count = 0;
397 for (i = 0; i < max_early_res && early_res[i].end; i++)
398 count++;
399
400 count *= 2;
401
402 size = sizeof(struct range) * count;
403 end = get_max_mapped();
404#ifdef MAX_DMA32_PFN
405 if (end > (MAX_DMA32_PFN << PAGE_SHIFT))
406 start = MAX_DMA32_PFN << PAGE_SHIFT;
407#endif
408 mem = find_fw_memmap_area(start, end, size, sizeof(struct range));
409 if (mem == -1ULL)
410 panic("can not find more space for range free");
411
412 range = __va(mem);
413 /* use early_node_map[] and early_res to get range array at first */
414 memset(range, 0, size);
415 nr_range = 0;
416
417 /* need to go over early_node_map to find out good range for node */
418 nr_range = add_from_early_node_map(range, count, nr_range, nodeid);
419#ifdef CONFIG_X86_32
420 subtract_range(range, count, max_low_pfn, -1ULL);
421#endif
422 subtract_early_res(range, count);
423 nr_range = clean_sort_range(range, count);
424
425 /* need to clear it ? */
426 if (nodeid == MAX_NUMNODES) {
427 memset(&early_res[0], 0,
428 sizeof(struct early_res) * max_early_res);
429 early_res = NULL;
430 max_early_res = 0;
431 }
432
433 *rangep = range;
434 return nr_range;
435}
436#else
437void __init early_res_to_bootmem(u64 start, u64 end)
438{
439 int i, count;
440 u64 final_start, final_end;
441 int idx = 0;
442
443 count = 0;
444 for (i = 0; i < max_early_res && early_res[i].end; i++)
445 count++;
446
447 /* need to skip first one ?*/
448 if (early_res != early_res_x)
449 idx = 1;
450
451 printk(KERN_INFO "(%d/%d early reservations) ==> bootmem [%010llx - %010llx]\n",
452 count - idx, max_early_res, start, end);
453 for (i = idx; i < count; i++) {
454 struct early_res *r = &early_res[i];
455 printk(KERN_INFO " #%d [%010llx - %010llx] %16s", i,
456 r->start, r->end, r->name);
457 final_start = max(start, r->start);
458 final_end = min(end, r->end);
459 if (final_start >= final_end) {
460 printk(KERN_CONT "\n");
461 continue;
462 }
463 printk(KERN_CONT " ==> [%010llx - %010llx]\n",
464 final_start, final_end);
465 reserve_bootmem_generic(final_start, final_end - final_start,
466 BOOTMEM_DEFAULT);
467 }
468 /* clear them */
469 memset(&early_res[0], 0, sizeof(struct early_res) * max_early_res);
470 early_res = NULL;
471 max_early_res = 0;
472 early_res_count = 0;
473}
474#endif
475
476/* Check for already reserved areas */
477static inline int __init bad_addr(u64 *addrp, u64 size, u64 align)
478{
479 int i;
480 u64 addr = *addrp;
481 int changed = 0;
482 struct early_res *r;
483again:
484 i = find_overlapped_early(addr, addr + size);
485 r = &early_res[i];
486 if (i < max_early_res && r->end) {
487 *addrp = addr = round_up(r->end, align);
488 changed = 1;
489 goto again;
490 }
491 return changed;
492}
493
494/* Check for already reserved areas */
495static inline int __init bad_addr_size(u64 *addrp, u64 *sizep, u64 align)
496{
497 int i;
498 u64 addr = *addrp, last;
499 u64 size = *sizep;
500 int changed = 0;
501again:
502 last = addr + size;
503 for (i = 0; i < max_early_res && early_res[i].end; i++) {
504 struct early_res *r = &early_res[i];
505 if (last > r->start && addr < r->start) {
506 size = r->start - addr;
507 changed = 1;
508 goto again;
509 }
510 if (last > r->end && addr < r->end) {
511 addr = round_up(r->end, align);
512 size = last - addr;
513 changed = 1;
514 goto again;
515 }
516 if (last <= r->end && addr >= r->start) {
517 (*sizep)++;
518 return 0;
519 }
520 }
521 if (changed) {
522 *addrp = addr;
523 *sizep = size;
524 }
525 return changed;
526}
527
528/*
529 * Find a free area with specified alignment in a specific range.
530 * only with the area.between start to end is active range from early_node_map
531 * so they are good as RAM
532 */
533u64 __init find_early_area(u64 ei_start, u64 ei_last, u64 start, u64 end,
534 u64 size, u64 align)
535{
536 u64 addr, last;
537
538 addr = round_up(ei_start, align);
539 if (addr < start)
540 addr = round_up(start, align);
541 if (addr >= ei_last)
542 goto out;
543 while (bad_addr(&addr, size, align) && addr+size <= ei_last)
544 ;
545 last = addr + size;
546 if (last > ei_last)
547 goto out;
548 if (last > end)
549 goto out;
550
551 return addr;
552
553out:
554 return -1ULL;
555}
556
557u64 __init find_early_area_size(u64 ei_start, u64 ei_last, u64 start,
558 u64 *sizep, u64 align)
559{
560 u64 addr, last;
561
562 addr = round_up(ei_start, align);
563 if (addr < start)
564 addr = round_up(start, align);
565 if (addr >= ei_last)
566 goto out;
567 *sizep = ei_last - addr;
568 while (bad_addr_size(&addr, sizep, align) && addr + *sizep <= ei_last)
569 ;
570 last = addr + *sizep;
571 if (last > ei_last)
572 goto out;
573
574 return addr;
575
576out:
577 return -1ULL;
578}
diff --git a/kernel/range.c b/kernel/range.c
new file mode 100644
index 000000000000..74e2e6114927
--- /dev/null
+++ b/kernel/range.c
@@ -0,0 +1,163 @@
1/*
2 * Range add and subtract
3 */
4#include <linux/module.h>
5#include <linux/init.h>
6#include <linux/sort.h>
7
8#include <linux/range.h>
9
10#ifndef ARRAY_SIZE
11#define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
12#endif
13
14int add_range(struct range *range, int az, int nr_range, u64 start, u64 end)
15{
16 if (start >= end)
17 return nr_range;
18
19 /* Out of slots: */
20 if (nr_range >= az)
21 return nr_range;
22
23 range[nr_range].start = start;
24 range[nr_range].end = end;
25
26 nr_range++;
27
28 return nr_range;
29}
30
31int add_range_with_merge(struct range *range, int az, int nr_range,
32 u64 start, u64 end)
33{
34 int i;
35
36 if (start >= end)
37 return nr_range;
38
39 /* Try to merge it with old one: */
40 for (i = 0; i < nr_range; i++) {
41 u64 final_start, final_end;
42 u64 common_start, common_end;
43
44 if (!range[i].end)
45 continue;
46
47 common_start = max(range[i].start, start);
48 common_end = min(range[i].end, end);
49 if (common_start > common_end)
50 continue;
51
52 final_start = min(range[i].start, start);
53 final_end = max(range[i].end, end);
54
55 range[i].start = final_start;
56 range[i].end = final_end;
57 return nr_range;
58 }
59
60 /* Need to add it: */
61 return add_range(range, az, nr_range, start, end);
62}
63
64void subtract_range(struct range *range, int az, u64 start, u64 end)
65{
66 int i, j;
67
68 if (start >= end)
69 return;
70
71 for (j = 0; j < az; j++) {
72 if (!range[j].end)
73 continue;
74
75 if (start <= range[j].start && end >= range[j].end) {
76 range[j].start = 0;
77 range[j].end = 0;
78 continue;
79 }
80
81 if (start <= range[j].start && end < range[j].end &&
82 range[j].start < end) {
83 range[j].start = end;
84 continue;
85 }
86
87
88 if (start > range[j].start && end >= range[j].end &&
89 range[j].end > start) {
90 range[j].end = start;
91 continue;
92 }
93
94 if (start > range[j].start && end < range[j].end) {
95 /* Find the new spare: */
96 for (i = 0; i < az; i++) {
97 if (range[i].end == 0)
98 break;
99 }
100 if (i < az) {
101 range[i].end = range[j].end;
102 range[i].start = end;
103 } else {
104 printk(KERN_ERR "run of slot in ranges\n");
105 }
106 range[j].end = start;
107 continue;
108 }
109 }
110}
111
112static int cmp_range(const void *x1, const void *x2)
113{
114 const struct range *r1 = x1;
115 const struct range *r2 = x2;
116 s64 start1, start2;
117
118 start1 = r1->start;
119 start2 = r2->start;
120
121 return start1 - start2;
122}
123
124int clean_sort_range(struct range *range, int az)
125{
126 int i, j, k = az - 1, nr_range = 0;
127
128 for (i = 0; i < k; i++) {
129 if (range[i].end)
130 continue;
131 for (j = k; j > i; j--) {
132 if (range[j].end) {
133 k = j;
134 break;
135 }
136 }
137 if (j == i)
138 break;
139 range[i].start = range[k].start;
140 range[i].end = range[k].end;
141 range[k].start = 0;
142 range[k].end = 0;
143 k--;
144 }
145 /* count it */
146 for (i = 0; i < az; i++) {
147 if (!range[i].end) {
148 nr_range = i;
149 break;
150 }
151 }
152
153 /* sort them */
154 sort(range, nr_range, sizeof(struct range), cmp_range, NULL);
155
156 return nr_range;
157}
158
159void sort_range(struct range *range, int nr_range)
160{
161 /* sort them */
162 sort(range, nr_range, sizeof(struct range), cmp_range, NULL);
163}
diff --git a/mm/Kconfig b/mm/Kconfig
index d34c2b971032..9c61158308dc 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -115,6 +115,10 @@ config SPARSEMEM_EXTREME
115config SPARSEMEM_VMEMMAP_ENABLE 115config SPARSEMEM_VMEMMAP_ENABLE
116 bool 116 bool
117 117
118config SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
119 def_bool y
120 depends on SPARSEMEM && X86_64
121
118config SPARSEMEM_VMEMMAP 122config SPARSEMEM_VMEMMAP
119 bool "Sparse Memory virtual memmap" 123 bool "Sparse Memory virtual memmap"
120 depends on SPARSEMEM && SPARSEMEM_VMEMMAP_ENABLE 124 depends on SPARSEMEM && SPARSEMEM_VMEMMAP_ENABLE
diff --git a/mm/bootmem.c b/mm/bootmem.c
index 7d1486875e1c..d7c791ef0036 100644
--- a/mm/bootmem.c
+++ b/mm/bootmem.c
@@ -13,6 +13,7 @@
13#include <linux/bootmem.h> 13#include <linux/bootmem.h>
14#include <linux/module.h> 14#include <linux/module.h>
15#include <linux/kmemleak.h> 15#include <linux/kmemleak.h>
16#include <linux/range.h>
16 17
17#include <asm/bug.h> 18#include <asm/bug.h>
18#include <asm/io.h> 19#include <asm/io.h>
@@ -32,6 +33,7 @@ unsigned long max_pfn;
32unsigned long saved_max_pfn; 33unsigned long saved_max_pfn;
33#endif 34#endif
34 35
36#ifndef CONFIG_NO_BOOTMEM
35bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata; 37bootmem_data_t bootmem_node_data[MAX_NUMNODES] __initdata;
36 38
37static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list); 39static struct list_head bdata_list __initdata = LIST_HEAD_INIT(bdata_list);
@@ -142,7 +144,7 @@ unsigned long __init init_bootmem(unsigned long start, unsigned long pages)
142 min_low_pfn = start; 144 min_low_pfn = start;
143 return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages); 145 return init_bootmem_core(NODE_DATA(0)->bdata, start, 0, pages);
144} 146}
145 147#endif
146/* 148/*
147 * free_bootmem_late - free bootmem pages directly to page allocator 149 * free_bootmem_late - free bootmem pages directly to page allocator
148 * @addr: starting address of the range 150 * @addr: starting address of the range
@@ -167,6 +169,60 @@ void __init free_bootmem_late(unsigned long addr, unsigned long size)
167 } 169 }
168} 170}
169 171
172#ifdef CONFIG_NO_BOOTMEM
173static void __init __free_pages_memory(unsigned long start, unsigned long end)
174{
175 int i;
176 unsigned long start_aligned, end_aligned;
177 int order = ilog2(BITS_PER_LONG);
178
179 start_aligned = (start + (BITS_PER_LONG - 1)) & ~(BITS_PER_LONG - 1);
180 end_aligned = end & ~(BITS_PER_LONG - 1);
181
182 if (end_aligned <= start_aligned) {
183#if 1
184 printk(KERN_DEBUG " %lx - %lx\n", start, end);
185#endif
186 for (i = start; i < end; i++)
187 __free_pages_bootmem(pfn_to_page(i), 0);
188
189 return;
190 }
191
192#if 1
193 printk(KERN_DEBUG " %lx %lx - %lx %lx\n",
194 start, start_aligned, end_aligned, end);
195#endif
196 for (i = start; i < start_aligned; i++)
197 __free_pages_bootmem(pfn_to_page(i), 0);
198
199 for (i = start_aligned; i < end_aligned; i += BITS_PER_LONG)
200 __free_pages_bootmem(pfn_to_page(i), order);
201
202 for (i = end_aligned; i < end; i++)
203 __free_pages_bootmem(pfn_to_page(i), 0);
204}
205
206unsigned long __init free_all_memory_core_early(int nodeid)
207{
208 int i;
209 u64 start, end;
210 unsigned long count = 0;
211 struct range *range = NULL;
212 int nr_range;
213
214 nr_range = get_free_all_memory_range(&range, nodeid);
215
216 for (i = 0; i < nr_range; i++) {
217 start = range[i].start;
218 end = range[i].end;
219 count += end - start;
220 __free_pages_memory(start, end);
221 }
222
223 return count;
224}
225#else
170static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata) 226static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
171{ 227{
172 int aligned; 228 int aligned;
@@ -227,6 +283,7 @@ static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
227 283
228 return count; 284 return count;
229} 285}
286#endif
230 287
231/** 288/**
232 * free_all_bootmem_node - release a node's free pages to the buddy allocator 289 * free_all_bootmem_node - release a node's free pages to the buddy allocator
@@ -237,7 +294,12 @@ static unsigned long __init free_all_bootmem_core(bootmem_data_t *bdata)
237unsigned long __init free_all_bootmem_node(pg_data_t *pgdat) 294unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
238{ 295{
239 register_page_bootmem_info_node(pgdat); 296 register_page_bootmem_info_node(pgdat);
297#ifdef CONFIG_NO_BOOTMEM
298 /* free_all_memory_core_early(MAX_NUMNODES) will be called later */
299 return 0;
300#else
240 return free_all_bootmem_core(pgdat->bdata); 301 return free_all_bootmem_core(pgdat->bdata);
302#endif
241} 303}
242 304
243/** 305/**
@@ -247,9 +309,14 @@ unsigned long __init free_all_bootmem_node(pg_data_t *pgdat)
247 */ 309 */
248unsigned long __init free_all_bootmem(void) 310unsigned long __init free_all_bootmem(void)
249{ 311{
312#ifdef CONFIG_NO_BOOTMEM
313 return free_all_memory_core_early(NODE_DATA(0)->node_id);
314#else
250 return free_all_bootmem_core(NODE_DATA(0)->bdata); 315 return free_all_bootmem_core(NODE_DATA(0)->bdata);
316#endif
251} 317}
252 318
319#ifndef CONFIG_NO_BOOTMEM
253static void __init __free(bootmem_data_t *bdata, 320static void __init __free(bootmem_data_t *bdata,
254 unsigned long sidx, unsigned long eidx) 321 unsigned long sidx, unsigned long eidx)
255{ 322{
@@ -344,6 +411,7 @@ static int __init mark_bootmem(unsigned long start, unsigned long end,
344 } 411 }
345 BUG(); 412 BUG();
346} 413}
414#endif
347 415
348/** 416/**
349 * free_bootmem_node - mark a page range as usable 417 * free_bootmem_node - mark a page range as usable
@@ -358,6 +426,12 @@ static int __init mark_bootmem(unsigned long start, unsigned long end,
358void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, 426void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
359 unsigned long size) 427 unsigned long size)
360{ 428{
429#ifdef CONFIG_NO_BOOTMEM
430 free_early(physaddr, physaddr + size);
431#if 0
432 printk(KERN_DEBUG "free %lx %lx\n", physaddr, size);
433#endif
434#else
361 unsigned long start, end; 435 unsigned long start, end;
362 436
363 kmemleak_free_part(__va(physaddr), size); 437 kmemleak_free_part(__va(physaddr), size);
@@ -366,6 +440,7 @@ void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
366 end = PFN_DOWN(physaddr + size); 440 end = PFN_DOWN(physaddr + size);
367 441
368 mark_bootmem_node(pgdat->bdata, start, end, 0, 0); 442 mark_bootmem_node(pgdat->bdata, start, end, 0, 0);
443#endif
369} 444}
370 445
371/** 446/**
@@ -379,6 +454,12 @@ void __init free_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
379 */ 454 */
380void __init free_bootmem(unsigned long addr, unsigned long size) 455void __init free_bootmem(unsigned long addr, unsigned long size)
381{ 456{
457#ifdef CONFIG_NO_BOOTMEM
458 free_early(addr, addr + size);
459#if 0
460 printk(KERN_DEBUG "free %lx %lx\n", addr, size);
461#endif
462#else
382 unsigned long start, end; 463 unsigned long start, end;
383 464
384 kmemleak_free_part(__va(addr), size); 465 kmemleak_free_part(__va(addr), size);
@@ -387,6 +468,7 @@ void __init free_bootmem(unsigned long addr, unsigned long size)
387 end = PFN_DOWN(addr + size); 468 end = PFN_DOWN(addr + size);
388 469
389 mark_bootmem(start, end, 0, 0); 470 mark_bootmem(start, end, 0, 0);
471#endif
390} 472}
391 473
392/** 474/**
@@ -403,12 +485,17 @@ void __init free_bootmem(unsigned long addr, unsigned long size)
403int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr, 485int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
404 unsigned long size, int flags) 486 unsigned long size, int flags)
405{ 487{
488#ifdef CONFIG_NO_BOOTMEM
489 panic("no bootmem");
490 return 0;
491#else
406 unsigned long start, end; 492 unsigned long start, end;
407 493
408 start = PFN_DOWN(physaddr); 494 start = PFN_DOWN(physaddr);
409 end = PFN_UP(physaddr + size); 495 end = PFN_UP(physaddr + size);
410 496
411 return mark_bootmem_node(pgdat->bdata, start, end, 1, flags); 497 return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
498#endif
412} 499}
413 500
414/** 501/**
@@ -424,14 +511,20 @@ int __init reserve_bootmem_node(pg_data_t *pgdat, unsigned long physaddr,
424int __init reserve_bootmem(unsigned long addr, unsigned long size, 511int __init reserve_bootmem(unsigned long addr, unsigned long size,
425 int flags) 512 int flags)
426{ 513{
514#ifdef CONFIG_NO_BOOTMEM
515 panic("no bootmem");
516 return 0;
517#else
427 unsigned long start, end; 518 unsigned long start, end;
428 519
429 start = PFN_DOWN(addr); 520 start = PFN_DOWN(addr);
430 end = PFN_UP(addr + size); 521 end = PFN_UP(addr + size);
431 522
432 return mark_bootmem(start, end, 1, flags); 523 return mark_bootmem(start, end, 1, flags);
524#endif
433} 525}
434 526
527#ifndef CONFIG_NO_BOOTMEM
435static unsigned long __init align_idx(struct bootmem_data *bdata, 528static unsigned long __init align_idx(struct bootmem_data *bdata,
436 unsigned long idx, unsigned long step) 529 unsigned long idx, unsigned long step)
437{ 530{
@@ -582,12 +675,33 @@ static void * __init alloc_arch_preferred_bootmem(bootmem_data_t *bdata,
582#endif 675#endif
583 return NULL; 676 return NULL;
584} 677}
678#endif
585 679
586static void * __init ___alloc_bootmem_nopanic(unsigned long size, 680static void * __init ___alloc_bootmem_nopanic(unsigned long size,
587 unsigned long align, 681 unsigned long align,
588 unsigned long goal, 682 unsigned long goal,
589 unsigned long limit) 683 unsigned long limit)
590{ 684{
685#ifdef CONFIG_NO_BOOTMEM
686 void *ptr;
687
688 if (WARN_ON_ONCE(slab_is_available()))
689 return kzalloc(size, GFP_NOWAIT);
690
691restart:
692
693 ptr = __alloc_memory_core_early(MAX_NUMNODES, size, align, goal, limit);
694
695 if (ptr)
696 return ptr;
697
698 if (goal != 0) {
699 goal = 0;
700 goto restart;
701 }
702
703 return NULL;
704#else
591 bootmem_data_t *bdata; 705 bootmem_data_t *bdata;
592 void *region; 706 void *region;
593 707
@@ -613,6 +727,7 @@ restart:
613 } 727 }
614 728
615 return NULL; 729 return NULL;
730#endif
616} 731}
617 732
618/** 733/**
@@ -631,7 +746,13 @@ restart:
631void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align, 746void * __init __alloc_bootmem_nopanic(unsigned long size, unsigned long align,
632 unsigned long goal) 747 unsigned long goal)
633{ 748{
634 return ___alloc_bootmem_nopanic(size, align, goal, 0); 749 unsigned long limit = 0;
750
751#ifdef CONFIG_NO_BOOTMEM
752 limit = -1UL;
753#endif
754
755 return ___alloc_bootmem_nopanic(size, align, goal, limit);
635} 756}
636 757
637static void * __init ___alloc_bootmem(unsigned long size, unsigned long align, 758static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
@@ -665,9 +786,16 @@ static void * __init ___alloc_bootmem(unsigned long size, unsigned long align,
665void * __init __alloc_bootmem(unsigned long size, unsigned long align, 786void * __init __alloc_bootmem(unsigned long size, unsigned long align,
666 unsigned long goal) 787 unsigned long goal)
667{ 788{
668 return ___alloc_bootmem(size, align, goal, 0); 789 unsigned long limit = 0;
790
791#ifdef CONFIG_NO_BOOTMEM
792 limit = -1UL;
793#endif
794
795 return ___alloc_bootmem(size, align, goal, limit);
669} 796}
670 797
798#ifndef CONFIG_NO_BOOTMEM
671static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata, 799static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata,
672 unsigned long size, unsigned long align, 800 unsigned long size, unsigned long align,
673 unsigned long goal, unsigned long limit) 801 unsigned long goal, unsigned long limit)
@@ -684,6 +812,7 @@ static void * __init ___alloc_bootmem_node(bootmem_data_t *bdata,
684 812
685 return ___alloc_bootmem(size, align, goal, limit); 813 return ___alloc_bootmem(size, align, goal, limit);
686} 814}
815#endif
687 816
688/** 817/**
689 * __alloc_bootmem_node - allocate boot memory from a specific node 818 * __alloc_bootmem_node - allocate boot memory from a specific node
@@ -706,7 +835,46 @@ void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
706 if (WARN_ON_ONCE(slab_is_available())) 835 if (WARN_ON_ONCE(slab_is_available()))
707 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); 836 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
708 837
838#ifdef CONFIG_NO_BOOTMEM
839 return __alloc_memory_core_early(pgdat->node_id, size, align,
840 goal, -1ULL);
841#else
709 return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0); 842 return ___alloc_bootmem_node(pgdat->bdata, size, align, goal, 0);
843#endif
844}
845
846void * __init __alloc_bootmem_node_high(pg_data_t *pgdat, unsigned long size,
847 unsigned long align, unsigned long goal)
848{
849#ifdef MAX_DMA32_PFN
850 unsigned long end_pfn;
851
852 if (WARN_ON_ONCE(slab_is_available()))
853 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
854
855 /* update goal according ...MAX_DMA32_PFN */
856 end_pfn = pgdat->node_start_pfn + pgdat->node_spanned_pages;
857
858 if (end_pfn > MAX_DMA32_PFN + (128 >> (20 - PAGE_SHIFT)) &&
859 (goal >> PAGE_SHIFT) < MAX_DMA32_PFN) {
860 void *ptr;
861 unsigned long new_goal;
862
863 new_goal = MAX_DMA32_PFN << PAGE_SHIFT;
864#ifdef CONFIG_NO_BOOTMEM
865 ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
866 new_goal, -1ULL);
867#else
868 ptr = alloc_bootmem_core(pgdat->bdata, size, align,
869 new_goal, 0);
870#endif
871 if (ptr)
872 return ptr;
873 }
874#endif
875
876 return __alloc_bootmem_node(pgdat, size, align, goal);
877
710} 878}
711 879
712#ifdef CONFIG_SPARSEMEM 880#ifdef CONFIG_SPARSEMEM
@@ -720,6 +888,16 @@ void * __init __alloc_bootmem_node(pg_data_t *pgdat, unsigned long size,
720void * __init alloc_bootmem_section(unsigned long size, 888void * __init alloc_bootmem_section(unsigned long size,
721 unsigned long section_nr) 889 unsigned long section_nr)
722{ 890{
891#ifdef CONFIG_NO_BOOTMEM
892 unsigned long pfn, goal, limit;
893
894 pfn = section_nr_to_pfn(section_nr);
895 goal = pfn << PAGE_SHIFT;
896 limit = section_nr_to_pfn(section_nr + 1) << PAGE_SHIFT;
897
898 return __alloc_memory_core_early(early_pfn_to_nid(pfn), size,
899 SMP_CACHE_BYTES, goal, limit);
900#else
723 bootmem_data_t *bdata; 901 bootmem_data_t *bdata;
724 unsigned long pfn, goal, limit; 902 unsigned long pfn, goal, limit;
725 903
@@ -729,6 +907,7 @@ void * __init alloc_bootmem_section(unsigned long size,
729 bdata = &bootmem_node_data[early_pfn_to_nid(pfn)]; 907 bdata = &bootmem_node_data[early_pfn_to_nid(pfn)];
730 908
731 return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit); 909 return alloc_bootmem_core(bdata, size, SMP_CACHE_BYTES, goal, limit);
910#endif
732} 911}
733#endif 912#endif
734 913
@@ -740,11 +919,16 @@ void * __init __alloc_bootmem_node_nopanic(pg_data_t *pgdat, unsigned long size,
740 if (WARN_ON_ONCE(slab_is_available())) 919 if (WARN_ON_ONCE(slab_is_available()))
741 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); 920 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
742 921
922#ifdef CONFIG_NO_BOOTMEM
923 ptr = __alloc_memory_core_early(pgdat->node_id, size, align,
924 goal, -1ULL);
925#else
743 ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0); 926 ptr = alloc_arch_preferred_bootmem(pgdat->bdata, size, align, goal, 0);
744 if (ptr) 927 if (ptr)
745 return ptr; 928 return ptr;
746 929
747 ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0); 930 ptr = alloc_bootmem_core(pgdat->bdata, size, align, goal, 0);
931#endif
748 if (ptr) 932 if (ptr)
749 return ptr; 933 return ptr;
750 934
@@ -795,6 +979,11 @@ void * __init __alloc_bootmem_low_node(pg_data_t *pgdat, unsigned long size,
795 if (WARN_ON_ONCE(slab_is_available())) 979 if (WARN_ON_ONCE(slab_is_available()))
796 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id); 980 return kzalloc_node(size, GFP_NOWAIT, pgdat->node_id);
797 981
982#ifdef CONFIG_NO_BOOTMEM
983 return __alloc_memory_core_early(pgdat->node_id, size, align,
984 goal, ARCH_LOW_ADDRESS_LIMIT);
985#else
798 return ___alloc_bootmem_node(pgdat->bdata, size, align, 986 return ___alloc_bootmem_node(pgdat->bdata, size, align,
799 goal, ARCH_LOW_ADDRESS_LIMIT); 987 goal, ARCH_LOW_ADDRESS_LIMIT);
988#endif
800} 989}
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 9a7aaae07ab4..a6b17aa4740b 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -3374,6 +3374,61 @@ void __init free_bootmem_with_active_regions(int nid,
3374 } 3374 }
3375} 3375}
3376 3376
3377int __init add_from_early_node_map(struct range *range, int az,
3378 int nr_range, int nid)
3379{
3380 int i;
3381 u64 start, end;
3382
3383 /* need to go over early_node_map to find out good range for node */
3384 for_each_active_range_index_in_nid(i, nid) {
3385 start = early_node_map[i].start_pfn;
3386 end = early_node_map[i].end_pfn;
3387 nr_range = add_range(range, az, nr_range, start, end);
3388 }
3389 return nr_range;
3390}
3391
3392#ifdef CONFIG_NO_BOOTMEM
3393void * __init __alloc_memory_core_early(int nid, u64 size, u64 align,
3394 u64 goal, u64 limit)
3395{
3396 int i;
3397 void *ptr;
3398
3399 /* need to go over early_node_map to find out good range for node */
3400 for_each_active_range_index_in_nid(i, nid) {
3401 u64 addr;
3402 u64 ei_start, ei_last;
3403
3404 ei_last = early_node_map[i].end_pfn;
3405 ei_last <<= PAGE_SHIFT;
3406 ei_start = early_node_map[i].start_pfn;
3407 ei_start <<= PAGE_SHIFT;
3408 addr = find_early_area(ei_start, ei_last,
3409 goal, limit, size, align);
3410
3411 if (addr == -1ULL)
3412 continue;
3413
3414#if 0
3415 printk(KERN_DEBUG "alloc (nid=%d %llx - %llx) (%llx - %llx) %llx %llx => %llx\n",
3416 nid,
3417 ei_start, ei_last, goal, limit, size,
3418 align, addr);
3419#endif
3420
3421 ptr = phys_to_virt(addr);
3422 memset(ptr, 0, size);
3423 reserve_early_without_check(addr, addr + size, "BOOTMEM");
3424 return ptr;
3425 }
3426
3427 return NULL;
3428}
3429#endif
3430
3431
3377void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data) 3432void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3378{ 3433{
3379 int i; 3434 int i;
@@ -4406,7 +4461,11 @@ void __init set_dma_reserve(unsigned long new_dma_reserve)
4406} 4461}
4407 4462
4408#ifndef CONFIG_NEED_MULTIPLE_NODES 4463#ifndef CONFIG_NEED_MULTIPLE_NODES
4409struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] }; 4464struct pglist_data __refdata contig_page_data = {
4465#ifndef CONFIG_NO_BOOTMEM
4466 .bdata = &bootmem_node_data[0]
4467#endif
4468 };
4410EXPORT_SYMBOL(contig_page_data); 4469EXPORT_SYMBOL(contig_page_data);
4411#endif 4470#endif
4412 4471
diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c
index d9714bdcb4a3..392b9bb5bc01 100644
--- a/mm/sparse-vmemmap.c
+++ b/mm/sparse-vmemmap.c
@@ -40,9 +40,11 @@ static void * __init_refok __earlyonly_bootmem_alloc(int node,
40 unsigned long align, 40 unsigned long align,
41 unsigned long goal) 41 unsigned long goal)
42{ 42{
43 return __alloc_bootmem_node(NODE_DATA(node), size, align, goal); 43 return __alloc_bootmem_node_high(NODE_DATA(node), size, align, goal);
44} 44}
45 45
46static void *vmemmap_buf;
47static void *vmemmap_buf_end;
46 48
47void * __meminit vmemmap_alloc_block(unsigned long size, int node) 49void * __meminit vmemmap_alloc_block(unsigned long size, int node)
48{ 50{
@@ -64,6 +66,24 @@ void * __meminit vmemmap_alloc_block(unsigned long size, int node)
64 __pa(MAX_DMA_ADDRESS)); 66 __pa(MAX_DMA_ADDRESS));
65} 67}
66 68
69/* need to make sure size is all the same during early stage */
70void * __meminit vmemmap_alloc_block_buf(unsigned long size, int node)
71{
72 void *ptr;
73
74 if (!vmemmap_buf)
75 return vmemmap_alloc_block(size, node);
76
77 /* take the from buf */
78 ptr = (void *)ALIGN((unsigned long)vmemmap_buf, size);
79 if (ptr + size > vmemmap_buf_end)
80 return vmemmap_alloc_block(size, node);
81
82 vmemmap_buf = ptr + size;
83
84 return ptr;
85}
86
67void __meminit vmemmap_verify(pte_t *pte, int node, 87void __meminit vmemmap_verify(pte_t *pte, int node,
68 unsigned long start, unsigned long end) 88 unsigned long start, unsigned long end)
69{ 89{
@@ -80,7 +100,7 @@ pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
80 pte_t *pte = pte_offset_kernel(pmd, addr); 100 pte_t *pte = pte_offset_kernel(pmd, addr);
81 if (pte_none(*pte)) { 101 if (pte_none(*pte)) {
82 pte_t entry; 102 pte_t entry;
83 void *p = vmemmap_alloc_block(PAGE_SIZE, node); 103 void *p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
84 if (!p) 104 if (!p)
85 return NULL; 105 return NULL;
86 entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL); 106 entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
@@ -163,3 +183,55 @@ struct page * __meminit sparse_mem_map_populate(unsigned long pnum, int nid)
163 183
164 return map; 184 return map;
165} 185}
186
187void __init sparse_mem_maps_populate_node(struct page **map_map,
188 unsigned long pnum_begin,
189 unsigned long pnum_end,
190 unsigned long map_count, int nodeid)
191{
192 unsigned long pnum;
193 unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
194 void *vmemmap_buf_start;
195
196 size = ALIGN(size, PMD_SIZE);
197 vmemmap_buf_start = __earlyonly_bootmem_alloc(nodeid, size * map_count,
198 PMD_SIZE, __pa(MAX_DMA_ADDRESS));
199
200 if (vmemmap_buf_start) {
201 vmemmap_buf = vmemmap_buf_start;
202 vmemmap_buf_end = vmemmap_buf_start + size * map_count;
203 }
204
205 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
206 struct mem_section *ms;
207
208 if (!present_section_nr(pnum))
209 continue;
210
211 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
212 if (map_map[pnum])
213 continue;
214 ms = __nr_to_section(pnum);
215 printk(KERN_ERR "%s: sparsemem memory map backing failed "
216 "some memory will not be available.\n", __func__);
217 ms->section_mem_map = 0;
218 }
219
220 if (vmemmap_buf_start) {
221 /* need to free left buf */
222#ifdef CONFIG_NO_BOOTMEM
223 free_early(__pa(vmemmap_buf_start), __pa(vmemmap_buf_end));
224 if (vmemmap_buf_start < vmemmap_buf) {
225 char name[15];
226
227 snprintf(name, sizeof(name), "MEMMAP %d", nodeid);
228 reserve_early_without_check(__pa(vmemmap_buf_start),
229 __pa(vmemmap_buf), name);
230 }
231#else
232 free_bootmem(__pa(vmemmap_buf), vmemmap_buf_end - vmemmap_buf);
233#endif
234 vmemmap_buf = NULL;
235 vmemmap_buf_end = NULL;
236 }
237}
diff --git a/mm/sparse.c b/mm/sparse.c
index 6ce4aab69e99..22896d589133 100644
--- a/mm/sparse.c
+++ b/mm/sparse.c
@@ -271,7 +271,8 @@ static unsigned long *__kmalloc_section_usemap(void)
271 271
272#ifdef CONFIG_MEMORY_HOTREMOVE 272#ifdef CONFIG_MEMORY_HOTREMOVE
273static unsigned long * __init 273static unsigned long * __init
274sparse_early_usemap_alloc_pgdat_section(struct pglist_data *pgdat) 274sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
275 unsigned long count)
275{ 276{
276 unsigned long section_nr; 277 unsigned long section_nr;
277 278
@@ -286,7 +287,7 @@ sparse_early_usemap_alloc_pgdat_section(struct pglist_data *pgdat)
286 * this problem. 287 * this problem.
287 */ 288 */
288 section_nr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT); 289 section_nr = pfn_to_section_nr(__pa(pgdat) >> PAGE_SHIFT);
289 return alloc_bootmem_section(usemap_size(), section_nr); 290 return alloc_bootmem_section(usemap_size() * count, section_nr);
290} 291}
291 292
292static void __init check_usemap_section_nr(int nid, unsigned long *usemap) 293static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
@@ -329,7 +330,8 @@ static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
329} 330}
330#else 331#else
331static unsigned long * __init 332static unsigned long * __init
332sparse_early_usemap_alloc_pgdat_section(struct pglist_data *pgdat) 333sparse_early_usemaps_alloc_pgdat_section(struct pglist_data *pgdat,
334 unsigned long count)
333{ 335{
334 return NULL; 336 return NULL;
335} 337}
@@ -339,27 +341,40 @@ static void __init check_usemap_section_nr(int nid, unsigned long *usemap)
339} 341}
340#endif /* CONFIG_MEMORY_HOTREMOVE */ 342#endif /* CONFIG_MEMORY_HOTREMOVE */
341 343
342static unsigned long *__init sparse_early_usemap_alloc(unsigned long pnum) 344static void __init sparse_early_usemaps_alloc_node(unsigned long**usemap_map,
345 unsigned long pnum_begin,
346 unsigned long pnum_end,
347 unsigned long usemap_count, int nodeid)
343{ 348{
344 unsigned long *usemap; 349 void *usemap;
345 struct mem_section *ms = __nr_to_section(pnum); 350 unsigned long pnum;
346 int nid = sparse_early_nid(ms); 351 int size = usemap_size();
347
348 usemap = sparse_early_usemap_alloc_pgdat_section(NODE_DATA(nid));
349 if (usemap)
350 return usemap;
351 352
352 usemap = alloc_bootmem_node(NODE_DATA(nid), usemap_size()); 353 usemap = sparse_early_usemaps_alloc_pgdat_section(NODE_DATA(nodeid),
354 usemap_count);
353 if (usemap) { 355 if (usemap) {
354 check_usemap_section_nr(nid, usemap); 356 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
355 return usemap; 357 if (!present_section_nr(pnum))
358 continue;
359 usemap_map[pnum] = usemap;
360 usemap += size;
361 }
362 return;
356 } 363 }
357 364
358 /* Stupid: suppress gcc warning for SPARSEMEM && !NUMA */ 365 usemap = alloc_bootmem_node(NODE_DATA(nodeid), size * usemap_count);
359 nid = 0; 366 if (usemap) {
367 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
368 if (!present_section_nr(pnum))
369 continue;
370 usemap_map[pnum] = usemap;
371 usemap += size;
372 check_usemap_section_nr(nodeid, usemap_map[pnum]);
373 }
374 return;
375 }
360 376
361 printk(KERN_WARNING "%s: allocation failed\n", __func__); 377 printk(KERN_WARNING "%s: allocation failed\n", __func__);
362 return NULL;
363} 378}
364 379
365#ifndef CONFIG_SPARSEMEM_VMEMMAP 380#ifndef CONFIG_SPARSEMEM_VMEMMAP
@@ -375,8 +390,65 @@ struct page __init *sparse_mem_map_populate(unsigned long pnum, int nid)
375 PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION)); 390 PAGE_ALIGN(sizeof(struct page) * PAGES_PER_SECTION));
376 return map; 391 return map;
377} 392}
393void __init sparse_mem_maps_populate_node(struct page **map_map,
394 unsigned long pnum_begin,
395 unsigned long pnum_end,
396 unsigned long map_count, int nodeid)
397{
398 void *map;
399 unsigned long pnum;
400 unsigned long size = sizeof(struct page) * PAGES_PER_SECTION;
401
402 map = alloc_remap(nodeid, size * map_count);
403 if (map) {
404 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
405 if (!present_section_nr(pnum))
406 continue;
407 map_map[pnum] = map;
408 map += size;
409 }
410 return;
411 }
412
413 size = PAGE_ALIGN(size);
414 map = alloc_bootmem_pages_node(NODE_DATA(nodeid), size * map_count);
415 if (map) {
416 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
417 if (!present_section_nr(pnum))
418 continue;
419 map_map[pnum] = map;
420 map += size;
421 }
422 return;
423 }
424
425 /* fallback */
426 for (pnum = pnum_begin; pnum < pnum_end; pnum++) {
427 struct mem_section *ms;
428
429 if (!present_section_nr(pnum))
430 continue;
431 map_map[pnum] = sparse_mem_map_populate(pnum, nodeid);
432 if (map_map[pnum])
433 continue;
434 ms = __nr_to_section(pnum);
435 printk(KERN_ERR "%s: sparsemem memory map backing failed "
436 "some memory will not be available.\n", __func__);
437 ms->section_mem_map = 0;
438 }
439}
378#endif /* !CONFIG_SPARSEMEM_VMEMMAP */ 440#endif /* !CONFIG_SPARSEMEM_VMEMMAP */
379 441
442#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
443static void __init sparse_early_mem_maps_alloc_node(struct page **map_map,
444 unsigned long pnum_begin,
445 unsigned long pnum_end,
446 unsigned long map_count, int nodeid)
447{
448 sparse_mem_maps_populate_node(map_map, pnum_begin, pnum_end,
449 map_count, nodeid);
450}
451#else
380static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum) 452static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum)
381{ 453{
382 struct page *map; 454 struct page *map;
@@ -392,10 +464,12 @@ static struct page __init *sparse_early_mem_map_alloc(unsigned long pnum)
392 ms->section_mem_map = 0; 464 ms->section_mem_map = 0;
393 return NULL; 465 return NULL;
394} 466}
467#endif
395 468
396void __attribute__((weak)) __meminit vmemmap_populate_print_last(void) 469void __attribute__((weak)) __meminit vmemmap_populate_print_last(void)
397{ 470{
398} 471}
472
399/* 473/*
400 * Allocate the accumulated non-linear sections, allocate a mem_map 474 * Allocate the accumulated non-linear sections, allocate a mem_map
401 * for each and record the physical to section mapping. 475 * for each and record the physical to section mapping.
@@ -407,6 +481,14 @@ void __init sparse_init(void)
407 unsigned long *usemap; 481 unsigned long *usemap;
408 unsigned long **usemap_map; 482 unsigned long **usemap_map;
409 int size; 483 int size;
484 int nodeid_begin = 0;
485 unsigned long pnum_begin = 0;
486 unsigned long usemap_count;
487#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
488 unsigned long map_count;
489 int size2;
490 struct page **map_map;
491#endif
410 492
411 /* 493 /*
412 * map is using big page (aka 2M in x86 64 bit) 494 * map is using big page (aka 2M in x86 64 bit)
@@ -425,10 +507,81 @@ void __init sparse_init(void)
425 panic("can not allocate usemap_map\n"); 507 panic("can not allocate usemap_map\n");
426 508
427 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) { 509 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
510 struct mem_section *ms;
511
428 if (!present_section_nr(pnum)) 512 if (!present_section_nr(pnum))
429 continue; 513 continue;
430 usemap_map[pnum] = sparse_early_usemap_alloc(pnum); 514 ms = __nr_to_section(pnum);
515 nodeid_begin = sparse_early_nid(ms);
516 pnum_begin = pnum;
517 break;
431 } 518 }
519 usemap_count = 1;
520 for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
521 struct mem_section *ms;
522 int nodeid;
523
524 if (!present_section_nr(pnum))
525 continue;
526 ms = __nr_to_section(pnum);
527 nodeid = sparse_early_nid(ms);
528 if (nodeid == nodeid_begin) {
529 usemap_count++;
530 continue;
531 }
532 /* ok, we need to take cake of from pnum_begin to pnum - 1*/
533 sparse_early_usemaps_alloc_node(usemap_map, pnum_begin, pnum,
534 usemap_count, nodeid_begin);
535 /* new start, update count etc*/
536 nodeid_begin = nodeid;
537 pnum_begin = pnum;
538 usemap_count = 1;
539 }
540 /* ok, last chunk */
541 sparse_early_usemaps_alloc_node(usemap_map, pnum_begin, NR_MEM_SECTIONS,
542 usemap_count, nodeid_begin);
543
544#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
545 size2 = sizeof(struct page *) * NR_MEM_SECTIONS;
546 map_map = alloc_bootmem(size2);
547 if (!map_map)
548 panic("can not allocate map_map\n");
549
550 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
551 struct mem_section *ms;
552
553 if (!present_section_nr(pnum))
554 continue;
555 ms = __nr_to_section(pnum);
556 nodeid_begin = sparse_early_nid(ms);
557 pnum_begin = pnum;
558 break;
559 }
560 map_count = 1;
561 for (pnum = pnum_begin + 1; pnum < NR_MEM_SECTIONS; pnum++) {
562 struct mem_section *ms;
563 int nodeid;
564
565 if (!present_section_nr(pnum))
566 continue;
567 ms = __nr_to_section(pnum);
568 nodeid = sparse_early_nid(ms);
569 if (nodeid == nodeid_begin) {
570 map_count++;
571 continue;
572 }
573 /* ok, we need to take cake of from pnum_begin to pnum - 1*/
574 sparse_early_mem_maps_alloc_node(map_map, pnum_begin, pnum,
575 map_count, nodeid_begin);
576 /* new start, update count etc*/
577 nodeid_begin = nodeid;
578 pnum_begin = pnum;
579 map_count = 1;
580 }
581 /* ok, last chunk */
582 sparse_early_mem_maps_alloc_node(map_map, pnum_begin, NR_MEM_SECTIONS,
583 map_count, nodeid_begin);
584#endif
432 585
433 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) { 586 for (pnum = 0; pnum < NR_MEM_SECTIONS; pnum++) {
434 if (!present_section_nr(pnum)) 587 if (!present_section_nr(pnum))
@@ -438,7 +591,11 @@ void __init sparse_init(void)
438 if (!usemap) 591 if (!usemap)
439 continue; 592 continue;
440 593
594#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
595 map = map_map[pnum];
596#else
441 map = sparse_early_mem_map_alloc(pnum); 597 map = sparse_early_mem_map_alloc(pnum);
598#endif
442 if (!map) 599 if (!map)
443 continue; 600 continue;
444 601
@@ -448,6 +605,9 @@ void __init sparse_init(void)
448 605
449 vmemmap_populate_print_last(); 606 vmemmap_populate_print_last();
450 607
608#ifdef CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER
609 free_bootmem(__pa(map_map), size2);
610#endif
451 free_bootmem(__pa(usemap_map), size); 611 free_bootmem(__pa(usemap_map), size);
452} 612}
453 613
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-17 21:04:42 -0500