aboutsummaryrefslogtreecommitdiffstats
path: root/lib
diff options
context:
space:
mode:
Diffstat (limited to 'lib')
-rw-r--r--lib/Kconfig6
-rw-r--r--lib/Kconfig.debug70
-rw-r--r--lib/Makefile3
-rw-r--r--lib/atomic64_test.c2
-rw-r--r--lib/bug.c12
-rw-r--r--lib/decompress_bunzip2.c10
-rw-r--r--lib/devres.c2
-rw-r--r--lib/dynamic_debug.c2
-rw-r--r--lib/flex_array.c25
-rw-r--r--lib/genalloc.c1
-rw-r--r--lib/idr.c4
-rw-r--r--lib/inflate.c2
-rw-r--r--lib/iommu-helper.c9
-rw-r--r--lib/ioremap.c10
-rw-r--r--lib/list_debug.c6
-rw-r--r--lib/lmb.c541
-rw-r--r--lib/percpu_counter.c27
-rw-r--r--lib/radix-tree.c94
-rw-r--r--lib/raid6/Makefile75
-rw-r--r--lib/raid6/algos.c154
-rw-r--r--lib/raid6/altivec.uc130
-rw-r--r--lib/raid6/int.uc117
-rw-r--r--lib/raid6/mktables.c132
-rw-r--r--lib/raid6/mmx.c142
-rw-r--r--lib/raid6/recov.c132
-rw-r--r--lib/raid6/sse1.c162
-rw-r--r--lib/raid6/sse2.c262
-rw-r--r--lib/raid6/test/Makefile72
-rw-r--r--lib/raid6/test/test.c124
-rw-r--r--lib/raid6/unroll.awk20
-rw-r--r--lib/raid6/x86.h61
-rw-r--r--lib/random32.c2
-rw-r--r--lib/rbtree.c116
-rw-r--r--lib/rwsem.c150
-rw-r--r--lib/scatterlist.c23
-rw-r--r--lib/swiotlb.c137
-rw-r--r--lib/vsprintf.c23
37 files changed, 2088 insertions, 772 deletions
diff --git a/lib/Kconfig b/lib/Kconfig
index 170d8ca901d8..fa9bf2c06199 100644
--- a/lib/Kconfig
+++ b/lib/Kconfig
@@ -7,6 +7,9 @@ config BINARY_PRINTF
7 7
8menu "Library routines" 8menu "Library routines"
9 9
10config RAID6_PQ
11 tristate
12
10config BITREVERSE 13config BITREVERSE
11 tristate 14 tristate
12 15
@@ -181,9 +184,6 @@ config HAS_DMA
181config CHECK_SIGNATURE 184config CHECK_SIGNATURE
182 bool 185 bool
183 186
184config HAVE_LMB
185 boolean
186
187config CPUMASK_OFFSTACK 187config CPUMASK_OFFSTACK
188 bool "Force CPU masks off stack" if DEBUG_PER_CPU_MAPS 188 bool "Force CPU masks off stack" if DEBUG_PER_CPU_MAPS
189 help 189 help
diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug
index e722e9d62221..9e06b7f5ecf1 100644
--- a/lib/Kconfig.debug
+++ b/lib/Kconfig.debug
@@ -76,7 +76,6 @@ config UNUSED_SYMBOLS
76 76
77config DEBUG_FS 77config DEBUG_FS
78 bool "Debug Filesystem" 78 bool "Debug Filesystem"
79 depends on SYSFS
80 help 79 help
81 debugfs is a virtual file system that kernel developers use to put 80 debugfs is a virtual file system that kernel developers use to put
82 debugging files into. Enable this option to be able to read and 81 debugging files into. Enable this option to be able to read and
@@ -152,28 +151,33 @@ config DEBUG_SHIRQ
152 Drivers ought to be able to handle interrupts coming in at those 151 Drivers ought to be able to handle interrupts coming in at those
153 points; some don't and need to be caught. 152 points; some don't and need to be caught.
154 153
155config DETECT_SOFTLOCKUP 154config LOCKUP_DETECTOR
156 bool "Detect Soft Lockups" 155 bool "Detect Hard and Soft Lockups"
157 depends on DEBUG_KERNEL && !S390 156 depends on DEBUG_KERNEL && !S390
158 default y
159 help 157 help
160 Say Y here to enable the kernel to detect "soft lockups", 158 Say Y here to enable the kernel to act as a watchdog to detect
161 which are bugs that cause the kernel to loop in kernel 159 hard and soft lockups.
160
161 Softlockups are bugs that cause the kernel to loop in kernel
162 mode for more than 60 seconds, without giving other tasks a 162 mode for more than 60 seconds, without giving other tasks a
163 chance to run. 163 chance to run. The current stack trace is displayed upon
164 detection and the system will stay locked up.
164 165
165 When a soft-lockup is detected, the kernel will print the 166 Hardlockups are bugs that cause the CPU to loop in kernel mode
166 current stack trace (which you should report), but the 167 for more than 60 seconds, without letting other interrupts have a
167 system will stay locked up. This feature has negligible 168 chance to run. The current stack trace is displayed upon detection
168 overhead. 169 and the system will stay locked up.
170
171 The overhead should be minimal. A periodic hrtimer runs to
172 generate interrupts and kick the watchdog task every 10-12 seconds.
173 An NMI is generated every 60 seconds or so to check for hardlockups.
169 174
170 (Note that "hard lockups" are separate type of bugs that 175config HARDLOCKUP_DETECTOR
171 can be detected via the NMI-watchdog, on platforms that 176 def_bool LOCKUP_DETECTOR && PERF_EVENTS && HAVE_PERF_EVENTS_NMI
172 support it.)
173 177
174config BOOTPARAM_SOFTLOCKUP_PANIC 178config BOOTPARAM_SOFTLOCKUP_PANIC
175 bool "Panic (Reboot) On Soft Lockups" 179 bool "Panic (Reboot) On Soft Lockups"
176 depends on DETECT_SOFTLOCKUP 180 depends on LOCKUP_DETECTOR
177 help 181 help
178 Say Y here to enable the kernel to panic on "soft lockups", 182 Say Y here to enable the kernel to panic on "soft lockups",
179 which are bugs that cause the kernel to loop in kernel 183 which are bugs that cause the kernel to loop in kernel
@@ -190,7 +194,7 @@ config BOOTPARAM_SOFTLOCKUP_PANIC
190 194
191config BOOTPARAM_SOFTLOCKUP_PANIC_VALUE 195config BOOTPARAM_SOFTLOCKUP_PANIC_VALUE
192 int 196 int
193 depends on DETECT_SOFTLOCKUP 197 depends on LOCKUP_DETECTOR
194 range 0 1 198 range 0 1
195 default 0 if !BOOTPARAM_SOFTLOCKUP_PANIC 199 default 0 if !BOOTPARAM_SOFTLOCKUP_PANIC
196 default 1 if BOOTPARAM_SOFTLOCKUP_PANIC 200 default 1 if BOOTPARAM_SOFTLOCKUP_PANIC
@@ -307,6 +311,12 @@ config DEBUG_OBJECTS_WORK
307 work queue routines to track the life time of work objects and 311 work queue routines to track the life time of work objects and
308 validate the work operations. 312 validate the work operations.
309 313
314config DEBUG_OBJECTS_RCU_HEAD
315 bool "Debug RCU callbacks objects"
316 depends on DEBUG_OBJECTS && PREEMPT
317 help
318 Enable this to turn on debugging of RCU list heads (call_rcu() usage).
319
310config DEBUG_OBJECTS_ENABLE_DEFAULT 320config DEBUG_OBJECTS_ENABLE_DEFAULT
311 int "debug_objects bootup default value (0-1)" 321 int "debug_objects bootup default value (0-1)"
312 range 0 1 322 range 0 1
@@ -400,6 +410,13 @@ config DEBUG_KMEMLEAK_TEST
400 410
401 If unsure, say N. 411 If unsure, say N.
402 412
413config DEBUG_KMEMLEAK_DEFAULT_OFF
414 bool "Default kmemleak to off"
415 depends on DEBUG_KMEMLEAK
416 help
417 Say Y here to disable kmemleak by default. It can then be enabled
418 on the command line via kmemleak=on.
419
403config DEBUG_PREEMPT 420config DEBUG_PREEMPT
404 bool "Debug preemptible kernel" 421 bool "Debug preemptible kernel"
405 depends on DEBUG_KERNEL && PREEMPT && TRACE_IRQFLAGS_SUPPORT 422 depends on DEBUG_KERNEL && PREEMPT && TRACE_IRQFLAGS_SUPPORT
@@ -528,7 +545,7 @@ config LOCKDEP
528 bool 545 bool
529 depends on DEBUG_KERNEL && TRACE_IRQFLAGS_SUPPORT && STACKTRACE_SUPPORT && LOCKDEP_SUPPORT 546 depends on DEBUG_KERNEL && TRACE_IRQFLAGS_SUPPORT && STACKTRACE_SUPPORT && LOCKDEP_SUPPORT
530 select STACKTRACE 547 select STACKTRACE
531 select FRAME_POINTER if !MIPS && !PPC && !ARM_UNWIND && !S390 548 select FRAME_POINTER if !MIPS && !PPC && !ARM_UNWIND && !S390 && !MICROBLAZE
532 select KALLSYMS 549 select KALLSYMS
533 select KALLSYMS_ALL 550 select KALLSYMS_ALL
534 551
@@ -628,6 +645,19 @@ config DEBUG_INFO
628 645
629 If unsure, say N. 646 If unsure, say N.
630 647
648config DEBUG_INFO_REDUCED
649 bool "Reduce debugging information"
650 depends on DEBUG_INFO
651 help
652 If you say Y here gcc is instructed to generate less debugging
653 information for structure types. This means that tools that
654 need full debugging information (like kgdb or systemtap) won't
655 be happy. But if you merely need debugging information to
656 resolve line numbers there is no loss. Advantage is that
657 build directory object sizes shrink dramatically over a full
658 DEBUG_INFO build and compile times are reduced too.
659 Only works with newer gcc versions.
660
631config DEBUG_VM 661config DEBUG_VM
632 bool "Debug VM" 662 bool "Debug VM"
633 depends on DEBUG_KERNEL 663 depends on DEBUG_KERNEL
@@ -937,7 +967,7 @@ config FAIL_MAKE_REQUEST
937 Provide fault-injection capability for disk IO. 967 Provide fault-injection capability for disk IO.
938 968
939config FAIL_IO_TIMEOUT 969config FAIL_IO_TIMEOUT
940 bool "Faul-injection capability for faking disk interrupts" 970 bool "Fault-injection capability for faking disk interrupts"
941 depends on FAULT_INJECTION && BLOCK 971 depends on FAULT_INJECTION && BLOCK
942 help 972 help
943 Provide fault-injection capability on end IO handling. This 973 Provide fault-injection capability on end IO handling. This
@@ -958,13 +988,13 @@ config FAULT_INJECTION_STACKTRACE_FILTER
958 depends on FAULT_INJECTION_DEBUG_FS && STACKTRACE_SUPPORT 988 depends on FAULT_INJECTION_DEBUG_FS && STACKTRACE_SUPPORT
959 depends on !X86_64 989 depends on !X86_64
960 select STACKTRACE 990 select STACKTRACE
961 select FRAME_POINTER if !PPC && !S390 991 select FRAME_POINTER if !PPC && !S390 && !MICROBLAZE
962 help 992 help
963 Provide stacktrace filter for fault-injection capabilities 993 Provide stacktrace filter for fault-injection capabilities
964 994
965config LATENCYTOP 995config LATENCYTOP
966 bool "Latency measuring infrastructure" 996 bool "Latency measuring infrastructure"
967 select FRAME_POINTER if !MIPS && !PPC && !S390 997 select FRAME_POINTER if !MIPS && !PPC && !S390 && !MICROBLAZE
968 select KALLSYMS 998 select KALLSYMS
969 select KALLSYMS_ALL 999 select KALLSYMS_ALL
970 select STACKTRACE 1000 select STACKTRACE
diff --git a/lib/Makefile b/lib/Makefile
index 3f1062cbbff4..e6a3763b8212 100644
--- a/lib/Makefile
+++ b/lib/Makefile
@@ -69,6 +69,7 @@ obj-$(CONFIG_ZLIB_DEFLATE) += zlib_deflate/
69obj-$(CONFIG_REED_SOLOMON) += reed_solomon/ 69obj-$(CONFIG_REED_SOLOMON) += reed_solomon/
70obj-$(CONFIG_LZO_COMPRESS) += lzo/ 70obj-$(CONFIG_LZO_COMPRESS) += lzo/
71obj-$(CONFIG_LZO_DECOMPRESS) += lzo/ 71obj-$(CONFIG_LZO_DECOMPRESS) += lzo/
72obj-$(CONFIG_RAID6_PQ) += raid6/
72 73
73lib-$(CONFIG_DECOMPRESS_GZIP) += decompress_inflate.o 74lib-$(CONFIG_DECOMPRESS_GZIP) += decompress_inflate.o
74lib-$(CONFIG_DECOMPRESS_BZIP2) += decompress_bunzip2.o 75lib-$(CONFIG_DECOMPRESS_BZIP2) += decompress_bunzip2.o
@@ -89,8 +90,6 @@ obj-$(CONFIG_CPU_NOTIFIER_ERROR_INJECT) += cpu-notifier-error-inject.o
89 90
90lib-$(CONFIG_GENERIC_BUG) += bug.o 91lib-$(CONFIG_GENERIC_BUG) += bug.o
91 92
92obj-$(CONFIG_HAVE_LMB) += lmb.o
93
94obj-$(CONFIG_HAVE_ARCH_TRACEHOOK) += syscall.o 93obj-$(CONFIG_HAVE_ARCH_TRACEHOOK) += syscall.o
95 94
96obj-$(CONFIG_DYNAMIC_DEBUG) += dynamic_debug.o 95obj-$(CONFIG_DYNAMIC_DEBUG) += dynamic_debug.o
diff --git a/lib/atomic64_test.c b/lib/atomic64_test.c
index 250ed11d3ed2..44524cc8c32a 100644
--- a/lib/atomic64_test.c
+++ b/lib/atomic64_test.c
@@ -114,7 +114,7 @@ static __init int test_atomic64(void)
114 BUG_ON(v.counter != r); 114 BUG_ON(v.counter != r);
115 115
116#if defined(CONFIG_X86) || defined(CONFIG_MIPS) || defined(CONFIG_PPC) || \ 116#if defined(CONFIG_X86) || defined(CONFIG_MIPS) || defined(CONFIG_PPC) || \
117 defined(CONFIG_S390) || defined(_ASM_GENERIC_ATOMIC64_H) 117 defined(CONFIG_S390) || defined(_ASM_GENERIC_ATOMIC64_H) || defined(CONFIG_ARM)
118 INIT(onestwos); 118 INIT(onestwos);
119 BUG_ON(atomic64_dec_if_positive(&v) != (onestwos - 1)); 119 BUG_ON(atomic64_dec_if_positive(&v) != (onestwos - 1));
120 r -= one; 120 r -= one;
diff --git a/lib/bug.c b/lib/bug.c
index f13daf435211..7cdfad88128f 100644
--- a/lib/bug.c
+++ b/lib/bug.c
@@ -136,8 +136,6 @@ enum bug_trap_type report_bug(unsigned long bugaddr, struct pt_regs *regs)
136 136
137 bug = find_bug(bugaddr); 137 bug = find_bug(bugaddr);
138 138
139 printk(KERN_EMERG "------------[ cut here ]------------\n");
140
141 file = NULL; 139 file = NULL;
142 line = 0; 140 line = 0;
143 warning = 0; 141 warning = 0;
@@ -156,19 +154,25 @@ enum bug_trap_type report_bug(unsigned long bugaddr, struct pt_regs *regs)
156 154
157 if (warning) { 155 if (warning) {
158 /* this is a WARN_ON rather than BUG/BUG_ON */ 156 /* this is a WARN_ON rather than BUG/BUG_ON */
157 printk(KERN_WARNING "------------[ cut here ]------------\n");
158
159 if (file) 159 if (file)
160 printk(KERN_ERR "Badness at %s:%u\n", 160 printk(KERN_WARNING "WARNING: at %s:%u\n",
161 file, line); 161 file, line);
162 else 162 else
163 printk(KERN_ERR "Badness at %p " 163 printk(KERN_WARNING "WARNING: at %p "
164 "[verbose debug info unavailable]\n", 164 "[verbose debug info unavailable]\n",
165 (void *)bugaddr); 165 (void *)bugaddr);
166 166
167 print_modules();
167 show_regs(regs); 168 show_regs(regs);
169 print_oops_end_marker();
168 add_taint(BUG_GET_TAINT(bug)); 170 add_taint(BUG_GET_TAINT(bug));
169 return BUG_TRAP_TYPE_WARN; 171 return BUG_TRAP_TYPE_WARN;
170 } 172 }
171 173
174 printk(KERN_EMERG "------------[ cut here ]------------\n");
175
172 if (file) 176 if (file)
173 printk(KERN_CRIT "kernel BUG at %s:%u!\n", 177 printk(KERN_CRIT "kernel BUG at %s:%u!\n",
174 file, line); 178 file, line);
diff --git a/lib/decompress_bunzip2.c b/lib/decompress_bunzip2.c
index a4e971dee102..81c8bb1cc6aa 100644
--- a/lib/decompress_bunzip2.c
+++ b/lib/decompress_bunzip2.c
@@ -107,6 +107,8 @@ struct bunzip_data {
107 unsigned char selectors[32768]; /* nSelectors = 15 bits */ 107 unsigned char selectors[32768]; /* nSelectors = 15 bits */
108 struct group_data groups[MAX_GROUPS]; /* Huffman coding tables */ 108 struct group_data groups[MAX_GROUPS]; /* Huffman coding tables */
109 int io_error; /* non-zero if we have IO error */ 109 int io_error; /* non-zero if we have IO error */
110 int byteCount[256];
111 unsigned char symToByte[256], mtfSymbol[256];
110}; 112};
111 113
112 114
@@ -158,14 +160,16 @@ static int INIT get_next_block(struct bunzip_data *bd)
158 int *base = NULL; 160 int *base = NULL;
159 int *limit = NULL; 161 int *limit = NULL;
160 int dbufCount, nextSym, dbufSize, groupCount, selector, 162 int dbufCount, nextSym, dbufSize, groupCount, selector,
161 i, j, k, t, runPos, symCount, symTotal, nSelectors, 163 i, j, k, t, runPos, symCount, symTotal, nSelectors, *byteCount;
162 byteCount[256]; 164 unsigned char uc, *symToByte, *mtfSymbol, *selectors;
163 unsigned char uc, symToByte[256], mtfSymbol[256], *selectors;
164 unsigned int *dbuf, origPtr; 165 unsigned int *dbuf, origPtr;
165 166
166 dbuf = bd->dbuf; 167 dbuf = bd->dbuf;
167 dbufSize = bd->dbufSize; 168 dbufSize = bd->dbufSize;
168 selectors = bd->selectors; 169 selectors = bd->selectors;
170 byteCount = bd->byteCount;
171 symToByte = bd->symToByte;
172 mtfSymbol = bd->mtfSymbol;
169 173
170 /* Read in header signature and CRC, then validate signature. 174 /* Read in header signature and CRC, then validate signature.
171 (last block signature means CRC is for whole file, return now) */ 175 (last block signature means CRC is for whole file, return now) */
diff --git a/lib/devres.c b/lib/devres.c
index 49368608f988..6efddf53b90c 100644
--- a/lib/devres.c
+++ b/lib/devres.c
@@ -328,7 +328,7 @@ EXPORT_SYMBOL(pcim_iomap_regions_request_all);
328 * @pdev: PCI device to map IO resources for 328 * @pdev: PCI device to map IO resources for
329 * @mask: Mask of BARs to unmap and release 329 * @mask: Mask of BARs to unmap and release
330 * 330 *
331 * Unamp and release regions specified by @mask. 331 * Unmap and release regions specified by @mask.
332 */ 332 */
333void pcim_iounmap_regions(struct pci_dev *pdev, u16 mask) 333void pcim_iounmap_regions(struct pci_dev *pdev, u16 mask)
334{ 334{
diff --git a/lib/dynamic_debug.c b/lib/dynamic_debug.c
index 3df8eb17a607..02afc2533728 100644
--- a/lib/dynamic_debug.c
+++ b/lib/dynamic_debug.c
@@ -692,7 +692,7 @@ static void ddebug_table_free(struct ddebug_table *dt)
692 * Called in response to a module being unloaded. Removes 692 * Called in response to a module being unloaded. Removes
693 * any ddebug_table's which point at the module. 693 * any ddebug_table's which point at the module.
694 */ 694 */
695int ddebug_remove_module(char *mod_name) 695int ddebug_remove_module(const char *mod_name)
696{ 696{
697 struct ddebug_table *dt, *nextdt; 697 struct ddebug_table *dt, *nextdt;
698 int ret = -ENOENT; 698 int ret = -ENOENT;
diff --git a/lib/flex_array.c b/lib/flex_array.c
index 41b1804fa728..77a6fea7481e 100644
--- a/lib/flex_array.c
+++ b/lib/flex_array.c
@@ -171,6 +171,8 @@ __fa_get_part(struct flex_array *fa, int part_nr, gfp_t flags)
171 * Note that this *copies* the contents of @src into 171 * Note that this *copies* the contents of @src into
172 * the array. If you are trying to store an array of 172 * the array. If you are trying to store an array of
173 * pointers, make sure to pass in &ptr instead of ptr. 173 * pointers, make sure to pass in &ptr instead of ptr.
174 * You may instead wish to use the flex_array_put_ptr()
175 * helper function.
174 * 176 *
175 * Locking must be provided by the caller. 177 * Locking must be provided by the caller.
176 */ 178 */
@@ -265,7 +267,8 @@ int flex_array_prealloc(struct flex_array *fa, unsigned int start,
265 * 267 *
266 * Returns a pointer to the data at index @element_nr. Note 268 * Returns a pointer to the data at index @element_nr. Note
267 * that this is a copy of the data that was passed in. If you 269 * that this is a copy of the data that was passed in. If you
268 * are using this to store pointers, you'll get back &ptr. 270 * are using this to store pointers, you'll get back &ptr. You
271 * may instead wish to use the flex_array_get_ptr helper.
269 * 272 *
270 * Locking must be provided by the caller. 273 * Locking must be provided by the caller.
271 */ 274 */
@@ -286,6 +289,26 @@ void *flex_array_get(struct flex_array *fa, unsigned int element_nr)
286 return &part->elements[index_inside_part(fa, element_nr)]; 289 return &part->elements[index_inside_part(fa, element_nr)];
287} 290}
288 291
292/**
293 * flex_array_get_ptr - pull a ptr back out of the array
294 * @fa: the flex array from which to extract data
295 * @element_nr: index of the element to fetch from the array
296 *
297 * Returns the pointer placed in the flex array at element_nr using
298 * flex_array_put_ptr(). This function should not be called if the
299 * element in question was not set using the _put_ptr() helper.
300 */
301void *flex_array_get_ptr(struct flex_array *fa, unsigned int element_nr)
302{
303 void **tmp;
304
305 tmp = flex_array_get(fa, element_nr);
306 if (!tmp)
307 return NULL;
308
309 return *tmp;
310}
311
289static int part_is_free(struct flex_array_part *part) 312static int part_is_free(struct flex_array_part *part)
290{ 313{
291 int i; 314 int i;
diff --git a/lib/genalloc.c b/lib/genalloc.c
index 736c3b06398e..1923f1490e72 100644
--- a/lib/genalloc.c
+++ b/lib/genalloc.c
@@ -128,7 +128,6 @@ unsigned long gen_pool_alloc(struct gen_pool *pool, size_t size)
128 chunk = list_entry(_chunk, struct gen_pool_chunk, next_chunk); 128 chunk = list_entry(_chunk, struct gen_pool_chunk, next_chunk);
129 129
130 end_bit = (chunk->end_addr - chunk->start_addr) >> order; 130 end_bit = (chunk->end_addr - chunk->start_addr) >> order;
131 end_bit -= nbits + 1;
132 131
133 spin_lock_irqsave(&chunk->lock, flags); 132 spin_lock_irqsave(&chunk->lock, flags);
134 start_bit = bitmap_find_next_zero_area(chunk->bits, end_bit, 0, 133 start_bit = bitmap_find_next_zero_area(chunk->bits, end_bit, 0,
diff --git a/lib/idr.c b/lib/idr.c
index c1a206901761..7f1a4f0acf50 100644
--- a/lib/idr.c
+++ b/lib/idr.c
@@ -602,7 +602,7 @@ void *idr_get_next(struct idr *idp, int *nextidp)
602 /* find first ent */ 602 /* find first ent */
603 n = idp->layers * IDR_BITS; 603 n = idp->layers * IDR_BITS;
604 max = 1 << n; 604 max = 1 << n;
605 p = rcu_dereference(idp->top); 605 p = rcu_dereference_raw(idp->top);
606 if (!p) 606 if (!p)
607 return NULL; 607 return NULL;
608 608
@@ -610,7 +610,7 @@ void *idr_get_next(struct idr *idp, int *nextidp)
610 while (n > 0 && p) { 610 while (n > 0 && p) {
611 n -= IDR_BITS; 611 n -= IDR_BITS;
612 *paa++ = p; 612 *paa++ = p;
613 p = rcu_dereference(p->ary[(id >> n) & IDR_MASK]); 613 p = rcu_dereference_raw(p->ary[(id >> n) & IDR_MASK]);
614 } 614 }
615 615
616 if (p) { 616 if (p) {
diff --git a/lib/inflate.c b/lib/inflate.c
index 677b738c2204..013a76193481 100644
--- a/lib/inflate.c
+++ b/lib/inflate.c
@@ -103,7 +103,9 @@
103 the two sets of lengths. 103 the two sets of lengths.
104 */ 104 */
105#include <linux/compiler.h> 105#include <linux/compiler.h>
106#ifdef NO_INFLATE_MALLOC
106#include <linux/slab.h> 107#include <linux/slab.h>
108#endif
107 109
108#ifdef RCSID 110#ifdef RCSID
109static char rcsid[] = "#Id: inflate.c,v 0.14 1993/06/10 13:27:04 jloup Exp #"; 111static char rcsid[] = "#Id: inflate.c,v 0.14 1993/06/10 13:27:04 jloup Exp #";
diff --git a/lib/iommu-helper.c b/lib/iommu-helper.c
index c0251f4ad08b..da053313ee5c 100644
--- a/lib/iommu-helper.c
+++ b/lib/iommu-helper.c
@@ -38,12 +38,3 @@ again:
38 return -1; 38 return -1;
39} 39}
40EXPORT_SYMBOL(iommu_area_alloc); 40EXPORT_SYMBOL(iommu_area_alloc);
41
42unsigned long iommu_num_pages(unsigned long addr, unsigned long len,
43 unsigned long io_page_size)
44{
45 unsigned long size = (addr & (io_page_size - 1)) + len;
46
47 return DIV_ROUND_UP(size, io_page_size);
48}
49EXPORT_SYMBOL(iommu_num_pages);
diff --git a/lib/ioremap.c b/lib/ioremap.c
index 14c6078f17a2..5730ecd3eb66 100644
--- a/lib/ioremap.c
+++ b/lib/ioremap.c
@@ -13,10 +13,10 @@
13#include <asm/pgtable.h> 13#include <asm/pgtable.h>
14 14
15static int ioremap_pte_range(pmd_t *pmd, unsigned long addr, 15static int ioremap_pte_range(pmd_t *pmd, unsigned long addr,
16 unsigned long end, unsigned long phys_addr, pgprot_t prot) 16 unsigned long end, phys_addr_t phys_addr, pgprot_t prot)
17{ 17{
18 pte_t *pte; 18 pte_t *pte;
19 unsigned long pfn; 19 u64 pfn;
20 20
21 pfn = phys_addr >> PAGE_SHIFT; 21 pfn = phys_addr >> PAGE_SHIFT;
22 pte = pte_alloc_kernel(pmd, addr); 22 pte = pte_alloc_kernel(pmd, addr);
@@ -31,7 +31,7 @@ static int ioremap_pte_range(pmd_t *pmd, unsigned long addr,
31} 31}
32 32
33static inline int ioremap_pmd_range(pud_t *pud, unsigned long addr, 33static inline int ioremap_pmd_range(pud_t *pud, unsigned long addr,
34 unsigned long end, unsigned long phys_addr, pgprot_t prot) 34 unsigned long end, phys_addr_t phys_addr, pgprot_t prot)
35{ 35{
36 pmd_t *pmd; 36 pmd_t *pmd;
37 unsigned long next; 37 unsigned long next;
@@ -49,7 +49,7 @@ static inline int ioremap_pmd_range(pud_t *pud, unsigned long addr,
49} 49}
50 50
51static inline int ioremap_pud_range(pgd_t *pgd, unsigned long addr, 51static inline int ioremap_pud_range(pgd_t *pgd, unsigned long addr,
52 unsigned long end, unsigned long phys_addr, pgprot_t prot) 52 unsigned long end, phys_addr_t phys_addr, pgprot_t prot)
53{ 53{
54 pud_t *pud; 54 pud_t *pud;
55 unsigned long next; 55 unsigned long next;
@@ -67,7 +67,7 @@ static inline int ioremap_pud_range(pgd_t *pgd, unsigned long addr,
67} 67}
68 68
69int ioremap_page_range(unsigned long addr, 69int ioremap_page_range(unsigned long addr,
70 unsigned long end, unsigned long phys_addr, pgprot_t prot) 70 unsigned long end, phys_addr_t phys_addr, pgprot_t prot)
71{ 71{
72 pgd_t *pgd; 72 pgd_t *pgd;
73 unsigned long start; 73 unsigned long start;
diff --git a/lib/list_debug.c b/lib/list_debug.c
index 1a39f4e3ae1f..344c710d16ca 100644
--- a/lib/list_debug.c
+++ b/lib/list_debug.c
@@ -43,6 +43,12 @@ EXPORT_SYMBOL(__list_add);
43 */ 43 */
44void list_del(struct list_head *entry) 44void list_del(struct list_head *entry)
45{ 45{
46 WARN(entry->next == LIST_POISON1,
47 "list_del corruption, next is LIST_POISON1 (%p)\n",
48 LIST_POISON1);
49 WARN(entry->next != LIST_POISON1 && entry->prev == LIST_POISON2,
50 "list_del corruption, prev is LIST_POISON2 (%p)\n",
51 LIST_POISON2);
46 WARN(entry->prev->next != entry, 52 WARN(entry->prev->next != entry,
47 "list_del corruption. prev->next should be %p, " 53 "list_del corruption. prev->next should be %p, "
48 "but was %p\n", entry, entry->prev->next); 54 "but was %p\n", entry, entry->prev->next);
diff --git a/lib/lmb.c b/lib/lmb.c
deleted file mode 100644
index b1fc52606524..000000000000
--- a/lib/lmb.c
+++ /dev/null
@@ -1,541 +0,0 @@
1/*
2 * Procedures for maintaining information about logical memory blocks.
3 *
4 * Peter Bergner, IBM Corp. June 2001.
5 * Copyright (C) 2001 Peter Bergner.
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
11 */
12
13#include <linux/kernel.h>
14#include <linux/init.h>
15#include <linux/bitops.h>
16#include <linux/lmb.h>
17
18#define LMB_ALLOC_ANYWHERE 0
19
20struct lmb lmb;
21
22static int lmb_debug;
23
24static int __init early_lmb(char *p)
25{
26 if (p && strstr(p, "debug"))
27 lmb_debug = 1;
28 return 0;
29}
30early_param("lmb", early_lmb);
31
32static void lmb_dump(struct lmb_region *region, char *name)
33{
34 unsigned long long base, size;
35 int i;
36
37 pr_info(" %s.cnt = 0x%lx\n", name, region->cnt);
38
39 for (i = 0; i < region->cnt; i++) {
40 base = region->region[i].base;
41 size = region->region[i].size;
42
43 pr_info(" %s[0x%x]\t0x%016llx - 0x%016llx, 0x%llx bytes\n",
44 name, i, base, base + size - 1, size);
45 }
46}
47
48void lmb_dump_all(void)
49{
50 if (!lmb_debug)
51 return;
52
53 pr_info("LMB configuration:\n");
54 pr_info(" rmo_size = 0x%llx\n", (unsigned long long)lmb.rmo_size);
55 pr_info(" memory.size = 0x%llx\n", (unsigned long long)lmb.memory.size);
56
57 lmb_dump(&lmb.memory, "memory");
58 lmb_dump(&lmb.reserved, "reserved");
59}
60
61static unsigned long lmb_addrs_overlap(u64 base1, u64 size1, u64 base2,
62 u64 size2)
63{
64 return ((base1 < (base2 + size2)) && (base2 < (base1 + size1)));
65}
66
67static long lmb_addrs_adjacent(u64 base1, u64 size1, u64 base2, u64 size2)
68{
69 if (base2 == base1 + size1)
70 return 1;
71 else if (base1 == base2 + size2)
72 return -1;
73
74 return 0;
75}
76
77static long lmb_regions_adjacent(struct lmb_region *rgn,
78 unsigned long r1, unsigned long r2)
79{
80 u64 base1 = rgn->region[r1].base;
81 u64 size1 = rgn->region[r1].size;
82 u64 base2 = rgn->region[r2].base;
83 u64 size2 = rgn->region[r2].size;
84
85 return lmb_addrs_adjacent(base1, size1, base2, size2);
86}
87
88static void lmb_remove_region(struct lmb_region *rgn, unsigned long r)
89{
90 unsigned long i;
91
92 for (i = r; i < rgn->cnt - 1; i++) {
93 rgn->region[i].base = rgn->region[i + 1].base;
94 rgn->region[i].size = rgn->region[i + 1].size;
95 }
96 rgn->cnt--;
97}
98
99/* Assumption: base addr of region 1 < base addr of region 2 */
100static void lmb_coalesce_regions(struct lmb_region *rgn,
101 unsigned long r1, unsigned long r2)
102{
103 rgn->region[r1].size += rgn->region[r2].size;
104 lmb_remove_region(rgn, r2);
105}
106
107void __init lmb_init(void)
108{
109 /* Create a dummy zero size LMB which will get coalesced away later.
110 * This simplifies the lmb_add() code below...
111 */
112 lmb.memory.region[0].base = 0;
113 lmb.memory.region[0].size = 0;
114 lmb.memory.cnt = 1;
115
116 /* Ditto. */
117 lmb.reserved.region[0].base = 0;
118 lmb.reserved.region[0].size = 0;
119 lmb.reserved.cnt = 1;
120}
121
122void __init lmb_analyze(void)
123{
124 int i;
125
126 lmb.memory.size = 0;
127
128 for (i = 0; i < lmb.memory.cnt; i++)
129 lmb.memory.size += lmb.memory.region[i].size;
130}
131
132static long lmb_add_region(struct lmb_region *rgn, u64 base, u64 size)
133{
134 unsigned long coalesced = 0;
135 long adjacent, i;
136
137 if ((rgn->cnt == 1) && (rgn->region[0].size == 0)) {
138 rgn->region[0].base = base;
139 rgn->region[0].size = size;
140 return 0;
141 }
142
143 /* First try and coalesce this LMB with another. */
144 for (i = 0; i < rgn->cnt; i++) {
145 u64 rgnbase = rgn->region[i].base;
146 u64 rgnsize = rgn->region[i].size;
147
148 if ((rgnbase == base) && (rgnsize == size))
149 /* Already have this region, so we're done */
150 return 0;
151
152 adjacent = lmb_addrs_adjacent(base, size, rgnbase, rgnsize);
153 if (adjacent > 0) {
154 rgn->region[i].base -= size;
155 rgn->region[i].size += size;
156 coalesced++;
157 break;
158 } else if (adjacent < 0) {
159 rgn->region[i].size += size;
160 coalesced++;
161 break;
162 }
163 }
164
165 if ((i < rgn->cnt - 1) && lmb_regions_adjacent(rgn, i, i+1)) {
166 lmb_coalesce_regions(rgn, i, i+1);
167 coalesced++;
168 }
169
170 if (coalesced)
171 return coalesced;
172 if (rgn->cnt >= MAX_LMB_REGIONS)
173 return -1;
174
175 /* Couldn't coalesce the LMB, so add it to the sorted table. */
176 for (i = rgn->cnt - 1; i >= 0; i--) {
177 if (base < rgn->region[i].base) {
178 rgn->region[i+1].base = rgn->region[i].base;
179 rgn->region[i+1].size = rgn->region[i].size;
180 } else {
181 rgn->region[i+1].base = base;
182 rgn->region[i+1].size = size;
183 break;
184 }
185 }
186
187 if (base < rgn->region[0].base) {
188 rgn->region[0].base = base;
189 rgn->region[0].size = size;
190 }
191 rgn->cnt++;
192
193 return 0;
194}
195
196long lmb_add(u64 base, u64 size)
197{
198 struct lmb_region *_rgn = &lmb.memory;
199
200 /* On pSeries LPAR systems, the first LMB is our RMO region. */
201 if (base == 0)
202 lmb.rmo_size = size;
203
204 return lmb_add_region(_rgn, base, size);
205
206}
207
208static long __lmb_remove(struct lmb_region *rgn, u64 base, u64 size)
209{
210 u64 rgnbegin, rgnend;
211 u64 end = base + size;
212 int i;
213
214 rgnbegin = rgnend = 0; /* supress gcc warnings */
215
216 /* Find the region where (base, size) belongs to */
217 for (i=0; i < rgn->cnt; i++) {
218 rgnbegin = rgn->region[i].base;
219 rgnend = rgnbegin + rgn->region[i].size;
220
221 if ((rgnbegin <= base) && (end <= rgnend))
222 break;
223 }
224
225 /* Didn't find the region */
226 if (i == rgn->cnt)
227 return -1;
228
229 /* Check to see if we are removing entire region */
230 if ((rgnbegin == base) && (rgnend == end)) {
231 lmb_remove_region(rgn, i);
232 return 0;
233 }
234
235 /* Check to see if region is matching at the front */
236 if (rgnbegin == base) {
237 rgn->region[i].base = end;
238 rgn->region[i].size -= size;
239 return 0;
240 }
241
242 /* Check to see if the region is matching at the end */
243 if (rgnend == end) {
244 rgn->region[i].size -= size;
245 return 0;
246 }
247
248 /*
249 * We need to split the entry - adjust the current one to the
250 * beginging of the hole and add the region after hole.
251 */
252 rgn->region[i].size = base - rgn->region[i].base;
253 return lmb_add_region(rgn, end, rgnend - end);
254}
255
256long lmb_remove(u64 base, u64 size)
257{
258 return __lmb_remove(&lmb.memory, base, size);
259}
260
261long __init lmb_free(u64 base, u64 size)
262{
263 return __lmb_remove(&lmb.reserved, base, size);
264}
265
266long __init lmb_reserve(u64 base, u64 size)
267{
268 struct lmb_region *_rgn = &lmb.reserved;
269
270 BUG_ON(0 == size);
271
272 return lmb_add_region(_rgn, base, size);
273}
274
275long lmb_overlaps_region(struct lmb_region *rgn, u64 base, u64 size)
276{
277 unsigned long i;
278
279 for (i = 0; i < rgn->cnt; i++) {
280 u64 rgnbase = rgn->region[i].base;
281 u64 rgnsize = rgn->region[i].size;
282 if (lmb_addrs_overlap(base, size, rgnbase, rgnsize))
283 break;
284 }
285
286 return (i < rgn->cnt) ? i : -1;
287}
288
289static u64 lmb_align_down(u64 addr, u64 size)
290{
291 return addr & ~(size - 1);
292}
293
294static u64 lmb_align_up(u64 addr, u64 size)
295{
296 return (addr + (size - 1)) & ~(size - 1);
297}
298
299static u64 __init lmb_alloc_nid_unreserved(u64 start, u64 end,
300 u64 size, u64 align)
301{
302 u64 base, res_base;
303 long j;
304
305 base = lmb_align_down((end - size), align);
306 while (start <= base) {
307 j = lmb_overlaps_region(&lmb.reserved, base, size);
308 if (j < 0) {
309 /* this area isn't reserved, take it */
310 if (lmb_add_region(&lmb.reserved, base, size) < 0)
311 base = ~(u64)0;
312 return base;
313 }
314 res_base = lmb.reserved.region[j].base;
315 if (res_base < size)
316 break;
317 base = lmb_align_down(res_base - size, align);
318 }
319
320 return ~(u64)0;
321}
322
323static u64 __init lmb_alloc_nid_region(struct lmb_property *mp,
324 u64 (*nid_range)(u64, u64, int *),
325 u64 size, u64 align, int nid)
326{
327 u64 start, end;
328
329 start = mp->base;
330 end = start + mp->size;
331
332 start = lmb_align_up(start, align);
333 while (start < end) {
334 u64 this_end;
335 int this_nid;
336
337 this_end = nid_range(start, end, &this_nid);
338 if (this_nid == nid) {
339 u64 ret = lmb_alloc_nid_unreserved(start, this_end,
340 size, align);
341 if (ret != ~(u64)0)
342 return ret;
343 }
344 start = this_end;
345 }
346
347 return ~(u64)0;
348}
349
350u64 __init lmb_alloc_nid(u64 size, u64 align, int nid,
351 u64 (*nid_range)(u64 start, u64 end, int *nid))
352{
353 struct lmb_region *mem = &lmb.memory;
354 int i;
355
356 BUG_ON(0 == size);
357
358 size = lmb_align_up(size, align);
359
360 for (i = 0; i < mem->cnt; i++) {
361 u64 ret = lmb_alloc_nid_region(&mem->region[i],
362 nid_range,
363 size, align, nid);
364 if (ret != ~(u64)0)
365 return ret;
366 }
367
368 return lmb_alloc(size, align);
369}
370
371u64 __init lmb_alloc(u64 size, u64 align)
372{
373 return lmb_alloc_base(size, align, LMB_ALLOC_ANYWHERE);
374}
375
376u64 __init lmb_alloc_base(u64 size, u64 align, u64 max_addr)
377{
378 u64 alloc;
379
380 alloc = __lmb_alloc_base(size, align, max_addr);
381
382 if (alloc == 0)
383 panic("ERROR: Failed to allocate 0x%llx bytes below 0x%llx.\n",
384 (unsigned long long) size, (unsigned long long) max_addr);
385
386 return alloc;
387}
388
389u64 __init __lmb_alloc_base(u64 size, u64 align, u64 max_addr)
390{
391 long i, j;
392 u64 base = 0;
393 u64 res_base;
394
395 BUG_ON(0 == size);
396
397 size = lmb_align_up(size, align);
398
399 /* On some platforms, make sure we allocate lowmem */
400 /* Note that LMB_REAL_LIMIT may be LMB_ALLOC_ANYWHERE */
401 if (max_addr == LMB_ALLOC_ANYWHERE)
402 max_addr = LMB_REAL_LIMIT;
403
404 for (i = lmb.memory.cnt - 1; i >= 0; i--) {
405 u64 lmbbase = lmb.memory.region[i].base;
406 u64 lmbsize = lmb.memory.region[i].size;
407
408 if (lmbsize < size)
409 continue;
410 if (max_addr == LMB_ALLOC_ANYWHERE)
411 base = lmb_align_down(lmbbase + lmbsize - size, align);
412 else if (lmbbase < max_addr) {
413 base = min(lmbbase + lmbsize, max_addr);
414 base = lmb_align_down(base - size, align);
415 } else
416 continue;
417
418 while (base && lmbbase <= base) {
419 j = lmb_overlaps_region(&lmb.reserved, base, size);
420 if (j < 0) {
421 /* this area isn't reserved, take it */
422 if (lmb_add_region(&lmb.reserved, base, size) < 0)
423 return 0;
424 return base;
425 }
426 res_base = lmb.reserved.region[j].base;
427 if (res_base < size)
428 break;
429 base = lmb_align_down(res_base - size, align);
430 }
431 }
432 return 0;
433}
434
435/* You must call lmb_analyze() before this. */
436u64 __init lmb_phys_mem_size(void)
437{
438 return lmb.memory.size;
439}
440
441u64 lmb_end_of_DRAM(void)
442{
443 int idx = lmb.memory.cnt - 1;
444
445 return (lmb.memory.region[idx].base + lmb.memory.region[idx].size);
446}
447
448/* You must call lmb_analyze() after this. */
449void __init lmb_enforce_memory_limit(u64 memory_limit)
450{
451 unsigned long i;
452 u64 limit;
453 struct lmb_property *p;
454
455 if (!memory_limit)
456 return;
457
458 /* Truncate the lmb regions to satisfy the memory limit. */
459 limit = memory_limit;
460 for (i = 0; i < lmb.memory.cnt; i++) {
461 if (limit > lmb.memory.region[i].size) {
462 limit -= lmb.memory.region[i].size;
463 continue;
464 }
465
466 lmb.memory.region[i].size = limit;
467 lmb.memory.cnt = i + 1;
468 break;
469 }
470
471 if (lmb.memory.region[0].size < lmb.rmo_size)
472 lmb.rmo_size = lmb.memory.region[0].size;
473
474 memory_limit = lmb_end_of_DRAM();
475
476 /* And truncate any reserves above the limit also. */
477 for (i = 0; i < lmb.reserved.cnt; i++) {
478 p = &lmb.reserved.region[i];
479
480 if (p->base > memory_limit)
481 p->size = 0;
482 else if ((p->base + p->size) > memory_limit)
483 p->size = memory_limit - p->base;
484
485 if (p->size == 0) {
486 lmb_remove_region(&lmb.reserved, i);
487 i--;
488 }
489 }
490}
491
492int __init lmb_is_reserved(u64 addr)
493{
494 int i;
495
496 for (i = 0; i < lmb.reserved.cnt; i++) {
497 u64 upper = lmb.reserved.region[i].base +
498 lmb.reserved.region[i].size - 1;
499 if ((addr >= lmb.reserved.region[i].base) && (addr <= upper))
500 return 1;
501 }
502 return 0;
503}
504
505int lmb_is_region_reserved(u64 base, u64 size)
506{
507 return lmb_overlaps_region(&lmb.reserved, base, size);
508}
509
510/*
511 * Given a <base, len>, find which memory regions belong to this range.
512 * Adjust the request and return a contiguous chunk.
513 */
514int lmb_find(struct lmb_property *res)
515{
516 int i;
517 u64 rstart, rend;
518
519 rstart = res->base;
520 rend = rstart + res->size - 1;
521
522 for (i = 0; i < lmb.memory.cnt; i++) {
523 u64 start = lmb.memory.region[i].base;
524 u64 end = start + lmb.memory.region[i].size - 1;
525
526 if (start > rend)
527 return -1;
528
529 if ((end >= rstart) && (start < rend)) {
530 /* adjust the request */
531 if (rstart < start)
532 rstart = start;
533 if (rend > end)
534 rend = end;
535 res->base = rstart;
536 res->size = rend - rstart + 1;
537 return 0;
538 }
539 }
540 return -1;
541}
diff --git a/lib/percpu_counter.c b/lib/percpu_counter.c
index aeaa6d734447..ec9048e74f44 100644
--- a/lib/percpu_counter.c
+++ b/lib/percpu_counter.c
@@ -137,6 +137,33 @@ static int __cpuinit percpu_counter_hotcpu_callback(struct notifier_block *nb,
137 return NOTIFY_OK; 137 return NOTIFY_OK;
138} 138}
139 139
140/*
141 * Compare counter against given value.
142 * Return 1 if greater, 0 if equal and -1 if less
143 */
144int percpu_counter_compare(struct percpu_counter *fbc, s64 rhs)
145{
146 s64 count;
147
148 count = percpu_counter_read(fbc);
149 /* Check to see if rough count will be sufficient for comparison */
150 if (abs(count - rhs) > (percpu_counter_batch*num_online_cpus())) {
151 if (count > rhs)
152 return 1;
153 else
154 return -1;
155 }
156 /* Need to use precise count */
157 count = percpu_counter_sum(fbc);
158 if (count > rhs)
159 return 1;
160 else if (count < rhs)
161 return -1;
162 else
163 return 0;
164}
165EXPORT_SYMBOL(percpu_counter_compare);
166
140static int __init percpu_counter_startup(void) 167static int __init percpu_counter_startup(void)
141{ 168{
142 compute_batch_value(); 169 compute_batch_value();
diff --git a/lib/radix-tree.c b/lib/radix-tree.c
index 05da38bcc298..e907858498a6 100644
--- a/lib/radix-tree.c
+++ b/lib/radix-tree.c
@@ -609,6 +609,100 @@ int radix_tree_tag_get(struct radix_tree_root *root,
609EXPORT_SYMBOL(radix_tree_tag_get); 609EXPORT_SYMBOL(radix_tree_tag_get);
610 610
611/** 611/**
612 * radix_tree_range_tag_if_tagged - for each item in given range set given
613 * tag if item has another tag set
614 * @root: radix tree root
615 * @first_indexp: pointer to a starting index of a range to scan
616 * @last_index: last index of a range to scan
617 * @nr_to_tag: maximum number items to tag
618 * @iftag: tag index to test
619 * @settag: tag index to set if tested tag is set
620 *
621 * This function scans range of radix tree from first_index to last_index
622 * (inclusive). For each item in the range if iftag is set, the function sets
623 * also settag. The function stops either after tagging nr_to_tag items or
624 * after reaching last_index.
625 *
626 * The function returns number of leaves where the tag was set and sets
627 * *first_indexp to the first unscanned index.
628 */
629unsigned long radix_tree_range_tag_if_tagged(struct radix_tree_root *root,
630 unsigned long *first_indexp, unsigned long last_index,
631 unsigned long nr_to_tag,
632 unsigned int iftag, unsigned int settag)
633{
634 unsigned int height = root->height, shift;
635 unsigned long tagged = 0, index = *first_indexp;
636 struct radix_tree_node *open_slots[height], *slot;
637
638 last_index = min(last_index, radix_tree_maxindex(height));
639 if (index > last_index)
640 return 0;
641 if (!nr_to_tag)
642 return 0;
643 if (!root_tag_get(root, iftag)) {
644 *first_indexp = last_index + 1;
645 return 0;
646 }
647 if (height == 0) {
648 *first_indexp = last_index + 1;
649 root_tag_set(root, settag);
650 return 1;
651 }
652
653 shift = (height - 1) * RADIX_TREE_MAP_SHIFT;
654 slot = radix_tree_indirect_to_ptr(root->rnode);
655
656 for (;;) {
657 int offset;
658
659 offset = (index >> shift) & RADIX_TREE_MAP_MASK;
660 if (!slot->slots[offset])
661 goto next;
662 if (!tag_get(slot, iftag, offset))
663 goto next;
664 tag_set(slot, settag, offset);
665 if (height == 1) {
666 tagged++;
667 goto next;
668 }
669 /* Go down one level */
670 height--;
671 shift -= RADIX_TREE_MAP_SHIFT;
672 open_slots[height] = slot;
673 slot = slot->slots[offset];
674 continue;
675next:
676 /* Go to next item at level determined by 'shift' */
677 index = ((index >> shift) + 1) << shift;
678 if (index > last_index)
679 break;
680 if (tagged >= nr_to_tag)
681 break;
682 while (((index >> shift) & RADIX_TREE_MAP_MASK) == 0) {
683 /*
684 * We've fully scanned this node. Go up. Because
685 * last_index is guaranteed to be in the tree, what
686 * we do below cannot wander astray.
687 */
688 slot = open_slots[height];
689 height++;
690 shift += RADIX_TREE_MAP_SHIFT;
691 }
692 }
693 /*
694 * The iftag must have been set somewhere because otherwise
695 * we would return immediated at the beginning of the function
696 */
697 root_tag_set(root, settag);
698 *first_indexp = index;
699
700 return tagged;
701}
702EXPORT_SYMBOL(radix_tree_range_tag_if_tagged);
703
704
705/**
612 * radix_tree_next_hole - find the next hole (not-present entry) 706 * radix_tree_next_hole - find the next hole (not-present entry)
613 * @root: tree root 707 * @root: tree root
614 * @index: index key 708 * @index: index key
diff --git a/lib/raid6/Makefile b/lib/raid6/Makefile
new file mode 100644
index 000000000000..8a38102770f3
--- /dev/null
+++ b/lib/raid6/Makefile
@@ -0,0 +1,75 @@
1obj-$(CONFIG_RAID6_PQ) += raid6_pq.o
2
3raid6_pq-y += algos.o recov.o tables.o int1.o int2.o int4.o \
4 int8.o int16.o int32.o altivec1.o altivec2.o altivec4.o \
5 altivec8.o mmx.o sse1.o sse2.o
6hostprogs-y += mktables
7
8quiet_cmd_unroll = UNROLL $@
9 cmd_unroll = $(AWK) -f$(srctree)/$(src)/unroll.awk -vN=$(UNROLL) \
10 < $< > $@ || ( rm -f $@ && exit 1 )
11
12ifeq ($(CONFIG_ALTIVEC),y)
13altivec_flags := -maltivec -mabi=altivec
14endif
15
16targets += int1.c
17$(obj)/int1.c: UNROLL := 1
18$(obj)/int1.c: $(src)/int.uc $(src)/unroll.awk FORCE
19 $(call if_changed,unroll)
20
21targets += int2.c
22$(obj)/int2.c: UNROLL := 2
23$(obj)/int2.c: $(src)/int.uc $(src)/unroll.awk FORCE
24 $(call if_changed,unroll)
25
26targets += int4.c
27$(obj)/int4.c: UNROLL := 4
28$(obj)/int4.c: $(src)/int.uc $(src)/unroll.awk FORCE
29 $(call if_changed,unroll)
30
31targets += int8.c
32$(obj)/int8.c: UNROLL := 8
33$(obj)/int8.c: $(src)/int.uc $(src)/unroll.awk FORCE
34 $(call if_changed,unroll)
35
36targets += int16.c
37$(obj)/int16.c: UNROLL := 16
38$(obj)/int16.c: $(src)/int.uc $(src)/unroll.awk FORCE
39 $(call if_changed,unroll)
40
41targets += int32.c
42$(obj)/int32.c: UNROLL := 32
43$(obj)/int32.c: $(src)/int.uc $(src)/unroll.awk FORCE
44 $(call if_changed,unroll)
45
46CFLAGS_altivec1.o += $(altivec_flags)
47targets += altivec1.c
48$(obj)/altivec1.c: UNROLL := 1
49$(obj)/altivec1.c: $(src)/altivec.uc $(src)/unroll.awk FORCE
50 $(call if_changed,unroll)
51
52CFLAGS_altivec2.o += $(altivec_flags)
53targets += altivec2.c
54$(obj)/altivec2.c: UNROLL := 2
55$(obj)/altivec2.c: $(src)/altivec.uc $(src)/unroll.awk FORCE
56 $(call if_changed,unroll)
57
58CFLAGS_altivec4.o += $(altivec_flags)
59targets += altivec4.c
60$(obj)/altivec4.c: UNROLL := 4
61$(obj)/altivec4.c: $(src)/altivec.uc $(src)/unroll.awk FORCE
62 $(call if_changed,unroll)
63
64CFLAGS_altivec8.o += $(altivec_flags)
65targets += altivec8.c
66$(obj)/altivec8.c: UNROLL := 8
67$(obj)/altivec8.c: $(src)/altivec.uc $(src)/unroll.awk FORCE
68 $(call if_changed,unroll)
69
70quiet_cmd_mktable = TABLE $@
71 cmd_mktable = $(obj)/mktables > $@ || ( rm -f $@ && exit 1 )
72
73targets += tables.c
74$(obj)/tables.c: $(obj)/mktables FORCE
75 $(call if_changed,mktable)
diff --git a/lib/raid6/algos.c b/lib/raid6/algos.c
new file mode 100644
index 000000000000..b595f560bee7
--- /dev/null
+++ b/lib/raid6/algos.c
@@ -0,0 +1,154 @@
1/* -*- linux-c -*- ------------------------------------------------------- *
2 *
3 * Copyright 2002 H. Peter Anvin - All Rights Reserved
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation, Inc., 53 Temple Place Ste 330,
8 * Boston MA 02111-1307, USA; either version 2 of the License, or
9 * (at your option) any later version; incorporated herein by reference.
10 *
11 * ----------------------------------------------------------------------- */
12
13/*
14 * raid6/algos.c
15 *
16 * Algorithm list and algorithm selection for RAID-6
17 */
18
19#include <linux/raid/pq.h>
20#ifndef __KERNEL__
21#include <sys/mman.h>
22#include <stdio.h>
23#else
24#include <linux/gfp.h>
25#if !RAID6_USE_EMPTY_ZERO_PAGE
26/* In .bss so it's zeroed */
27const char raid6_empty_zero_page[PAGE_SIZE] __attribute__((aligned(256)));
28EXPORT_SYMBOL(raid6_empty_zero_page);
29#endif
30#endif
31
32struct raid6_calls raid6_call;
33EXPORT_SYMBOL_GPL(raid6_call);
34
35const struct raid6_calls * const raid6_algos[] = {
36 &raid6_intx1,
37 &raid6_intx2,
38 &raid6_intx4,
39 &raid6_intx8,
40#if defined(__ia64__)
41 &raid6_intx16,
42 &raid6_intx32,
43#endif
44#if defined(__i386__) && !defined(__arch_um__)
45 &raid6_mmxx1,
46 &raid6_mmxx2,
47 &raid6_sse1x1,
48 &raid6_sse1x2,
49 &raid6_sse2x1,
50 &raid6_sse2x2,
51#endif
52#if defined(__x86_64__) && !defined(__arch_um__)
53 &raid6_sse2x1,
54 &raid6_sse2x2,
55 &raid6_sse2x4,
56#endif
57#ifdef CONFIG_ALTIVEC
58 &raid6_altivec1,
59 &raid6_altivec2,
60 &raid6_altivec4,
61 &raid6_altivec8,
62#endif
63 NULL
64};
65
66#ifdef __KERNEL__
67#define RAID6_TIME_JIFFIES_LG2 4
68#else
69/* Need more time to be stable in userspace */
70#define RAID6_TIME_JIFFIES_LG2 9
71#define time_before(x, y) ((x) < (y))
72#endif
73
74/* Try to pick the best algorithm */
75/* This code uses the gfmul table as convenient data set to abuse */
76
77int __init raid6_select_algo(void)
78{
79 const struct raid6_calls * const * algo;
80 const struct raid6_calls * best;
81 char *syndromes;
82 void *dptrs[(65536/PAGE_SIZE)+2];
83 int i, disks;
84 unsigned long perf, bestperf;
85 int bestprefer;
86 unsigned long j0, j1;
87
88 disks = (65536/PAGE_SIZE)+2;
89 for ( i = 0 ; i < disks-2 ; i++ ) {
90 dptrs[i] = ((char *)raid6_gfmul) + PAGE_SIZE*i;
91 }
92
93 /* Normal code - use a 2-page allocation to avoid D$ conflict */
94 syndromes = (void *) __get_free_pages(GFP_KERNEL, 1);
95
96 if ( !syndromes ) {
97 printk("raid6: Yikes! No memory available.\n");
98 return -ENOMEM;
99 }
100
101 dptrs[disks-2] = syndromes;
102 dptrs[disks-1] = syndromes + PAGE_SIZE;
103
104 bestperf = 0; bestprefer = 0; best = NULL;
105
106 for ( algo = raid6_algos ; *algo ; algo++ ) {
107 if ( !(*algo)->valid || (*algo)->valid() ) {
108 perf = 0;
109
110 preempt_disable();
111 j0 = jiffies;
112 while ( (j1 = jiffies) == j0 )
113 cpu_relax();
114 while (time_before(jiffies,
115 j1 + (1<<RAID6_TIME_JIFFIES_LG2))) {
116 (*algo)->gen_syndrome(disks, PAGE_SIZE, dptrs);
117 perf++;
118 }
119 preempt_enable();
120
121 if ( (*algo)->prefer > bestprefer ||
122 ((*algo)->prefer == bestprefer &&
123 perf > bestperf) ) {
124 best = *algo;
125 bestprefer = best->prefer;
126 bestperf = perf;
127 }
128 printk("raid6: %-8s %5ld MB/s\n", (*algo)->name,
129 (perf*HZ) >> (20-16+RAID6_TIME_JIFFIES_LG2));
130 }
131 }
132
133 if (best) {
134 printk("raid6: using algorithm %s (%ld MB/s)\n",
135 best->name,
136 (bestperf*HZ) >> (20-16+RAID6_TIME_JIFFIES_LG2));
137 raid6_call = *best;
138 } else
139 printk("raid6: Yikes! No algorithm found!\n");
140
141 free_pages((unsigned long)syndromes, 1);
142
143 return best ? 0 : -EINVAL;
144}
145
146static void raid6_exit(void)
147{
148 do { } while (0);
149}
150
151subsys_initcall(raid6_select_algo);
152module_exit(raid6_exit);
153MODULE_LICENSE("GPL");
154MODULE_DESCRIPTION("RAID6 Q-syndrome calculations");
diff --git a/lib/raid6/altivec.uc b/lib/raid6/altivec.uc
new file mode 100644
index 000000000000..2654d5c854be
--- /dev/null
+++ b/lib/raid6/altivec.uc
@@ -0,0 +1,130 @@
1/* -*- linux-c -*- ------------------------------------------------------- *
2 *
3 * Copyright 2002-2004 H. Peter Anvin - All Rights Reserved
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation, Inc., 53 Temple Place Ste 330,
8 * Boston MA 02111-1307, USA; either version 2 of the License, or
9 * (at your option) any later version; incorporated herein by reference.
10 *
11 * ----------------------------------------------------------------------- */
12
13/*
14 * raid6altivec$#.c
15 *
16 * $#-way unrolled portable integer math RAID-6 instruction set
17 *
18 * This file is postprocessed using unroll.awk
19 *
20 * <benh> hpa: in process,
21 * you can just "steal" the vec unit with enable_kernel_altivec() (but
22 * bracked this with preempt_disable/enable or in a lock)
23 */
24
25#include <linux/raid/pq.h>
26
27#ifdef CONFIG_ALTIVEC
28
29#include <altivec.h>
30#ifdef __KERNEL__
31# include <asm/system.h>
32# include <asm/cputable.h>
33#endif
34
35/*
36 * This is the C data type to use. We use a vector of
37 * signed char so vec_cmpgt() will generate the right
38 * instruction.
39 */
40
41typedef vector signed char unative_t;
42
43#define NBYTES(x) ((vector signed char) {x,x,x,x, x,x,x,x, x,x,x,x, x,x,x,x})
44#define NSIZE sizeof(unative_t)
45
46/*
47 * The SHLBYTE() operation shifts each byte left by 1, *not*
48 * rolling over into the next byte
49 */
50static inline __attribute_const__ unative_t SHLBYTE(unative_t v)
51{
52 return vec_add(v,v);
53}
54
55/*
56 * The MASK() operation returns 0xFF in any byte for which the high
57 * bit is 1, 0x00 for any byte for which the high bit is 0.
58 */
59static inline __attribute_const__ unative_t MASK(unative_t v)
60{
61 unative_t zv = NBYTES(0);
62
63 /* vec_cmpgt returns a vector bool char; thus the need for the cast */
64 return (unative_t)vec_cmpgt(zv, v);
65}
66
67
68/* This is noinline to make damned sure that gcc doesn't move any of the
69 Altivec code around the enable/disable code */
70static void noinline
71raid6_altivec$#_gen_syndrome_real(int disks, size_t bytes, void **ptrs)
72{
73 u8 **dptr = (u8 **)ptrs;
74 u8 *p, *q;
75 int d, z, z0;
76
77 unative_t wd$$, wq$$, wp$$, w1$$, w2$$;
78 unative_t x1d = NBYTES(0x1d);
79
80 z0 = disks - 3; /* Highest data disk */
81 p = dptr[z0+1]; /* XOR parity */
82 q = dptr[z0+2]; /* RS syndrome */
83
84 for ( d = 0 ; d < bytes ; d += NSIZE*$# ) {
85 wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE];
86 for ( z = z0-1 ; z >= 0 ; z-- ) {
87 wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE];
88 wp$$ = vec_xor(wp$$, wd$$);
89 w2$$ = MASK(wq$$);
90 w1$$ = SHLBYTE(wq$$);
91 w2$$ = vec_and(w2$$, x1d);
92 w1$$ = vec_xor(w1$$, w2$$);
93 wq$$ = vec_xor(w1$$, wd$$);
94 }
95 *(unative_t *)&p[d+NSIZE*$$] = wp$$;
96 *(unative_t *)&q[d+NSIZE*$$] = wq$$;
97 }
98}
99
100static void raid6_altivec$#_gen_syndrome(int disks, size_t bytes, void **ptrs)
101{
102 preempt_disable();
103 enable_kernel_altivec();
104
105 raid6_altivec$#_gen_syndrome_real(disks, bytes, ptrs);
106
107 preempt_enable();
108}
109
110int raid6_have_altivec(void);
111#if $# == 1
112int raid6_have_altivec(void)
113{
114 /* This assumes either all CPUs have Altivec or none does */
115# ifdef __KERNEL__
116 return cpu_has_feature(CPU_FTR_ALTIVEC);
117# else
118 return 1;
119# endif
120}
121#endif
122
123const struct raid6_calls raid6_altivec$# = {
124 raid6_altivec$#_gen_syndrome,
125 raid6_have_altivec,
126 "altivecx$#",
127 0
128};
129
130#endif /* CONFIG_ALTIVEC */
diff --git a/lib/raid6/int.uc b/lib/raid6/int.uc
new file mode 100644
index 000000000000..d1e276a14fab
--- /dev/null
+++ b/lib/raid6/int.uc
@@ -0,0 +1,117 @@
1/* -*- linux-c -*- ------------------------------------------------------- *
2 *
3 * Copyright 2002-2004 H. Peter Anvin - All Rights Reserved
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation, Inc., 53 Temple Place Ste 330,
8 * Boston MA 02111-1307, USA; either version 2 of the License, or
9 * (at your option) any later version; incorporated herein by reference.
10 *
11 * ----------------------------------------------------------------------- */
12
13/*
14 * raid6int$#.c
15 *
16 * $#-way unrolled portable integer math RAID-6 instruction set
17 *
18 * This file is postprocessed using unroll.awk
19 */
20
21#include <linux/raid/pq.h>
22
23/*
24 * This is the C data type to use
25 */
26
27/* Change this from BITS_PER_LONG if there is something better... */
28#if BITS_PER_LONG == 64
29# define NBYTES(x) ((x) * 0x0101010101010101UL)
30# define NSIZE 8
31# define NSHIFT 3
32# define NSTRING "64"
33typedef u64 unative_t;
34#else
35# define NBYTES(x) ((x) * 0x01010101U)
36# define NSIZE 4
37# define NSHIFT 2
38# define NSTRING "32"
39typedef u32 unative_t;
40#endif
41
42
43
44/*
45 * IA-64 wants insane amounts of unrolling. On other architectures that
46 * is just a waste of space.
47 */
48#if ($# <= 8) || defined(__ia64__)
49
50
51/*
52 * These sub-operations are separate inlines since they can sometimes be
53 * specially optimized using architecture-specific hacks.
54 */
55
56/*
57 * The SHLBYTE() operation shifts each byte left by 1, *not*
58 * rolling over into the next byte
59 */
60static inline __attribute_const__ unative_t SHLBYTE(unative_t v)
61{
62 unative_t vv;
63
64 vv = (v << 1) & NBYTES(0xfe);
65 return vv;
66}
67
68/*
69 * The MASK() operation returns 0xFF in any byte for which the high
70 * bit is 1, 0x00 for any byte for which the high bit is 0.
71 */
72static inline __attribute_const__ unative_t MASK(unative_t v)
73{
74 unative_t vv;
75
76 vv = v & NBYTES(0x80);
77 vv = (vv << 1) - (vv >> 7); /* Overflow on the top bit is OK */
78 return vv;
79}
80
81
82static void raid6_int$#_gen_syndrome(int disks, size_t bytes, void **ptrs)
83{
84 u8 **dptr = (u8 **)ptrs;
85 u8 *p, *q;
86 int d, z, z0;
87
88 unative_t wd$$, wq$$, wp$$, w1$$, w2$$;
89
90 z0 = disks - 3; /* Highest data disk */
91 p = dptr[z0+1]; /* XOR parity */
92 q = dptr[z0+2]; /* RS syndrome */
93
94 for ( d = 0 ; d < bytes ; d += NSIZE*$# ) {
95 wq$$ = wp$$ = *(unative_t *)&dptr[z0][d+$$*NSIZE];
96 for ( z = z0-1 ; z >= 0 ; z-- ) {
97 wd$$ = *(unative_t *)&dptr[z][d+$$*NSIZE];
98 wp$$ ^= wd$$;
99 w2$$ = MASK(wq$$);
100 w1$$ = SHLBYTE(wq$$);
101 w2$$ &= NBYTES(0x1d);
102 w1$$ ^= w2$$;
103 wq$$ = w1$$ ^ wd$$;
104 }
105 *(unative_t *)&p[d+NSIZE*$$] = wp$$;
106 *(unative_t *)&q[d+NSIZE*$$] = wq$$;
107 }
108}
109
110const struct raid6_calls raid6_intx$# = {
111 raid6_int$#_gen_syndrome,
112 NULL, /* always valid */
113 "int" NSTRING "x$#",
114 0
115};
116
117#endif
diff --git a/lib/raid6/mktables.c b/lib/raid6/mktables.c
new file mode 100644
index 000000000000..3b1500843bba
--- /dev/null
+++ b/lib/raid6/mktables.c
@@ -0,0 +1,132 @@
1/* -*- linux-c -*- ------------------------------------------------------- *
2 *
3 * Copyright 2002-2007 H. Peter Anvin - All Rights Reserved
4 *
5 * This file is part of the Linux kernel, and is made available under
6 * the terms of the GNU General Public License version 2 or (at your
7 * option) any later version; incorporated herein by reference.
8 *
9 * ----------------------------------------------------------------------- */
10
11/*
12 * mktables.c
13 *
14 * Make RAID-6 tables. This is a host user space program to be run at
15 * compile time.
16 */
17
18#include <stdio.h>
19#include <string.h>
20#include <inttypes.h>
21#include <stdlib.h>
22#include <time.h>
23
24static uint8_t gfmul(uint8_t a, uint8_t b)
25{
26 uint8_t v = 0;
27
28 while (b) {
29 if (b & 1)
30 v ^= a;
31 a = (a << 1) ^ (a & 0x80 ? 0x1d : 0);
32 b >>= 1;
33 }
34
35 return v;
36}
37
38static uint8_t gfpow(uint8_t a, int b)
39{
40 uint8_t v = 1;
41
42 b %= 255;
43 if (b < 0)
44 b += 255;
45
46 while (b) {
47 if (b & 1)
48 v = gfmul(v, a);
49 a = gfmul(a, a);
50 b >>= 1;
51 }
52
53 return v;
54}
55
56int main(int argc, char *argv[])
57{
58 int i, j, k;
59 uint8_t v;
60 uint8_t exptbl[256], invtbl[256];
61
62 printf("#include <linux/raid/pq.h>\n");
63
64 /* Compute multiplication table */
65 printf("\nconst u8 __attribute__((aligned(256)))\n"
66 "raid6_gfmul[256][256] =\n"
67 "{\n");
68 for (i = 0; i < 256; i++) {
69 printf("\t{\n");
70 for (j = 0; j < 256; j += 8) {
71 printf("\t\t");
72 for (k = 0; k < 8; k++)
73 printf("0x%02x,%c", gfmul(i, j + k),
74 (k == 7) ? '\n' : ' ');
75 }
76 printf("\t},\n");
77 }
78 printf("};\n");
79 printf("#ifdef __KERNEL__\n");
80 printf("EXPORT_SYMBOL(raid6_gfmul);\n");
81 printf("#endif\n");
82
83 /* Compute power-of-2 table (exponent) */
84 v = 1;
85 printf("\nconst u8 __attribute__((aligned(256)))\n"
86 "raid6_gfexp[256] =\n" "{\n");
87 for (i = 0; i < 256; i += 8) {
88 printf("\t");
89 for (j = 0; j < 8; j++) {
90 exptbl[i + j] = v;
91 printf("0x%02x,%c", v, (j == 7) ? '\n' : ' ');
92 v = gfmul(v, 2);
93 if (v == 1)
94 v = 0; /* For entry 255, not a real entry */
95 }
96 }
97 printf("};\n");
98 printf("#ifdef __KERNEL__\n");
99 printf("EXPORT_SYMBOL(raid6_gfexp);\n");
100 printf("#endif\n");
101
102 /* Compute inverse table x^-1 == x^254 */
103 printf("\nconst u8 __attribute__((aligned(256)))\n"
104 "raid6_gfinv[256] =\n" "{\n");
105 for (i = 0; i < 256; i += 8) {
106 printf("\t");
107 for (j = 0; j < 8; j++) {
108 invtbl[i + j] = v = gfpow(i + j, 254);
109 printf("0x%02x,%c", v, (j == 7) ? '\n' : ' ');
110 }
111 }
112 printf("};\n");
113 printf("#ifdef __KERNEL__\n");
114 printf("EXPORT_SYMBOL(raid6_gfinv);\n");
115 printf("#endif\n");
116
117 /* Compute inv(2^x + 1) (exponent-xor-inverse) table */
118 printf("\nconst u8 __attribute__((aligned(256)))\n"
119 "raid6_gfexi[256] =\n" "{\n");
120 for (i = 0; i < 256; i += 8) {
121 printf("\t");
122 for (j = 0; j < 8; j++)
123 printf("0x%02x,%c", invtbl[exptbl[i + j] ^ 1],
124 (j == 7) ? '\n' : ' ');
125 }
126 printf("};\n");
127 printf("#ifdef __KERNEL__\n");
128 printf("EXPORT_SYMBOL(raid6_gfexi);\n");
129 printf("#endif\n");
130
131 return 0;
132}
diff --git a/lib/raid6/mmx.c b/lib/raid6/mmx.c
new file mode 100644
index 000000000000..279347f23094
--- /dev/null
+++ b/lib/raid6/mmx.c
@@ -0,0 +1,142 @@
1/* -*- linux-c -*- ------------------------------------------------------- *
2 *
3 * Copyright 2002 H. Peter Anvin - All Rights Reserved
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation, Inc., 53 Temple Place Ste 330,
8 * Boston MA 02111-1307, USA; either version 2 of the License, or
9 * (at your option) any later version; incorporated herein by reference.
10 *
11 * ----------------------------------------------------------------------- */
12
13/*
14 * raid6/mmx.c
15 *
16 * MMX implementation of RAID-6 syndrome functions
17 */
18
19#if defined(__i386__) && !defined(__arch_um__)
20
21#include <linux/raid/pq.h>
22#include "x86.h"
23
24/* Shared with raid6/sse1.c */
25const struct raid6_mmx_constants {
26 u64 x1d;
27} raid6_mmx_constants = {
28 0x1d1d1d1d1d1d1d1dULL,
29};
30
31static int raid6_have_mmx(void)
32{
33 /* Not really "boot_cpu" but "all_cpus" */
34 return boot_cpu_has(X86_FEATURE_MMX);
35}
36
37/*
38 * Plain MMX implementation
39 */
40static void raid6_mmx1_gen_syndrome(int disks, size_t bytes, void **ptrs)
41{
42 u8 **dptr = (u8 **)ptrs;
43 u8 *p, *q;
44 int d, z, z0;
45
46 z0 = disks - 3; /* Highest data disk */
47 p = dptr[z0+1]; /* XOR parity */
48 q = dptr[z0+2]; /* RS syndrome */
49
50 kernel_fpu_begin();
51
52 asm volatile("movq %0,%%mm0" : : "m" (raid6_mmx_constants.x1d));
53 asm volatile("pxor %mm5,%mm5"); /* Zero temp */
54
55 for ( d = 0 ; d < bytes ; d += 8 ) {
56 asm volatile("movq %0,%%mm2" : : "m" (dptr[z0][d])); /* P[0] */
57 asm volatile("movq %mm2,%mm4"); /* Q[0] */
58 for ( z = z0-1 ; z >= 0 ; z-- ) {
59 asm volatile("movq %0,%%mm6" : : "m" (dptr[z][d]));
60 asm volatile("pcmpgtb %mm4,%mm5");
61 asm volatile("paddb %mm4,%mm4");
62 asm volatile("pand %mm0,%mm5");
63 asm volatile("pxor %mm5,%mm4");
64 asm volatile("pxor %mm5,%mm5");
65 asm volatile("pxor %mm6,%mm2");
66 asm volatile("pxor %mm6,%mm4");
67 }
68 asm volatile("movq %%mm2,%0" : "=m" (p[d]));
69 asm volatile("pxor %mm2,%mm2");
70 asm volatile("movq %%mm4,%0" : "=m" (q[d]));
71 asm volatile("pxor %mm4,%mm4");
72 }
73
74 kernel_fpu_end();
75}
76
77const struct raid6_calls raid6_mmxx1 = {
78 raid6_mmx1_gen_syndrome,
79 raid6_have_mmx,
80 "mmxx1",
81 0
82};
83
84/*
85 * Unrolled-by-2 MMX implementation
86 */
87static void raid6_mmx2_gen_syndrome(int disks, size_t bytes, void **ptrs)
88{
89 u8 **dptr = (u8 **)ptrs;
90 u8 *p, *q;
91 int d, z, z0;
92
93 z0 = disks - 3; /* Highest data disk */
94 p = dptr[z0+1]; /* XOR parity */
95 q = dptr[z0+2]; /* RS syndrome */
96
97 kernel_fpu_begin();
98
99 asm volatile("movq %0,%%mm0" : : "m" (raid6_mmx_constants.x1d));
100 asm volatile("pxor %mm5,%mm5"); /* Zero temp */
101 asm volatile("pxor %mm7,%mm7"); /* Zero temp */
102
103 for ( d = 0 ; d < bytes ; d += 16 ) {
104 asm volatile("movq %0,%%mm2" : : "m" (dptr[z0][d])); /* P[0] */
105 asm volatile("movq %0,%%mm3" : : "m" (dptr[z0][d+8]));
106 asm volatile("movq %mm2,%mm4"); /* Q[0] */
107 asm volatile("movq %mm3,%mm6"); /* Q[1] */
108 for ( z = z0-1 ; z >= 0 ; z-- ) {
109 asm volatile("pcmpgtb %mm4,%mm5");
110 asm volatile("pcmpgtb %mm6,%mm7");
111 asm volatile("paddb %mm4,%mm4");
112 asm volatile("paddb %mm6,%mm6");
113 asm volatile("pand %mm0,%mm5");
114 asm volatile("pand %mm0,%mm7");
115 asm volatile("pxor %mm5,%mm4");
116 asm volatile("pxor %mm7,%mm6");
117 asm volatile("movq %0,%%mm5" : : "m" (dptr[z][d]));
118 asm volatile("movq %0,%%mm7" : : "m" (dptr[z][d+8]));
119 asm volatile("pxor %mm5,%mm2");
120 asm volatile("pxor %mm7,%mm3");
121 asm volatile("pxor %mm5,%mm4");
122 asm volatile("pxor %mm7,%mm6");
123 asm volatile("pxor %mm5,%mm5");
124 asm volatile("pxor %mm7,%mm7");
125 }
126 asm volatile("movq %%mm2,%0" : "=m" (p[d]));
127 asm volatile("movq %%mm3,%0" : "=m" (p[d+8]));
128 asm volatile("movq %%mm4,%0" : "=m" (q[d]));
129 asm volatile("movq %%mm6,%0" : "=m" (q[d+8]));
130 }
131
132 kernel_fpu_end();
133}
134
135const struct raid6_calls raid6_mmxx2 = {
136 raid6_mmx2_gen_syndrome,
137 raid6_have_mmx,
138 "mmxx2",
139 0
140};
141
142#endif
diff --git a/lib/raid6/recov.c b/lib/raid6/recov.c
new file mode 100644
index 000000000000..8590d19cf522
--- /dev/null
+++ b/lib/raid6/recov.c
@@ -0,0 +1,132 @@
1/* -*- linux-c -*- ------------------------------------------------------- *
2 *
3 * Copyright 2002 H. Peter Anvin - All Rights Reserved
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation, Inc., 53 Temple Place Ste 330,
8 * Boston MA 02111-1307, USA; either version 2 of the License, or
9 * (at your option) any later version; incorporated herein by reference.
10 *
11 * ----------------------------------------------------------------------- */
12
13/*
14 * raid6/recov.c
15 *
16 * RAID-6 data recovery in dual failure mode. In single failure mode,
17 * use the RAID-5 algorithm (or, in the case of Q failure, just reconstruct
18 * the syndrome.)
19 */
20
21#include <linux/raid/pq.h>
22
23/* Recover two failed data blocks. */
24void raid6_2data_recov(int disks, size_t bytes, int faila, int failb,
25 void **ptrs)
26{
27 u8 *p, *q, *dp, *dq;
28 u8 px, qx, db;
29 const u8 *pbmul; /* P multiplier table for B data */
30 const u8 *qmul; /* Q multiplier table (for both) */
31
32 p = (u8 *)ptrs[disks-2];
33 q = (u8 *)ptrs[disks-1];
34
35 /* Compute syndrome with zero for the missing data pages
36 Use the dead data pages as temporary storage for
37 delta p and delta q */
38 dp = (u8 *)ptrs[faila];
39 ptrs[faila] = (void *)raid6_empty_zero_page;
40 ptrs[disks-2] = dp;
41 dq = (u8 *)ptrs[failb];
42 ptrs[failb] = (void *)raid6_empty_zero_page;
43 ptrs[disks-1] = dq;
44
45 raid6_call.gen_syndrome(disks, bytes, ptrs);
46
47 /* Restore pointer table */
48 ptrs[faila] = dp;
49 ptrs[failb] = dq;
50 ptrs[disks-2] = p;
51 ptrs[disks-1] = q;
52
53 /* Now, pick the proper data tables */
54 pbmul = raid6_gfmul[raid6_gfexi[failb-faila]];
55 qmul = raid6_gfmul[raid6_gfinv[raid6_gfexp[faila]^raid6_gfexp[failb]]];
56
57 /* Now do it... */
58 while ( bytes-- ) {
59 px = *p ^ *dp;
60 qx = qmul[*q ^ *dq];
61 *dq++ = db = pbmul[px] ^ qx; /* Reconstructed B */
62 *dp++ = db ^ px; /* Reconstructed A */
63 p++; q++;
64 }
65}
66EXPORT_SYMBOL_GPL(raid6_2data_recov);
67
68/* Recover failure of one data block plus the P block */
69void raid6_datap_recov(int disks, size_t bytes, int faila, void **ptrs)
70{
71 u8 *p, *q, *dq;
72 const u8 *qmul; /* Q multiplier table */
73
74 p = (u8 *)ptrs[disks-2];
75 q = (u8 *)ptrs[disks-1];
76
77 /* Compute syndrome with zero for the missing data page
78 Use the dead data page as temporary storage for delta q */
79 dq = (u8 *)ptrs[faila];
80 ptrs[faila] = (void *)raid6_empty_zero_page;
81 ptrs[disks-1] = dq;
82
83 raid6_call.gen_syndrome(disks, bytes, ptrs);
84
85 /* Restore pointer table */
86 ptrs[faila] = dq;
87 ptrs[disks-1] = q;
88
89 /* Now, pick the proper data tables */
90 qmul = raid6_gfmul[raid6_gfinv[raid6_gfexp[faila]]];
91
92 /* Now do it... */
93 while ( bytes-- ) {
94 *p++ ^= *dq = qmul[*q ^ *dq];
95 q++; dq++;
96 }
97}
98EXPORT_SYMBOL_GPL(raid6_datap_recov);
99
100#ifndef __KERNEL__
101/* Testing only */
102
103/* Recover two failed blocks. */
104void raid6_dual_recov(int disks, size_t bytes, int faila, int failb, void **ptrs)
105{
106 if ( faila > failb ) {
107 int tmp = faila;
108 faila = failb;
109 failb = tmp;
110 }
111
112 if ( failb == disks-1 ) {
113 if ( faila == disks-2 ) {
114 /* P+Q failure. Just rebuild the syndrome. */
115 raid6_call.gen_syndrome(disks, bytes, ptrs);
116 } else {
117 /* data+Q failure. Reconstruct data from P,
118 then rebuild syndrome. */
119 /* NOT IMPLEMENTED - equivalent to RAID-5 */
120 }
121 } else {
122 if ( failb == disks-2 ) {
123 /* data+P failure. */
124 raid6_datap_recov(disks, bytes, faila, ptrs);
125 } else {
126 /* data+data failure. */
127 raid6_2data_recov(disks, bytes, faila, failb, ptrs);
128 }
129 }
130}
131
132#endif
diff --git a/lib/raid6/sse1.c b/lib/raid6/sse1.c
new file mode 100644
index 000000000000..10dd91948c07
--- /dev/null
+++ b/lib/raid6/sse1.c
@@ -0,0 +1,162 @@
1/* -*- linux-c -*- ------------------------------------------------------- *
2 *
3 * Copyright 2002 H. Peter Anvin - All Rights Reserved
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation, Inc., 53 Temple Place Ste 330,
8 * Boston MA 02111-1307, USA; either version 2 of the License, or
9 * (at your option) any later version; incorporated herein by reference.
10 *
11 * ----------------------------------------------------------------------- */
12
13/*
14 * raid6/sse1.c
15 *
16 * SSE-1/MMXEXT implementation of RAID-6 syndrome functions
17 *
18 * This is really an MMX implementation, but it requires SSE-1 or
19 * AMD MMXEXT for prefetch support and a few other features. The
20 * support for nontemporal memory accesses is enough to make this
21 * worthwhile as a separate implementation.
22 */
23
24#if defined(__i386__) && !defined(__arch_um__)
25
26#include <linux/raid/pq.h>
27#include "x86.h"
28
29/* Defined in raid6/mmx.c */
30extern const struct raid6_mmx_constants {
31 u64 x1d;
32} raid6_mmx_constants;
33
34static int raid6_have_sse1_or_mmxext(void)
35{
36 /* Not really boot_cpu but "all_cpus" */
37 return boot_cpu_has(X86_FEATURE_MMX) &&
38 (boot_cpu_has(X86_FEATURE_XMM) ||
39 boot_cpu_has(X86_FEATURE_MMXEXT));
40}
41
42/*
43 * Plain SSE1 implementation
44 */
45static void raid6_sse11_gen_syndrome(int disks, size_t bytes, void **ptrs)
46{
47 u8 **dptr = (u8 **)ptrs;
48 u8 *p, *q;
49 int d, z, z0;
50
51 z0 = disks - 3; /* Highest data disk */
52 p = dptr[z0+1]; /* XOR parity */
53 q = dptr[z0+2]; /* RS syndrome */
54
55 kernel_fpu_begin();
56
57 asm volatile("movq %0,%%mm0" : : "m" (raid6_mmx_constants.x1d));
58 asm volatile("pxor %mm5,%mm5"); /* Zero temp */
59
60 for ( d = 0 ; d < bytes ; d += 8 ) {
61 asm volatile("prefetchnta %0" : : "m" (dptr[z0][d]));
62 asm volatile("movq %0,%%mm2" : : "m" (dptr[z0][d])); /* P[0] */
63 asm volatile("prefetchnta %0" : : "m" (dptr[z0-1][d]));
64 asm volatile("movq %mm2,%mm4"); /* Q[0] */
65 asm volatile("movq %0,%%mm6" : : "m" (dptr[z0-1][d]));
66 for ( z = z0-2 ; z >= 0 ; z-- ) {
67 asm volatile("prefetchnta %0" : : "m" (dptr[z][d]));
68 asm volatile("pcmpgtb %mm4,%mm5");
69 asm volatile("paddb %mm4,%mm4");
70 asm volatile("pand %mm0,%mm5");
71 asm volatile("pxor %mm5,%mm4");
72 asm volatile("pxor %mm5,%mm5");
73 asm volatile("pxor %mm6,%mm2");
74 asm volatile("pxor %mm6,%mm4");
75 asm volatile("movq %0,%%mm6" : : "m" (dptr[z][d]));
76 }
77 asm volatile("pcmpgtb %mm4,%mm5");
78 asm volatile("paddb %mm4,%mm4");
79 asm volatile("pand %mm0,%mm5");
80 asm volatile("pxor %mm5,%mm4");
81 asm volatile("pxor %mm5,%mm5");
82 asm volatile("pxor %mm6,%mm2");
83 asm volatile("pxor %mm6,%mm4");
84
85 asm volatile("movntq %%mm2,%0" : "=m" (p[d]));
86 asm volatile("movntq %%mm4,%0" : "=m" (q[d]));
87 }
88
89 asm volatile("sfence" : : : "memory");
90 kernel_fpu_end();
91}
92
93const struct raid6_calls raid6_sse1x1 = {
94 raid6_sse11_gen_syndrome,
95 raid6_have_sse1_or_mmxext,
96 "sse1x1",
97 1 /* Has cache hints */
98};
99
100/*
101 * Unrolled-by-2 SSE1 implementation
102 */
103static void raid6_sse12_gen_syndrome(int disks, size_t bytes, void **ptrs)
104{
105 u8 **dptr = (u8 **)ptrs;
106 u8 *p, *q;
107 int d, z, z0;
108
109 z0 = disks - 3; /* Highest data disk */
110 p = dptr[z0+1]; /* XOR parity */
111 q = dptr[z0+2]; /* RS syndrome */
112
113 kernel_fpu_begin();
114
115 asm volatile("movq %0,%%mm0" : : "m" (raid6_mmx_constants.x1d));
116 asm volatile("pxor %mm5,%mm5"); /* Zero temp */
117 asm volatile("pxor %mm7,%mm7"); /* Zero temp */
118
119 /* We uniformly assume a single prefetch covers at least 16 bytes */
120 for ( d = 0 ; d < bytes ; d += 16 ) {
121 asm volatile("prefetchnta %0" : : "m" (dptr[z0][d]));
122 asm volatile("movq %0,%%mm2" : : "m" (dptr[z0][d])); /* P[0] */
123 asm volatile("movq %0,%%mm3" : : "m" (dptr[z0][d+8])); /* P[1] */
124 asm volatile("movq %mm2,%mm4"); /* Q[0] */
125 asm volatile("movq %mm3,%mm6"); /* Q[1] */
126 for ( z = z0-1 ; z >= 0 ; z-- ) {
127 asm volatile("prefetchnta %0" : : "m" (dptr[z][d]));
128 asm volatile("pcmpgtb %mm4,%mm5");
129 asm volatile("pcmpgtb %mm6,%mm7");
130 asm volatile("paddb %mm4,%mm4");
131 asm volatile("paddb %mm6,%mm6");
132 asm volatile("pand %mm0,%mm5");
133 asm volatile("pand %mm0,%mm7");
134 asm volatile("pxor %mm5,%mm4");
135 asm volatile("pxor %mm7,%mm6");
136 asm volatile("movq %0,%%mm5" : : "m" (dptr[z][d]));
137 asm volatile("movq %0,%%mm7" : : "m" (dptr[z][d+8]));
138 asm volatile("pxor %mm5,%mm2");
139 asm volatile("pxor %mm7,%mm3");
140 asm volatile("pxor %mm5,%mm4");
141 asm volatile("pxor %mm7,%mm6");
142 asm volatile("pxor %mm5,%mm5");
143 asm volatile("pxor %mm7,%mm7");
144 }
145 asm volatile("movntq %%mm2,%0" : "=m" (p[d]));
146 asm volatile("movntq %%mm3,%0" : "=m" (p[d+8]));
147 asm volatile("movntq %%mm4,%0" : "=m" (q[d]));
148 asm volatile("movntq %%mm6,%0" : "=m" (q[d+8]));
149 }
150
151 asm volatile("sfence" : :: "memory");
152 kernel_fpu_end();
153}
154
155const struct raid6_calls raid6_sse1x2 = {
156 raid6_sse12_gen_syndrome,
157 raid6_have_sse1_or_mmxext,
158 "sse1x2",
159 1 /* Has cache hints */
160};
161
162#endif
diff --git a/lib/raid6/sse2.c b/lib/raid6/sse2.c
new file mode 100644
index 000000000000..bc2d57daa589
--- /dev/null
+++ b/lib/raid6/sse2.c
@@ -0,0 +1,262 @@
1/* -*- linux-c -*- ------------------------------------------------------- *
2 *
3 * Copyright 2002 H. Peter Anvin - All Rights Reserved
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation, Inc., 53 Temple Place Ste 330,
8 * Boston MA 02111-1307, USA; either version 2 of the License, or
9 * (at your option) any later version; incorporated herein by reference.
10 *
11 * ----------------------------------------------------------------------- */
12
13/*
14 * raid6/sse2.c
15 *
16 * SSE-2 implementation of RAID-6 syndrome functions
17 *
18 */
19
20#if (defined(__i386__) || defined(__x86_64__)) && !defined(__arch_um__)
21
22#include <linux/raid/pq.h>
23#include "x86.h"
24
25static const struct raid6_sse_constants {
26 u64 x1d[2];
27} raid6_sse_constants __attribute__((aligned(16))) = {
28 { 0x1d1d1d1d1d1d1d1dULL, 0x1d1d1d1d1d1d1d1dULL },
29};
30
31static int raid6_have_sse2(void)
32{
33 /* Not really boot_cpu but "all_cpus" */
34 return boot_cpu_has(X86_FEATURE_MMX) &&
35 boot_cpu_has(X86_FEATURE_FXSR) &&
36 boot_cpu_has(X86_FEATURE_XMM) &&
37 boot_cpu_has(X86_FEATURE_XMM2);
38}
39
40/*
41 * Plain SSE2 implementation
42 */
43static void raid6_sse21_gen_syndrome(int disks, size_t bytes, void **ptrs)
44{
45 u8 **dptr = (u8 **)ptrs;
46 u8 *p, *q;
47 int d, z, z0;
48
49 z0 = disks - 3; /* Highest data disk */
50 p = dptr[z0+1]; /* XOR parity */
51 q = dptr[z0+2]; /* RS syndrome */
52
53 kernel_fpu_begin();
54
55 asm volatile("movdqa %0,%%xmm0" : : "m" (raid6_sse_constants.x1d[0]));
56 asm volatile("pxor %xmm5,%xmm5"); /* Zero temp */
57
58 for ( d = 0 ; d < bytes ; d += 16 ) {
59 asm volatile("prefetchnta %0" : : "m" (dptr[z0][d]));
60 asm volatile("movdqa %0,%%xmm2" : : "m" (dptr[z0][d])); /* P[0] */
61 asm volatile("prefetchnta %0" : : "m" (dptr[z0-1][d]));
62 asm volatile("movdqa %xmm2,%xmm4"); /* Q[0] */
63 asm volatile("movdqa %0,%%xmm6" : : "m" (dptr[z0-1][d]));
64 for ( z = z0-2 ; z >= 0 ; z-- ) {
65 asm volatile("prefetchnta %0" : : "m" (dptr[z][d]));
66 asm volatile("pcmpgtb %xmm4,%xmm5");
67 asm volatile("paddb %xmm4,%xmm4");
68 asm volatile("pand %xmm0,%xmm5");
69 asm volatile("pxor %xmm5,%xmm4");
70 asm volatile("pxor %xmm5,%xmm5");
71 asm volatile("pxor %xmm6,%xmm2");
72 asm volatile("pxor %xmm6,%xmm4");
73 asm volatile("movdqa %0,%%xmm6" : : "m" (dptr[z][d]));
74 }
75 asm volatile("pcmpgtb %xmm4,%xmm5");
76 asm volatile("paddb %xmm4,%xmm4");
77 asm volatile("pand %xmm0,%xmm5");
78 asm volatile("pxor %xmm5,%xmm4");
79 asm volatile("pxor %xmm5,%xmm5");
80 asm volatile("pxor %xmm6,%xmm2");
81 asm volatile("pxor %xmm6,%xmm4");
82
83 asm volatile("movntdq %%xmm2,%0" : "=m" (p[d]));
84 asm volatile("pxor %xmm2,%xmm2");
85 asm volatile("movntdq %%xmm4,%0" : "=m" (q[d]));
86 asm volatile("pxor %xmm4,%xmm4");
87 }
88
89 asm volatile("sfence" : : : "memory");
90 kernel_fpu_end();
91}
92
93const struct raid6_calls raid6_sse2x1 = {
94 raid6_sse21_gen_syndrome,
95 raid6_have_sse2,
96 "sse2x1",
97 1 /* Has cache hints */
98};
99
100/*
101 * Unrolled-by-2 SSE2 implementation
102 */
103static void raid6_sse22_gen_syndrome(int disks, size_t bytes, void **ptrs)
104{
105 u8 **dptr = (u8 **)ptrs;
106 u8 *p, *q;
107 int d, z, z0;
108
109 z0 = disks - 3; /* Highest data disk */
110 p = dptr[z0+1]; /* XOR parity */
111 q = dptr[z0+2]; /* RS syndrome */
112
113 kernel_fpu_begin();
114
115 asm volatile("movdqa %0,%%xmm0" : : "m" (raid6_sse_constants.x1d[0]));
116 asm volatile("pxor %xmm5,%xmm5"); /* Zero temp */
117 asm volatile("pxor %xmm7,%xmm7"); /* Zero temp */
118
119 /* We uniformly assume a single prefetch covers at least 32 bytes */
120 for ( d = 0 ; d < bytes ; d += 32 ) {
121 asm volatile("prefetchnta %0" : : "m" (dptr[z0][d]));
122 asm volatile("movdqa %0,%%xmm2" : : "m" (dptr[z0][d])); /* P[0] */
123 asm volatile("movdqa %0,%%xmm3" : : "m" (dptr[z0][d+16])); /* P[1] */
124 asm volatile("movdqa %xmm2,%xmm4"); /* Q[0] */
125 asm volatile("movdqa %xmm3,%xmm6"); /* Q[1] */
126 for ( z = z0-1 ; z >= 0 ; z-- ) {
127 asm volatile("prefetchnta %0" : : "m" (dptr[z][d]));
128 asm volatile("pcmpgtb %xmm4,%xmm5");
129 asm volatile("pcmpgtb %xmm6,%xmm7");
130 asm volatile("paddb %xmm4,%xmm4");
131 asm volatile("paddb %xmm6,%xmm6");
132 asm volatile("pand %xmm0,%xmm5");
133 asm volatile("pand %xmm0,%xmm7");
134 asm volatile("pxor %xmm5,%xmm4");
135 asm volatile("pxor %xmm7,%xmm6");
136 asm volatile("movdqa %0,%%xmm5" : : "m" (dptr[z][d]));
137 asm volatile("movdqa %0,%%xmm7" : : "m" (dptr[z][d+16]));
138 asm volatile("pxor %xmm5,%xmm2");
139 asm volatile("pxor %xmm7,%xmm3");
140 asm volatile("pxor %xmm5,%xmm4");
141 asm volatile("pxor %xmm7,%xmm6");
142 asm volatile("pxor %xmm5,%xmm5");
143 asm volatile("pxor %xmm7,%xmm7");
144 }
145 asm volatile("movntdq %%xmm2,%0" : "=m" (p[d]));
146 asm volatile("movntdq %%xmm3,%0" : "=m" (p[d+16]));
147 asm volatile("movntdq %%xmm4,%0" : "=m" (q[d]));
148 asm volatile("movntdq %%xmm6,%0" : "=m" (q[d+16]));
149 }
150
151 asm volatile("sfence" : : : "memory");
152 kernel_fpu_end();
153}
154
155const struct raid6_calls raid6_sse2x2 = {
156 raid6_sse22_gen_syndrome,
157 raid6_have_sse2,
158 "sse2x2",
159 1 /* Has cache hints */
160};
161
162#endif
163
164#if defined(__x86_64__) && !defined(__arch_um__)
165
166/*
167 * Unrolled-by-4 SSE2 implementation
168 */
169static void raid6_sse24_gen_syndrome(int disks, size_t bytes, void **ptrs)
170{
171 u8 **dptr = (u8 **)ptrs;
172 u8 *p, *q;
173 int d, z, z0;
174
175 z0 = disks - 3; /* Highest data disk */
176 p = dptr[z0+1]; /* XOR parity */
177 q = dptr[z0+2]; /* RS syndrome */
178
179 kernel_fpu_begin();
180
181 asm volatile("movdqa %0,%%xmm0" :: "m" (raid6_sse_constants.x1d[0]));
182 asm volatile("pxor %xmm2,%xmm2"); /* P[0] */
183 asm volatile("pxor %xmm3,%xmm3"); /* P[1] */
184 asm volatile("pxor %xmm4,%xmm4"); /* Q[0] */
185 asm volatile("pxor %xmm5,%xmm5"); /* Zero temp */
186 asm volatile("pxor %xmm6,%xmm6"); /* Q[1] */
187 asm volatile("pxor %xmm7,%xmm7"); /* Zero temp */
188 asm volatile("pxor %xmm10,%xmm10"); /* P[2] */
189 asm volatile("pxor %xmm11,%xmm11"); /* P[3] */
190 asm volatile("pxor %xmm12,%xmm12"); /* Q[2] */
191 asm volatile("pxor %xmm13,%xmm13"); /* Zero temp */
192 asm volatile("pxor %xmm14,%xmm14"); /* Q[3] */
193 asm volatile("pxor %xmm15,%xmm15"); /* Zero temp */
194
195 for ( d = 0 ; d < bytes ; d += 64 ) {
196 for ( z = z0 ; z >= 0 ; z-- ) {
197 /* The second prefetch seems to improve performance... */
198 asm volatile("prefetchnta %0" :: "m" (dptr[z][d]));
199 asm volatile("prefetchnta %0" :: "m" (dptr[z][d+32]));
200 asm volatile("pcmpgtb %xmm4,%xmm5");
201 asm volatile("pcmpgtb %xmm6,%xmm7");
202 asm volatile("pcmpgtb %xmm12,%xmm13");
203 asm volatile("pcmpgtb %xmm14,%xmm15");
204 asm volatile("paddb %xmm4,%xmm4");
205 asm volatile("paddb %xmm6,%xmm6");
206 asm volatile("paddb %xmm12,%xmm12");
207 asm volatile("paddb %xmm14,%xmm14");
208 asm volatile("pand %xmm0,%xmm5");
209 asm volatile("pand %xmm0,%xmm7");
210 asm volatile("pand %xmm0,%xmm13");
211 asm volatile("pand %xmm0,%xmm15");
212 asm volatile("pxor %xmm5,%xmm4");
213 asm volatile("pxor %xmm7,%xmm6");
214 asm volatile("pxor %xmm13,%xmm12");
215 asm volatile("pxor %xmm15,%xmm14");
216 asm volatile("movdqa %0,%%xmm5" :: "m" (dptr[z][d]));
217 asm volatile("movdqa %0,%%xmm7" :: "m" (dptr[z][d+16]));
218 asm volatile("movdqa %0,%%xmm13" :: "m" (dptr[z][d+32]));
219 asm volatile("movdqa %0,%%xmm15" :: "m" (dptr[z][d+48]));
220 asm volatile("pxor %xmm5,%xmm2");
221 asm volatile("pxor %xmm7,%xmm3");
222 asm volatile("pxor %xmm13,%xmm10");
223 asm volatile("pxor %xmm15,%xmm11");
224 asm volatile("pxor %xmm5,%xmm4");
225 asm volatile("pxor %xmm7,%xmm6");
226 asm volatile("pxor %xmm13,%xmm12");
227 asm volatile("pxor %xmm15,%xmm14");
228 asm volatile("pxor %xmm5,%xmm5");
229 asm volatile("pxor %xmm7,%xmm7");
230 asm volatile("pxor %xmm13,%xmm13");
231 asm volatile("pxor %xmm15,%xmm15");
232 }
233 asm volatile("movntdq %%xmm2,%0" : "=m" (p[d]));
234 asm volatile("pxor %xmm2,%xmm2");
235 asm volatile("movntdq %%xmm3,%0" : "=m" (p[d+16]));
236 asm volatile("pxor %xmm3,%xmm3");
237 asm volatile("movntdq %%xmm10,%0" : "=m" (p[d+32]));
238 asm volatile("pxor %xmm10,%xmm10");
239 asm volatile("movntdq %%xmm11,%0" : "=m" (p[d+48]));
240 asm volatile("pxor %xmm11,%xmm11");
241 asm volatile("movntdq %%xmm4,%0" : "=m" (q[d]));
242 asm volatile("pxor %xmm4,%xmm4");
243 asm volatile("movntdq %%xmm6,%0" : "=m" (q[d+16]));
244 asm volatile("pxor %xmm6,%xmm6");
245 asm volatile("movntdq %%xmm12,%0" : "=m" (q[d+32]));
246 asm volatile("pxor %xmm12,%xmm12");
247 asm volatile("movntdq %%xmm14,%0" : "=m" (q[d+48]));
248 asm volatile("pxor %xmm14,%xmm14");
249 }
250
251 asm volatile("sfence" : : : "memory");
252 kernel_fpu_end();
253}
254
255const struct raid6_calls raid6_sse2x4 = {
256 raid6_sse24_gen_syndrome,
257 raid6_have_sse2,
258 "sse2x4",
259 1 /* Has cache hints */
260};
261
262#endif
diff --git a/lib/raid6/test/Makefile b/lib/raid6/test/Makefile
new file mode 100644
index 000000000000..aa651697b6dc
--- /dev/null
+++ b/lib/raid6/test/Makefile
@@ -0,0 +1,72 @@
1#
2# This is a simple Makefile to test some of the RAID-6 code
3# from userspace.
4#
5
6CC = gcc
7OPTFLAGS = -O2 # Adjust as desired
8CFLAGS = -I.. -I ../../../include -g $(OPTFLAGS)
9LD = ld
10AWK = awk -f
11AR = ar
12RANLIB = ranlib
13
14.c.o:
15 $(CC) $(CFLAGS) -c -o $@ $<
16
17%.c: ../%.c
18 cp -f $< $@
19
20%.uc: ../%.uc
21 cp -f $< $@
22
23all: raid6.a raid6test
24
25raid6.a: int1.o int2.o int4.o int8.o int16.o int32.o mmx.o sse1.o sse2.o \
26 altivec1.o altivec2.o altivec4.o altivec8.o recov.o algos.o \
27 tables.o
28 rm -f $@
29 $(AR) cq $@ $^
30 $(RANLIB) $@
31
32raid6test: test.c raid6.a
33 $(CC) $(CFLAGS) -o raid6test $^
34
35altivec1.c: altivec.uc ../unroll.awk
36 $(AWK) ../unroll.awk -vN=1 < altivec.uc > $@
37
38altivec2.c: altivec.uc ../unroll.awk
39 $(AWK) ../unroll.awk -vN=2 < altivec.uc > $@
40
41altivec4.c: altivec.uc ../unroll.awk
42 $(AWK) ../unroll.awk -vN=4 < altivec.uc > $@
43
44altivec8.c: altivec.uc ../unroll.awk
45 $(AWK) ../unroll.awk -vN=8 < altivec.uc > $@
46
47int1.c: int.uc ../unroll.awk
48 $(AWK) ../unroll.awk -vN=1 < int.uc > $@
49
50int2.c: int.uc ../unroll.awk
51 $(AWK) ../unroll.awk -vN=2 < int.uc > $@
52
53int4.c: int.uc ../unroll.awk
54 $(AWK) ../unroll.awk -vN=4 < int.uc > $@
55
56int8.c: int.uc ../unroll.awk
57 $(AWK) ../unroll.awk -vN=8 < int.uc > $@
58
59int16.c: int.uc ../unroll.awk
60 $(AWK) ../unroll.awk -vN=16 < int.uc > $@
61
62int32.c: int.uc ../unroll.awk
63 $(AWK) ../unroll.awk -vN=32 < int.uc > $@
64
65tables.c: mktables
66 ./mktables > tables.c
67
68clean:
69 rm -f *.o *.a mktables mktables.c *.uc int*.c altivec*.c tables.c raid6test
70
71spotless: clean
72 rm -f *~
diff --git a/lib/raid6/test/test.c b/lib/raid6/test/test.c
new file mode 100644
index 000000000000..7a930318b17d
--- /dev/null
+++ b/lib/raid6/test/test.c
@@ -0,0 +1,124 @@
1/* -*- linux-c -*- ------------------------------------------------------- *
2 *
3 * Copyright 2002-2007 H. Peter Anvin - All Rights Reserved
4 *
5 * This file is part of the Linux kernel, and is made available under
6 * the terms of the GNU General Public License version 2 or (at your
7 * option) any later version; incorporated herein by reference.
8 *
9 * ----------------------------------------------------------------------- */
10
11/*
12 * raid6test.c
13 *
14 * Test RAID-6 recovery with various algorithms
15 */
16
17#include <stdlib.h>
18#include <stdio.h>
19#include <string.h>
20#include <linux/raid/pq.h>
21
22#define NDISKS 16 /* Including P and Q */
23
24const char raid6_empty_zero_page[PAGE_SIZE] __attribute__((aligned(256)));
25struct raid6_calls raid6_call;
26
27char *dataptrs[NDISKS];
28char data[NDISKS][PAGE_SIZE];
29char recovi[PAGE_SIZE], recovj[PAGE_SIZE];
30
31static void makedata(void)
32{
33 int i, j;
34
35 for (i = 0; i < NDISKS; i++) {
36 for (j = 0; j < PAGE_SIZE; j++)
37 data[i][j] = rand();
38
39 dataptrs[i] = data[i];
40 }
41}
42
43static char disk_type(int d)
44{
45 switch (d) {
46 case NDISKS-2:
47 return 'P';
48 case NDISKS-1:
49 return 'Q';
50 default:
51 return 'D';
52 }
53}
54
55static int test_disks(int i, int j)
56{
57 int erra, errb;
58
59 memset(recovi, 0xf0, PAGE_SIZE);
60 memset(recovj, 0xba, PAGE_SIZE);
61
62 dataptrs[i] = recovi;
63 dataptrs[j] = recovj;
64
65 raid6_dual_recov(NDISKS, PAGE_SIZE, i, j, (void **)&dataptrs);
66
67 erra = memcmp(data[i], recovi, PAGE_SIZE);
68 errb = memcmp(data[j], recovj, PAGE_SIZE);
69
70 if (i < NDISKS-2 && j == NDISKS-1) {
71 /* We don't implement the DQ failure scenario, since it's
72 equivalent to a RAID-5 failure (XOR, then recompute Q) */
73 erra = errb = 0;
74 } else {
75 printf("algo=%-8s faila=%3d(%c) failb=%3d(%c) %s\n",
76 raid6_call.name,
77 i, disk_type(i),
78 j, disk_type(j),
79 (!erra && !errb) ? "OK" :
80 !erra ? "ERRB" :
81 !errb ? "ERRA" : "ERRAB");
82 }
83
84 dataptrs[i] = data[i];
85 dataptrs[j] = data[j];
86
87 return erra || errb;
88}
89
90int main(int argc, char *argv[])
91{
92 const struct raid6_calls *const *algo;
93 int i, j;
94 int err = 0;
95
96 makedata();
97
98 for (algo = raid6_algos; *algo; algo++) {
99 if (!(*algo)->valid || (*algo)->valid()) {
100 raid6_call = **algo;
101
102 /* Nuke syndromes */
103 memset(data[NDISKS-2], 0xee, 2*PAGE_SIZE);
104
105 /* Generate assumed good syndrome */
106 raid6_call.gen_syndrome(NDISKS, PAGE_SIZE,
107 (void **)&dataptrs);
108
109 for (i = 0; i < NDISKS-1; i++)
110 for (j = i+1; j < NDISKS; j++)
111 err += test_disks(i, j);
112 }
113 printf("\n");
114 }
115
116 printf("\n");
117 /* Pick the best algorithm test */
118 raid6_select_algo();
119
120 if (err)
121 printf("\n*** ERRORS FOUND ***\n");
122
123 return err;
124}
diff --git a/lib/raid6/unroll.awk b/lib/raid6/unroll.awk
new file mode 100644
index 000000000000..c6aa03631df8
--- /dev/null
+++ b/lib/raid6/unroll.awk
@@ -0,0 +1,20 @@
1
2# This filter requires one command line option of form -vN=n
3# where n must be a decimal number.
4#
5# Repeat each input line containing $$ n times, replacing $$ with 0...n-1.
6# Replace each $# with n, and each $* with a single $.
7
8BEGIN {
9 n = N + 0
10}
11{
12 if (/\$\$/) { rep = n } else { rep = 1 }
13 for (i = 0; i < rep; ++i) {
14 tmp = $0
15 gsub(/\$\$/, i, tmp)
16 gsub(/\$\#/, n, tmp)
17 gsub(/\$\*/, "$", tmp)
18 print tmp
19 }
20}
diff --git a/lib/raid6/x86.h b/lib/raid6/x86.h
new file mode 100644
index 000000000000..cb2a8c91c886
--- /dev/null
+++ b/lib/raid6/x86.h
@@ -0,0 +1,61 @@
1/* ----------------------------------------------------------------------- *
2 *
3 * Copyright 2002-2004 H. Peter Anvin - All Rights Reserved
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation, Inc., 53 Temple Place Ste 330,
8 * Boston MA 02111-1307, USA; either version 2 of the License, or
9 * (at your option) any later version; incorporated herein by reference.
10 *
11 * ----------------------------------------------------------------------- */
12
13/*
14 * raid6/x86.h
15 *
16 * Definitions common to x86 and x86-64 RAID-6 code only
17 */
18
19#ifndef LINUX_RAID_RAID6X86_H
20#define LINUX_RAID_RAID6X86_H
21
22#if (defined(__i386__) || defined(__x86_64__)) && !defined(__arch_um__)
23
24#ifdef __KERNEL__ /* Real code */
25
26#include <asm/i387.h>
27
28#else /* Dummy code for user space testing */
29
30static inline void kernel_fpu_begin(void)
31{
32}
33
34static inline void kernel_fpu_end(void)
35{
36}
37
38#define X86_FEATURE_MMX (0*32+23) /* Multimedia Extensions */
39#define X86_FEATURE_FXSR (0*32+24) /* FXSAVE and FXRSTOR instructions
40 * (fast save and restore) */
41#define X86_FEATURE_XMM (0*32+25) /* Streaming SIMD Extensions */
42#define X86_FEATURE_XMM2 (0*32+26) /* Streaming SIMD Extensions-2 */
43#define X86_FEATURE_MMXEXT (1*32+22) /* AMD MMX extensions */
44
45/* Should work well enough on modern CPUs for testing */
46static inline int boot_cpu_has(int flag)
47{
48 u32 eax = (flag >> 5) ? 0x80000001 : 1;
49 u32 edx;
50
51 asm volatile("cpuid"
52 : "+a" (eax), "=d" (edx)
53 : : "ecx", "ebx");
54
55 return (edx >> (flag & 31)) & 1;
56}
57
58#endif /* ndef __KERNEL__ */
59
60#endif
61#endif
diff --git a/lib/random32.c b/lib/random32.c
index 870dc3fc0f0f..fc3545a32771 100644
--- a/lib/random32.c
+++ b/lib/random32.c
@@ -127,7 +127,7 @@ core_initcall(random32_init);
127 127
128/* 128/*
129 * Generate better values after random number generator 129 * Generate better values after random number generator
130 * is fully initalized. 130 * is fully initialized.
131 */ 131 */
132static int __init random32_reseed(void) 132static int __init random32_reseed(void)
133{ 133{
diff --git a/lib/rbtree.c b/lib/rbtree.c
index 15e10b1afdd2..4693f79195d3 100644
--- a/lib/rbtree.c
+++ b/lib/rbtree.c
@@ -44,11 +44,6 @@ static void __rb_rotate_left(struct rb_node *node, struct rb_root *root)
44 else 44 else
45 root->rb_node = right; 45 root->rb_node = right;
46 rb_set_parent(node, right); 46 rb_set_parent(node, right);
47
48 if (root->augment_cb) {
49 root->augment_cb(node);
50 root->augment_cb(right);
51 }
52} 47}
53 48
54static void __rb_rotate_right(struct rb_node *node, struct rb_root *root) 49static void __rb_rotate_right(struct rb_node *node, struct rb_root *root)
@@ -72,20 +67,12 @@ static void __rb_rotate_right(struct rb_node *node, struct rb_root *root)
72 else 67 else
73 root->rb_node = left; 68 root->rb_node = left;
74 rb_set_parent(node, left); 69 rb_set_parent(node, left);
75
76 if (root->augment_cb) {
77 root->augment_cb(node);
78 root->augment_cb(left);
79 }
80} 70}
81 71
82void rb_insert_color(struct rb_node *node, struct rb_root *root) 72void rb_insert_color(struct rb_node *node, struct rb_root *root)
83{ 73{
84 struct rb_node *parent, *gparent; 74 struct rb_node *parent, *gparent;
85 75
86 if (root->augment_cb)
87 root->augment_cb(node);
88
89 while ((parent = rb_parent(node)) && rb_is_red(parent)) 76 while ((parent = rb_parent(node)) && rb_is_red(parent))
90 { 77 {
91 gparent = rb_parent(parent); 78 gparent = rb_parent(parent);
@@ -240,15 +227,12 @@ void rb_erase(struct rb_node *node, struct rb_root *root)
240 else 227 else
241 { 228 {
242 struct rb_node *old = node, *left; 229 struct rb_node *old = node, *left;
243 int old_parent_cb = 0;
244 int successor_parent_cb = 0;
245 230
246 node = node->rb_right; 231 node = node->rb_right;
247 while ((left = node->rb_left) != NULL) 232 while ((left = node->rb_left) != NULL)
248 node = left; 233 node = left;
249 234
250 if (rb_parent(old)) { 235 if (rb_parent(old)) {
251 old_parent_cb = 1;
252 if (rb_parent(old)->rb_left == old) 236 if (rb_parent(old)->rb_left == old)
253 rb_parent(old)->rb_left = node; 237 rb_parent(old)->rb_left = node;
254 else 238 else
@@ -263,10 +247,8 @@ void rb_erase(struct rb_node *node, struct rb_root *root)
263 if (parent == old) { 247 if (parent == old) {
264 parent = node; 248 parent = node;
265 } else { 249 } else {
266 successor_parent_cb = 1;
267 if (child) 250 if (child)
268 rb_set_parent(child, parent); 251 rb_set_parent(child, parent);
269
270 parent->rb_left = child; 252 parent->rb_left = child;
271 253
272 node->rb_right = old->rb_right; 254 node->rb_right = old->rb_right;
@@ -277,24 +259,6 @@ void rb_erase(struct rb_node *node, struct rb_root *root)
277 node->rb_left = old->rb_left; 259 node->rb_left = old->rb_left;
278 rb_set_parent(old->rb_left, node); 260 rb_set_parent(old->rb_left, node);
279 261
280 if (root->augment_cb) {
281 /*
282 * Here, three different nodes can have new children.
283 * The parent of the successor node that was selected
284 * to replace the node to be erased.
285 * The node that is getting erased and is now replaced
286 * by its successor.
287 * The parent of the node getting erased-replaced.
288 */
289 if (successor_parent_cb)
290 root->augment_cb(parent);
291
292 root->augment_cb(node);
293
294 if (old_parent_cb)
295 root->augment_cb(rb_parent(old));
296 }
297
298 goto color; 262 goto color;
299 } 263 }
300 264
@@ -303,19 +267,15 @@ void rb_erase(struct rb_node *node, struct rb_root *root)
303 267
304 if (child) 268 if (child)
305 rb_set_parent(child, parent); 269 rb_set_parent(child, parent);
306 270 if (parent)
307 if (parent) { 271 {
308 if (parent->rb_left == node) 272 if (parent->rb_left == node)
309 parent->rb_left = child; 273 parent->rb_left = child;
310 else 274 else
311 parent->rb_right = child; 275 parent->rb_right = child;
312
313 if (root->augment_cb)
314 root->augment_cb(parent);
315
316 } else {
317 root->rb_node = child;
318 } 276 }
277 else
278 root->rb_node = child;
319 279
320 color: 280 color:
321 if (color == RB_BLACK) 281 if (color == RB_BLACK)
@@ -323,6 +283,74 @@ void rb_erase(struct rb_node *node, struct rb_root *root)
323} 283}
324EXPORT_SYMBOL(rb_erase); 284EXPORT_SYMBOL(rb_erase);
325 285
286static void rb_augment_path(struct rb_node *node, rb_augment_f func, void *data)
287{
288 struct rb_node *parent;
289
290up:
291 func(node, data);
292 parent = rb_parent(node);
293 if (!parent)
294 return;
295
296 if (node == parent->rb_left && parent->rb_right)
297 func(parent->rb_right, data);
298 else if (parent->rb_left)
299 func(parent->rb_left, data);
300
301 node = parent;
302 goto up;
303}
304
305/*
306 * after inserting @node into the tree, update the tree to account for
307 * both the new entry and any damage done by rebalance
308 */
309void rb_augment_insert(struct rb_node *node, rb_augment_f func, void *data)
310{
311 if (node->rb_left)
312 node = node->rb_left;
313 else if (node->rb_right)
314 node = node->rb_right;
315
316 rb_augment_path(node, func, data);
317}
318
319/*
320 * before removing the node, find the deepest node on the rebalance path
321 * that will still be there after @node gets removed
322 */
323struct rb_node *rb_augment_erase_begin(struct rb_node *node)
324{
325 struct rb_node *deepest;
326
327 if (!node->rb_right && !node->rb_left)
328 deepest = rb_parent(node);
329 else if (!node->rb_right)
330 deepest = node->rb_left;
331 else if (!node->rb_left)
332 deepest = node->rb_right;
333 else {
334 deepest = rb_next(node);
335 if (deepest->rb_right)
336 deepest = deepest->rb_right;
337 else if (rb_parent(deepest) != node)
338 deepest = rb_parent(deepest);
339 }
340
341 return deepest;
342}
343
344/*
345 * after removal, update the tree to account for the removed entry
346 * and any rebalance damage.
347 */
348void rb_augment_erase_end(struct rb_node *node, rb_augment_f func, void *data)
349{
350 if (node)
351 rb_augment_path(node, func, data);
352}
353
326/* 354/*
327 * This function returns the first node (in sort order) of the tree. 355 * This function returns the first node (in sort order) of the tree.
328 */ 356 */
diff --git a/lib/rwsem.c b/lib/rwsem.c
index ceba8e28807a..f236d7cd5cf3 100644
--- a/lib/rwsem.c
+++ b/lib/rwsem.c
@@ -36,45 +36,56 @@ struct rwsem_waiter {
36#define RWSEM_WAITING_FOR_WRITE 0x00000002 36#define RWSEM_WAITING_FOR_WRITE 0x00000002
37}; 37};
38 38
39/* Wake types for __rwsem_do_wake(). Note that RWSEM_WAKE_NO_ACTIVE and
40 * RWSEM_WAKE_READ_OWNED imply that the spinlock must have been kept held
41 * since the rwsem value was observed.
42 */
43#define RWSEM_WAKE_ANY 0 /* Wake whatever's at head of wait list */
44#define RWSEM_WAKE_NO_ACTIVE 1 /* rwsem was observed with no active thread */
45#define RWSEM_WAKE_READ_OWNED 2 /* rwsem was observed to be read owned */
46
39/* 47/*
40 * handle the lock release when processes blocked on it that can now run 48 * handle the lock release when processes blocked on it that can now run
41 * - if we come here from up_xxxx(), then: 49 * - if we come here from up_xxxx(), then:
42 * - the 'active part' of count (&0x0000ffff) reached 0 (but may have changed) 50 * - the 'active part' of count (&0x0000ffff) reached 0 (but may have changed)
43 * - the 'waiting part' of count (&0xffff0000) is -ve (and will still be so) 51 * - the 'waiting part' of count (&0xffff0000) is -ve (and will still be so)
44 * - there must be someone on the queue 52 * - there must be someone on the queue
45 * - the spinlock must be held by the caller 53 * - the spinlock must be held by the caller
46 * - woken process blocks are discarded from the list after having task zeroed 54 * - woken process blocks are discarded from the list after having task zeroed
47 * - writers are only woken if downgrading is false 55 * - writers are only woken if downgrading is false
48 */ 56 */
49static inline struct rw_semaphore * 57static struct rw_semaphore *
50__rwsem_do_wake(struct rw_semaphore *sem, int downgrading) 58__rwsem_do_wake(struct rw_semaphore *sem, int wake_type)
51{ 59{
52 struct rwsem_waiter *waiter; 60 struct rwsem_waiter *waiter;
53 struct task_struct *tsk; 61 struct task_struct *tsk;
54 struct list_head *next; 62 struct list_head *next;
55 signed long oldcount, woken, loop; 63 signed long oldcount, woken, loop, adjustment;
56
57 if (downgrading)
58 goto dont_wake_writers;
59
60 /* if we came through an up_xxxx() call, we only only wake someone up
61 * if we can transition the active part of the count from 0 -> 1
62 */
63 try_again:
64 oldcount = rwsem_atomic_update(RWSEM_ACTIVE_BIAS, sem)
65 - RWSEM_ACTIVE_BIAS;
66 if (oldcount & RWSEM_ACTIVE_MASK)
67 goto undo;
68 64
69 waiter = list_entry(sem->wait_list.next, struct rwsem_waiter, list); 65 waiter = list_entry(sem->wait_list.next, struct rwsem_waiter, list);
70
71 /* try to grant a single write lock if there's a writer at the front
72 * of the queue - note we leave the 'active part' of the count
73 * incremented by 1 and the waiting part incremented by 0x00010000
74 */
75 if (!(waiter->flags & RWSEM_WAITING_FOR_WRITE)) 66 if (!(waiter->flags & RWSEM_WAITING_FOR_WRITE))
76 goto readers_only; 67 goto readers_only;
77 68
69 if (wake_type == RWSEM_WAKE_READ_OWNED)
70 /* Another active reader was observed, so wakeup is not
71 * likely to succeed. Save the atomic op.
72 */
73 goto out;
74
75 /* There's a writer at the front of the queue - try to grant it the
76 * write lock. However, we only wake this writer if we can transition
77 * the active part of the count from 0 -> 1
78 */
79 adjustment = RWSEM_ACTIVE_WRITE_BIAS;
80 if (waiter->list.next == &sem->wait_list)
81 adjustment -= RWSEM_WAITING_BIAS;
82
83 try_again_write:
84 oldcount = rwsem_atomic_update(adjustment, sem) - adjustment;
85 if (oldcount & RWSEM_ACTIVE_MASK)
86 /* Someone grabbed the sem already */
87 goto undo_write;
88
78 /* We must be careful not to touch 'waiter' after we set ->task = NULL. 89 /* We must be careful not to touch 'waiter' after we set ->task = NULL.
79 * It is an allocated on the waiter's stack and may become invalid at 90 * It is an allocated on the waiter's stack and may become invalid at
80 * any time after that point (due to a wakeup from another source). 91 * any time after that point (due to a wakeup from another source).
@@ -87,18 +98,30 @@ __rwsem_do_wake(struct rw_semaphore *sem, int downgrading)
87 put_task_struct(tsk); 98 put_task_struct(tsk);
88 goto out; 99 goto out;
89 100
90 /* don't want to wake any writers */ 101 readers_only:
91 dont_wake_writers: 102 /* If we come here from up_xxxx(), another thread might have reached
92 waiter = list_entry(sem->wait_list.next, struct rwsem_waiter, list); 103 * rwsem_down_failed_common() before we acquired the spinlock and
93 if (waiter->flags & RWSEM_WAITING_FOR_WRITE) 104 * woken up a waiter, making it now active. We prefer to check for
105 * this first in order to not spend too much time with the spinlock
106 * held if we're not going to be able to wake up readers in the end.
107 *
108 * Note that we do not need to update the rwsem count: any writer
109 * trying to acquire rwsem will run rwsem_down_write_failed() due
110 * to the waiting threads and block trying to acquire the spinlock.
111 *
112 * We use a dummy atomic update in order to acquire the cache line
113 * exclusively since we expect to succeed and run the final rwsem
114 * count adjustment pretty soon.
115 */
116 if (wake_type == RWSEM_WAKE_ANY &&
117 rwsem_atomic_update(0, sem) < RWSEM_WAITING_BIAS)
118 /* Someone grabbed the sem for write already */
94 goto out; 119 goto out;
95 120
96 /* grant an infinite number of read locks to the readers at the front 121 /* Grant an infinite number of read locks to the readers at the front
97 * of the queue 122 * of the queue. Note we increment the 'active part' of the count by
98 * - note we increment the 'active part' of the count by the number of 123 * the number of readers before waking any processes up.
99 * readers before waking any processes up
100 */ 124 */
101 readers_only:
102 woken = 0; 125 woken = 0;
103 do { 126 do {
104 woken++; 127 woken++;
@@ -111,16 +134,15 @@ __rwsem_do_wake(struct rw_semaphore *sem, int downgrading)
111 134
112 } while (waiter->flags & RWSEM_WAITING_FOR_READ); 135 } while (waiter->flags & RWSEM_WAITING_FOR_READ);
113 136
114 loop = woken; 137 adjustment = woken * RWSEM_ACTIVE_READ_BIAS;
115 woken *= RWSEM_ACTIVE_BIAS - RWSEM_WAITING_BIAS; 138 if (waiter->flags & RWSEM_WAITING_FOR_READ)
116 if (!downgrading) 139 /* hit end of list above */
117 /* we'd already done one increment earlier */ 140 adjustment -= RWSEM_WAITING_BIAS;
118 woken -= RWSEM_ACTIVE_BIAS;
119 141
120 rwsem_atomic_add(woken, sem); 142 rwsem_atomic_add(adjustment, sem);
121 143
122 next = sem->wait_list.next; 144 next = sem->wait_list.next;
123 for (; loop > 0; loop--) { 145 for (loop = woken; loop > 0; loop--) {
124 waiter = list_entry(next, struct rwsem_waiter, list); 146 waiter = list_entry(next, struct rwsem_waiter, list);
125 next = waiter->list.next; 147 next = waiter->list.next;
126 tsk = waiter->task; 148 tsk = waiter->task;
@@ -138,10 +160,10 @@ __rwsem_do_wake(struct rw_semaphore *sem, int downgrading)
138 160
139 /* undo the change to the active count, but check for a transition 161 /* undo the change to the active count, but check for a transition
140 * 1->0 */ 162 * 1->0 */
141 undo: 163 undo_write:
142 if (rwsem_atomic_update(-RWSEM_ACTIVE_BIAS, sem) & RWSEM_ACTIVE_MASK) 164 if (rwsem_atomic_update(-adjustment, sem) & RWSEM_ACTIVE_MASK)
143 goto out; 165 goto out;
144 goto try_again; 166 goto try_again_write;
145} 167}
146 168
147/* 169/*
@@ -149,8 +171,9 @@ __rwsem_do_wake(struct rw_semaphore *sem, int downgrading)
149 */ 171 */
150static struct rw_semaphore __sched * 172static struct rw_semaphore __sched *
151rwsem_down_failed_common(struct rw_semaphore *sem, 173rwsem_down_failed_common(struct rw_semaphore *sem,
152 struct rwsem_waiter *waiter, signed long adjustment) 174 unsigned int flags, signed long adjustment)
153{ 175{
176 struct rwsem_waiter waiter;
154 struct task_struct *tsk = current; 177 struct task_struct *tsk = current;
155 signed long count; 178 signed long count;
156 179
@@ -158,23 +181,34 @@ rwsem_down_failed_common(struct rw_semaphore *sem,
158 181
159 /* set up my own style of waitqueue */ 182 /* set up my own style of waitqueue */
160 spin_lock_irq(&sem->wait_lock); 183 spin_lock_irq(&sem->wait_lock);
161 waiter->task = tsk; 184 waiter.task = tsk;
185 waiter.flags = flags;
162 get_task_struct(tsk); 186 get_task_struct(tsk);
163 187
164 list_add_tail(&waiter->list, &sem->wait_list); 188 if (list_empty(&sem->wait_list))
189 adjustment += RWSEM_WAITING_BIAS;
190 list_add_tail(&waiter.list, &sem->wait_list);
165 191
166 /* we're now waiting on the lock, but no longer actively read-locking */ 192 /* we're now waiting on the lock, but no longer actively locking */
167 count = rwsem_atomic_update(adjustment, sem); 193 count = rwsem_atomic_update(adjustment, sem);
168 194
169 /* if there are no active locks, wake the front queued process(es) up */ 195 /* If there are no active locks, wake the front queued process(es) up.
170 if (!(count & RWSEM_ACTIVE_MASK)) 196 *
171 sem = __rwsem_do_wake(sem, 0); 197 * Alternatively, if we're called from a failed down_write(), there
198 * were already threads queued before us and there are no active
199 * writers, the lock must be read owned; so we try to wake any read
200 * locks that were queued ahead of us. */
201 if (count == RWSEM_WAITING_BIAS)
202 sem = __rwsem_do_wake(sem, RWSEM_WAKE_NO_ACTIVE);
203 else if (count > RWSEM_WAITING_BIAS &&
204 adjustment == -RWSEM_ACTIVE_WRITE_BIAS)
205 sem = __rwsem_do_wake(sem, RWSEM_WAKE_READ_OWNED);
172 206
173 spin_unlock_irq(&sem->wait_lock); 207 spin_unlock_irq(&sem->wait_lock);
174 208
175 /* wait to be given the lock */ 209 /* wait to be given the lock */
176 for (;;) { 210 for (;;) {
177 if (!waiter->task) 211 if (!waiter.task)
178 break; 212 break;
179 schedule(); 213 schedule();
180 set_task_state(tsk, TASK_UNINTERRUPTIBLE); 214 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
@@ -191,12 +225,8 @@ rwsem_down_failed_common(struct rw_semaphore *sem,
191asmregparm struct rw_semaphore __sched * 225asmregparm struct rw_semaphore __sched *
192rwsem_down_read_failed(struct rw_semaphore *sem) 226rwsem_down_read_failed(struct rw_semaphore *sem)
193{ 227{
194 struct rwsem_waiter waiter; 228 return rwsem_down_failed_common(sem, RWSEM_WAITING_FOR_READ,
195 229 -RWSEM_ACTIVE_READ_BIAS);
196 waiter.flags = RWSEM_WAITING_FOR_READ;
197 rwsem_down_failed_common(sem, &waiter,
198 RWSEM_WAITING_BIAS - RWSEM_ACTIVE_BIAS);
199 return sem;
200} 230}
201 231
202/* 232/*
@@ -205,12 +235,8 @@ rwsem_down_read_failed(struct rw_semaphore *sem)
205asmregparm struct rw_semaphore __sched * 235asmregparm struct rw_semaphore __sched *
206rwsem_down_write_failed(struct rw_semaphore *sem) 236rwsem_down_write_failed(struct rw_semaphore *sem)
207{ 237{
208 struct rwsem_waiter waiter; 238 return rwsem_down_failed_common(sem, RWSEM_WAITING_FOR_WRITE,
209 239 -RWSEM_ACTIVE_WRITE_BIAS);
210 waiter.flags = RWSEM_WAITING_FOR_WRITE;
211 rwsem_down_failed_common(sem, &waiter, -RWSEM_ACTIVE_BIAS);
212
213 return sem;
214} 240}
215 241
216/* 242/*
@@ -225,7 +251,7 @@ asmregparm struct rw_semaphore *rwsem_wake(struct rw_semaphore *sem)
225 251
226 /* do nothing if list empty */ 252 /* do nothing if list empty */
227 if (!list_empty(&sem->wait_list)) 253 if (!list_empty(&sem->wait_list))
228 sem = __rwsem_do_wake(sem, 0); 254 sem = __rwsem_do_wake(sem, RWSEM_WAKE_ANY);
229 255
230 spin_unlock_irqrestore(&sem->wait_lock, flags); 256 spin_unlock_irqrestore(&sem->wait_lock, flags);
231 257
@@ -245,7 +271,7 @@ asmregparm struct rw_semaphore *rwsem_downgrade_wake(struct rw_semaphore *sem)
245 271
246 /* do nothing if list empty */ 272 /* do nothing if list empty */
247 if (!list_empty(&sem->wait_list)) 273 if (!list_empty(&sem->wait_list))
248 sem = __rwsem_do_wake(sem, 1); 274 sem = __rwsem_do_wake(sem, RWSEM_WAKE_READ_OWNED);
249 275
250 spin_unlock_irqrestore(&sem->wait_lock, flags); 276 spin_unlock_irqrestore(&sem->wait_lock, flags);
251 277
diff --git a/lib/scatterlist.c b/lib/scatterlist.c
index 9afa25b52a83..a5ec42868f99 100644
--- a/lib/scatterlist.c
+++ b/lib/scatterlist.c
@@ -10,6 +10,7 @@
10#include <linux/slab.h> 10#include <linux/slab.h>
11#include <linux/scatterlist.h> 11#include <linux/scatterlist.h>
12#include <linux/highmem.h> 12#include <linux/highmem.h>
13#include <linux/kmemleak.h>
13 14
14/** 15/**
15 * sg_next - return the next scatterlist entry in a list 16 * sg_next - return the next scatterlist entry in a list
@@ -115,17 +116,29 @@ EXPORT_SYMBOL(sg_init_one);
115 */ 116 */
116static struct scatterlist *sg_kmalloc(unsigned int nents, gfp_t gfp_mask) 117static struct scatterlist *sg_kmalloc(unsigned int nents, gfp_t gfp_mask)
117{ 118{
118 if (nents == SG_MAX_SINGLE_ALLOC) 119 if (nents == SG_MAX_SINGLE_ALLOC) {
119 return (struct scatterlist *) __get_free_page(gfp_mask); 120 /*
120 else 121 * Kmemleak doesn't track page allocations as they are not
122 * commonly used (in a raw form) for kernel data structures.
123 * As we chain together a list of pages and then a normal
124 * kmalloc (tracked by kmemleak), in order to for that last
125 * allocation not to become decoupled (and thus a
126 * false-positive) we need to inform kmemleak of all the
127 * intermediate allocations.
128 */
129 void *ptr = (void *) __get_free_page(gfp_mask);
130 kmemleak_alloc(ptr, PAGE_SIZE, 1, gfp_mask);
131 return ptr;
132 } else
121 return kmalloc(nents * sizeof(struct scatterlist), gfp_mask); 133 return kmalloc(nents * sizeof(struct scatterlist), gfp_mask);
122} 134}
123 135
124static void sg_kfree(struct scatterlist *sg, unsigned int nents) 136static void sg_kfree(struct scatterlist *sg, unsigned int nents)
125{ 137{
126 if (nents == SG_MAX_SINGLE_ALLOC) 138 if (nents == SG_MAX_SINGLE_ALLOC) {
139 kmemleak_free(sg);
127 free_page((unsigned long) sg); 140 free_page((unsigned long) sg);
128 else 141 } else
129 kfree(sg); 142 kfree(sg);
130} 143}
131 144
diff --git a/lib/swiotlb.c b/lib/swiotlb.c
index a009055140ec..34e3082632d8 100644
--- a/lib/swiotlb.c
+++ b/lib/swiotlb.c
@@ -50,19 +50,11 @@
50 */ 50 */
51#define IO_TLB_MIN_SLABS ((1<<20) >> IO_TLB_SHIFT) 51#define IO_TLB_MIN_SLABS ((1<<20) >> IO_TLB_SHIFT)
52 52
53/*
54 * Enumeration for sync targets
55 */
56enum dma_sync_target {
57 SYNC_FOR_CPU = 0,
58 SYNC_FOR_DEVICE = 1,
59};
60
61int swiotlb_force; 53int swiotlb_force;
62 54
63/* 55/*
64 * Used to do a quick range check in unmap_single and 56 * Used to do a quick range check in swiotlb_tbl_unmap_single and
65 * sync_single_*, to see if the memory was in fact allocated by this 57 * swiotlb_tbl_sync_single_*, to see if the memory was in fact allocated by this
66 * API. 58 * API.
67 */ 59 */
68static char *io_tlb_start, *io_tlb_end; 60static char *io_tlb_start, *io_tlb_end;
@@ -140,28 +132,14 @@ void swiotlb_print_info(void)
140 (unsigned long long)pend); 132 (unsigned long long)pend);
141} 133}
142 134
143/* 135void __init swiotlb_init_with_tbl(char *tlb, unsigned long nslabs, int verbose)
144 * Statically reserve bounce buffer space and initialize bounce buffer data
145 * structures for the software IO TLB used to implement the DMA API.
146 */
147void __init
148swiotlb_init_with_default_size(size_t default_size, int verbose)
149{ 136{
150 unsigned long i, bytes; 137 unsigned long i, bytes;
151 138
152 if (!io_tlb_nslabs) { 139 bytes = nslabs << IO_TLB_SHIFT;
153 io_tlb_nslabs = (default_size >> IO_TLB_SHIFT);
154 io_tlb_nslabs = ALIGN(io_tlb_nslabs, IO_TLB_SEGSIZE);
155 }
156
157 bytes = io_tlb_nslabs << IO_TLB_SHIFT;
158 140
159 /* 141 io_tlb_nslabs = nslabs;
160 * Get IO TLB memory from the low pages 142 io_tlb_start = tlb;
161 */
162 io_tlb_start = alloc_bootmem_low_pages(bytes);
163 if (!io_tlb_start)
164 panic("Cannot allocate SWIOTLB buffer");
165 io_tlb_end = io_tlb_start + bytes; 143 io_tlb_end = io_tlb_start + bytes;
166 144
167 /* 145 /*
@@ -185,6 +163,32 @@ swiotlb_init_with_default_size(size_t default_size, int verbose)
185 swiotlb_print_info(); 163 swiotlb_print_info();
186} 164}
187 165
166/*
167 * Statically reserve bounce buffer space and initialize bounce buffer data
168 * structures for the software IO TLB used to implement the DMA API.
169 */
170void __init
171swiotlb_init_with_default_size(size_t default_size, int verbose)
172{
173 unsigned long bytes;
174
175 if (!io_tlb_nslabs) {
176 io_tlb_nslabs = (default_size >> IO_TLB_SHIFT);
177 io_tlb_nslabs = ALIGN(io_tlb_nslabs, IO_TLB_SEGSIZE);
178 }
179
180 bytes = io_tlb_nslabs << IO_TLB_SHIFT;
181
182 /*
183 * Get IO TLB memory from the low pages
184 */
185 io_tlb_start = alloc_bootmem_low_pages(bytes);
186 if (!io_tlb_start)
187 panic("Cannot allocate SWIOTLB buffer");
188
189 swiotlb_init_with_tbl(io_tlb_start, io_tlb_nslabs, verbose);
190}
191
188void __init 192void __init
189swiotlb_init(int verbose) 193swiotlb_init(int verbose)
190{ 194{
@@ -323,8 +327,8 @@ static int is_swiotlb_buffer(phys_addr_t paddr)
323/* 327/*
324 * Bounce: copy the swiotlb buffer back to the original dma location 328 * Bounce: copy the swiotlb buffer back to the original dma location
325 */ 329 */
326static void swiotlb_bounce(phys_addr_t phys, char *dma_addr, size_t size, 330void swiotlb_bounce(phys_addr_t phys, char *dma_addr, size_t size,
327 enum dma_data_direction dir) 331 enum dma_data_direction dir)
328{ 332{
329 unsigned long pfn = PFN_DOWN(phys); 333 unsigned long pfn = PFN_DOWN(phys);
330 334
@@ -360,26 +364,25 @@ static void swiotlb_bounce(phys_addr_t phys, char *dma_addr, size_t size,
360 memcpy(phys_to_virt(phys), dma_addr, size); 364 memcpy(phys_to_virt(phys), dma_addr, size);
361 } 365 }
362} 366}
367EXPORT_SYMBOL_GPL(swiotlb_bounce);
363 368
364/* 369void *swiotlb_tbl_map_single(struct device *hwdev, dma_addr_t tbl_dma_addr,
365 * Allocates bounce buffer and returns its kernel virtual address. 370 phys_addr_t phys, size_t size,
366 */ 371 enum dma_data_direction dir)
367static void *
368map_single(struct device *hwdev, phys_addr_t phys, size_t size, int dir)
369{ 372{
370 unsigned long flags; 373 unsigned long flags;
371 char *dma_addr; 374 char *dma_addr;
372 unsigned int nslots, stride, index, wrap; 375 unsigned int nslots, stride, index, wrap;
373 int i; 376 int i;
374 unsigned long start_dma_addr;
375 unsigned long mask; 377 unsigned long mask;
376 unsigned long offset_slots; 378 unsigned long offset_slots;
377 unsigned long max_slots; 379 unsigned long max_slots;
378 380
379 mask = dma_get_seg_boundary(hwdev); 381 mask = dma_get_seg_boundary(hwdev);
380 start_dma_addr = swiotlb_virt_to_bus(hwdev, io_tlb_start) & mask;
381 382
382 offset_slots = ALIGN(start_dma_addr, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT; 383 tbl_dma_addr &= mask;
384
385 offset_slots = ALIGN(tbl_dma_addr, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT;
383 386
384 /* 387 /*
385 * Carefully handle integer overflow which can occur when mask == ~0UL. 388 * Carefully handle integer overflow which can occur when mask == ~0UL.
@@ -466,12 +469,27 @@ found:
466 469
467 return dma_addr; 470 return dma_addr;
468} 471}
472EXPORT_SYMBOL_GPL(swiotlb_tbl_map_single);
473
474/*
475 * Allocates bounce buffer and returns its kernel virtual address.
476 */
477
478static void *
479map_single(struct device *hwdev, phys_addr_t phys, size_t size,
480 enum dma_data_direction dir)
481{
482 dma_addr_t start_dma_addr = swiotlb_virt_to_bus(hwdev, io_tlb_start);
483
484 return swiotlb_tbl_map_single(hwdev, start_dma_addr, phys, size, dir);
485}
469 486
470/* 487/*
471 * dma_addr is the kernel virtual address of the bounce buffer to unmap. 488 * dma_addr is the kernel virtual address of the bounce buffer to unmap.
472 */ 489 */
473static void 490void
474do_unmap_single(struct device *hwdev, char *dma_addr, size_t size, int dir) 491swiotlb_tbl_unmap_single(struct device *hwdev, char *dma_addr, size_t size,
492 enum dma_data_direction dir)
475{ 493{
476 unsigned long flags; 494 unsigned long flags;
477 int i, count, nslots = ALIGN(size, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT; 495 int i, count, nslots = ALIGN(size, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT;
@@ -509,10 +527,12 @@ do_unmap_single(struct device *hwdev, char *dma_addr, size_t size, int dir)
509 } 527 }
510 spin_unlock_irqrestore(&io_tlb_lock, flags); 528 spin_unlock_irqrestore(&io_tlb_lock, flags);
511} 529}
530EXPORT_SYMBOL_GPL(swiotlb_tbl_unmap_single);
512 531
513static void 532void
514sync_single(struct device *hwdev, char *dma_addr, size_t size, 533swiotlb_tbl_sync_single(struct device *hwdev, char *dma_addr, size_t size,
515 int dir, int target) 534 enum dma_data_direction dir,
535 enum dma_sync_target target)
516{ 536{
517 int index = (dma_addr - io_tlb_start) >> IO_TLB_SHIFT; 537 int index = (dma_addr - io_tlb_start) >> IO_TLB_SHIFT;
518 phys_addr_t phys = io_tlb_orig_addr[index]; 538 phys_addr_t phys = io_tlb_orig_addr[index];
@@ -536,6 +556,7 @@ sync_single(struct device *hwdev, char *dma_addr, size_t size,
536 BUG(); 556 BUG();
537 } 557 }
538} 558}
559EXPORT_SYMBOL_GPL(swiotlb_tbl_sync_single);
539 560
540void * 561void *
541swiotlb_alloc_coherent(struct device *hwdev, size_t size, 562swiotlb_alloc_coherent(struct device *hwdev, size_t size,
@@ -559,8 +580,8 @@ swiotlb_alloc_coherent(struct device *hwdev, size_t size,
559 } 580 }
560 if (!ret) { 581 if (!ret) {
561 /* 582 /*
562 * We are either out of memory or the device can't DMA 583 * We are either out of memory or the device can't DMA to
563 * to GFP_DMA memory; fall back on map_single(), which 584 * GFP_DMA memory; fall back on map_single(), which
564 * will grab memory from the lowest available address range. 585 * will grab memory from the lowest available address range.
565 */ 586 */
566 ret = map_single(hwdev, 0, size, DMA_FROM_DEVICE); 587 ret = map_single(hwdev, 0, size, DMA_FROM_DEVICE);
@@ -578,7 +599,7 @@ swiotlb_alloc_coherent(struct device *hwdev, size_t size,
578 (unsigned long long)dev_addr); 599 (unsigned long long)dev_addr);
579 600
580 /* DMA_TO_DEVICE to avoid memcpy in unmap_single */ 601 /* DMA_TO_DEVICE to avoid memcpy in unmap_single */
581 do_unmap_single(hwdev, ret, size, DMA_TO_DEVICE); 602 swiotlb_tbl_unmap_single(hwdev, ret, size, DMA_TO_DEVICE);
582 return NULL; 603 return NULL;
583 } 604 }
584 *dma_handle = dev_addr; 605 *dma_handle = dev_addr;
@@ -596,13 +617,14 @@ swiotlb_free_coherent(struct device *hwdev, size_t size, void *vaddr,
596 if (!is_swiotlb_buffer(paddr)) 617 if (!is_swiotlb_buffer(paddr))
597 free_pages((unsigned long)vaddr, get_order(size)); 618 free_pages((unsigned long)vaddr, get_order(size));
598 else 619 else
599 /* DMA_TO_DEVICE to avoid memcpy in unmap_single */ 620 /* DMA_TO_DEVICE to avoid memcpy in swiotlb_tbl_unmap_single */
600 do_unmap_single(hwdev, vaddr, size, DMA_TO_DEVICE); 621 swiotlb_tbl_unmap_single(hwdev, vaddr, size, DMA_TO_DEVICE);
601} 622}
602EXPORT_SYMBOL(swiotlb_free_coherent); 623EXPORT_SYMBOL(swiotlb_free_coherent);
603 624
604static void 625static void
605swiotlb_full(struct device *dev, size_t size, int dir, int do_panic) 626swiotlb_full(struct device *dev, size_t size, enum dma_data_direction dir,
627 int do_panic)
606{ 628{
607 /* 629 /*
608 * Ran out of IOMMU space for this operation. This is very bad. 630 * Ran out of IOMMU space for this operation. This is very bad.
@@ -680,14 +702,14 @@ EXPORT_SYMBOL_GPL(swiotlb_map_page);
680 * whatever the device wrote there. 702 * whatever the device wrote there.
681 */ 703 */
682static void unmap_single(struct device *hwdev, dma_addr_t dev_addr, 704static void unmap_single(struct device *hwdev, dma_addr_t dev_addr,
683 size_t size, int dir) 705 size_t size, enum dma_data_direction dir)
684{ 706{
685 phys_addr_t paddr = dma_to_phys(hwdev, dev_addr); 707 phys_addr_t paddr = dma_to_phys(hwdev, dev_addr);
686 708
687 BUG_ON(dir == DMA_NONE); 709 BUG_ON(dir == DMA_NONE);
688 710
689 if (is_swiotlb_buffer(paddr)) { 711 if (is_swiotlb_buffer(paddr)) {
690 do_unmap_single(hwdev, phys_to_virt(paddr), size, dir); 712 swiotlb_tbl_unmap_single(hwdev, phys_to_virt(paddr), size, dir);
691 return; 713 return;
692 } 714 }
693 715
@@ -723,14 +745,16 @@ EXPORT_SYMBOL_GPL(swiotlb_unmap_page);
723 */ 745 */
724static void 746static void
725swiotlb_sync_single(struct device *hwdev, dma_addr_t dev_addr, 747swiotlb_sync_single(struct device *hwdev, dma_addr_t dev_addr,
726 size_t size, int dir, int target) 748 size_t size, enum dma_data_direction dir,
749 enum dma_sync_target target)
727{ 750{
728 phys_addr_t paddr = dma_to_phys(hwdev, dev_addr); 751 phys_addr_t paddr = dma_to_phys(hwdev, dev_addr);
729 752
730 BUG_ON(dir == DMA_NONE); 753 BUG_ON(dir == DMA_NONE);
731 754
732 if (is_swiotlb_buffer(paddr)) { 755 if (is_swiotlb_buffer(paddr)) {
733 sync_single(hwdev, phys_to_virt(paddr), size, dir, target); 756 swiotlb_tbl_sync_single(hwdev, phys_to_virt(paddr), size, dir,
757 target);
734 return; 758 return;
735 } 759 }
736 760
@@ -809,7 +833,7 @@ EXPORT_SYMBOL(swiotlb_map_sg_attrs);
809 833
810int 834int
811swiotlb_map_sg(struct device *hwdev, struct scatterlist *sgl, int nelems, 835swiotlb_map_sg(struct device *hwdev, struct scatterlist *sgl, int nelems,
812 int dir) 836 enum dma_data_direction dir)
813{ 837{
814 return swiotlb_map_sg_attrs(hwdev, sgl, nelems, dir, NULL); 838 return swiotlb_map_sg_attrs(hwdev, sgl, nelems, dir, NULL);
815} 839}
@@ -836,7 +860,7 @@ EXPORT_SYMBOL(swiotlb_unmap_sg_attrs);
836 860
837void 861void
838swiotlb_unmap_sg(struct device *hwdev, struct scatterlist *sgl, int nelems, 862swiotlb_unmap_sg(struct device *hwdev, struct scatterlist *sgl, int nelems,
839 int dir) 863 enum dma_data_direction dir)
840{ 864{
841 return swiotlb_unmap_sg_attrs(hwdev, sgl, nelems, dir, NULL); 865 return swiotlb_unmap_sg_attrs(hwdev, sgl, nelems, dir, NULL);
842} 866}
@@ -851,7 +875,8 @@ EXPORT_SYMBOL(swiotlb_unmap_sg);
851 */ 875 */
852static void 876static void
853swiotlb_sync_sg(struct device *hwdev, struct scatterlist *sgl, 877swiotlb_sync_sg(struct device *hwdev, struct scatterlist *sgl,
854 int nelems, int dir, int target) 878 int nelems, enum dma_data_direction dir,
879 enum dma_sync_target target)
855{ 880{
856 struct scatterlist *sg; 881 struct scatterlist *sg;
857 int i; 882 int i;
diff --git a/lib/vsprintf.c b/lib/vsprintf.c
index b8a2f549ab0e..7af9d841c43b 100644
--- a/lib/vsprintf.c
+++ b/lib/vsprintf.c
@@ -146,19 +146,16 @@ int strict_strtoul(const char *cp, unsigned int base, unsigned long *res)
146{ 146{
147 char *tail; 147 char *tail;
148 unsigned long val; 148 unsigned long val;
149 size_t len;
150 149
151 *res = 0; 150 *res = 0;
152 len = strlen(cp); 151 if (!*cp)
153 if (len == 0)
154 return -EINVAL; 152 return -EINVAL;
155 153
156 val = simple_strtoul(cp, &tail, base); 154 val = simple_strtoul(cp, &tail, base);
157 if (tail == cp) 155 if (tail == cp)
158 return -EINVAL; 156 return -EINVAL;
159 157
160 if ((*tail == '\0') || 158 if ((tail[0] == '\0') || (tail[0] == '\n' && tail[1] == '\0')) {
161 ((len == (size_t)(tail - cp) + 1) && (*tail == '\n'))) {
162 *res = val; 159 *res = val;
163 return 0; 160 return 0;
164 } 161 }
@@ -220,18 +217,15 @@ int strict_strtoull(const char *cp, unsigned int base, unsigned long long *res)
220{ 217{
221 char *tail; 218 char *tail;
222 unsigned long long val; 219 unsigned long long val;
223 size_t len;
224 220
225 *res = 0; 221 *res = 0;
226 len = strlen(cp); 222 if (!*cp)
227 if (len == 0)
228 return -EINVAL; 223 return -EINVAL;
229 224
230 val = simple_strtoull(cp, &tail, base); 225 val = simple_strtoull(cp, &tail, base);
231 if (tail == cp) 226 if (tail == cp)
232 return -EINVAL; 227 return -EINVAL;
233 if ((*tail == '\0') || 228 if ((tail[0] == '\0') || (tail[0] == '\n' && tail[1] == '\0')) {
234 ((len == (size_t)(tail - cp) + 1) && (*tail == '\n'))) {
235 *res = val; 229 *res = val;
236 return 0; 230 return 0;
237 } 231 }
@@ -980,6 +974,11 @@ char *uuid_string(char *buf, char *end, const u8 *addr,
980 * [0][1][2][3]-[4][5]-[6][7]-[8][9]-[10][11][12][13][14][15] 974 * [0][1][2][3]-[4][5]-[6][7]-[8][9]-[10][11][12][13][14][15]
981 * little endian output byte order is: 975 * little endian output byte order is:
982 * [3][2][1][0]-[5][4]-[7][6]-[8][9]-[10][11][12][13][14][15] 976 * [3][2][1][0]-[5][4]-[7][6]-[8][9]-[10][11][12][13][14][15]
977 * - 'V' For a struct va_format which contains a format string * and va_list *,
978 * call vsnprintf(->format, *->va_list).
979 * Implements a "recursive vsnprintf".
980 * Do not use this feature without some mechanism to verify the
981 * correctness of the format string and va_list arguments.
983 * 982 *
984 * Note: The difference between 'S' and 'F' is that on ia64 and ppc64 983 * Note: The difference between 'S' and 'F' is that on ia64 and ppc64
985 * function pointers are really function descriptors, which contain a 984 * function pointers are really function descriptors, which contain a
@@ -1025,6 +1024,10 @@ char *pointer(const char *fmt, char *buf, char *end, void *ptr,
1025 break; 1024 break;
1026 case 'U': 1025 case 'U':
1027 return uuid_string(buf, end, ptr, spec, fmt); 1026 return uuid_string(buf, end, ptr, spec, fmt);
1027 case 'V':
1028 return buf + vsnprintf(buf, end - buf,
1029 ((struct va_format *)ptr)->fmt,
1030 *(((struct va_format *)ptr)->va));
1028 } 1031 }
1029 spec.flags |= SMALL; 1032 spec.flags |= SMALL;
1030 if (spec.field_width == -1) { 1033 if (spec.field_width == -1) {
generated by cgit v1.2.2 (git 2.25.0) at 2025-11-11 06:34:34 -0500