aboutsummaryrefslogtreecommitdiffstats
path: root/kernel
diff options
context:
space:
mode:
Diffstat (limited to 'kernel')
-rw-r--r--kernel/audit.c11
-rw-r--r--kernel/audit_tree.c5
-rw-r--r--kernel/capability.c111
-rw-r--r--kernel/cgroup.c2
-rw-r--r--kernel/cpuset.c10
-rw-r--r--kernel/exit.c7
-rw-r--r--kernel/fork.c130
-rw-r--r--kernel/kgdb.c16
-rw-r--r--kernel/kprobes.c15
-rw-r--r--kernel/module.c18
-rw-r--r--kernel/relay.c2
-rw-r--r--kernel/sched.c469
-rw-r--r--kernel/sched_clock.c18
-rw-r--r--kernel/sched_debug.c5
-rw-r--r--kernel/sched_fair.c254
-rw-r--r--kernel/sched_rt.c4
-rw-r--r--kernel/sched_stats.h1
-rw-r--r--kernel/signal.c51
-rw-r--r--kernel/stop_machine.c7
-rw-r--r--kernel/sys.c6
-rw-r--r--kernel/sysctl.c5
21 files changed, 342 insertions, 805 deletions
diff --git a/kernel/audit.c b/kernel/audit.c
index b7d3709cc452..e8692a5748c2 100644
--- a/kernel/audit.c
+++ b/kernel/audit.c
@@ -572,16 +572,17 @@ void audit_send_reply(int pid, int seq, int type, int done, int multi,
572 572
573 skb = audit_make_reply(pid, seq, type, done, multi, payload, size); 573 skb = audit_make_reply(pid, seq, type, done, multi, payload, size);
574 if (!skb) 574 if (!skb)
575 return; 575 goto out;
576 576
577 reply->pid = pid; 577 reply->pid = pid;
578 reply->skb = skb; 578 reply->skb = skb;
579 579
580 tsk = kthread_run(audit_send_reply_thread, reply, "audit_send_reply"); 580 tsk = kthread_run(audit_send_reply_thread, reply, "audit_send_reply");
581 if (IS_ERR(tsk)) { 581 if (!IS_ERR(tsk))
582 kfree(reply); 582 return;
583 kfree_skb(skb); 583 kfree_skb(skb);
584 } 584out:
585 kfree(reply);
585} 586}
586 587
587/* 588/*
diff --git a/kernel/audit_tree.c b/kernel/audit_tree.c
index 9ef5e0aacc3c..f7921a2ecf16 100644
--- a/kernel/audit_tree.c
+++ b/kernel/audit_tree.c
@@ -172,10 +172,9 @@ static void insert_hash(struct audit_chunk *chunk)
172struct audit_chunk *audit_tree_lookup(const struct inode *inode) 172struct audit_chunk *audit_tree_lookup(const struct inode *inode)
173{ 173{
174 struct list_head *list = chunk_hash(inode); 174 struct list_head *list = chunk_hash(inode);
175 struct list_head *pos; 175 struct audit_chunk *p;
176 176
177 list_for_each_rcu(pos, list) { 177 list_for_each_entry_rcu(p, list, hash) {
178 struct audit_chunk *p = container_of(pos, struct audit_chunk, hash);
179 if (p->watch.inode == inode) { 178 if (p->watch.inode == inode) {
180 get_inotify_watch(&p->watch); 179 get_inotify_watch(&p->watch);
181 return p; 180 return p;
diff --git a/kernel/capability.c b/kernel/capability.c
index 39e8193b41ea..cfbe44299488 100644
--- a/kernel/capability.c
+++ b/kernel/capability.c
@@ -53,6 +53,69 @@ static void warn_legacy_capability_use(void)
53} 53}
54 54
55/* 55/*
56 * Version 2 capabilities worked fine, but the linux/capability.h file
57 * that accompanied their introduction encouraged their use without
58 * the necessary user-space source code changes. As such, we have
59 * created a version 3 with equivalent functionality to version 2, but
60 * with a header change to protect legacy source code from using
61 * version 2 when it wanted to use version 1. If your system has code
62 * that trips the following warning, it is using version 2 specific
63 * capabilities and may be doing so insecurely.
64 *
65 * The remedy is to either upgrade your version of libcap (to 2.10+,
66 * if the application is linked against it), or recompile your
67 * application with modern kernel headers and this warning will go
68 * away.
69 */
70
71static void warn_deprecated_v2(void)
72{
73 static int warned;
74
75 if (!warned) {
76 char name[sizeof(current->comm)];
77
78 printk(KERN_INFO "warning: `%s' uses deprecated v2"
79 " capabilities in a way that may be insecure.\n",
80 get_task_comm(name, current));
81 warned = 1;
82 }
83}
84
85/*
86 * Version check. Return the number of u32s in each capability flag
87 * array, or a negative value on error.
88 */
89static int cap_validate_magic(cap_user_header_t header, unsigned *tocopy)
90{
91 __u32 version;
92
93 if (get_user(version, &header->version))
94 return -EFAULT;
95
96 switch (version) {
97 case _LINUX_CAPABILITY_VERSION_1:
98 warn_legacy_capability_use();
99 *tocopy = _LINUX_CAPABILITY_U32S_1;
100 break;
101 case _LINUX_CAPABILITY_VERSION_2:
102 warn_deprecated_v2();
103 /*
104 * fall through - v3 is otherwise equivalent to v2.
105 */
106 case _LINUX_CAPABILITY_VERSION_3:
107 *tocopy = _LINUX_CAPABILITY_U32S_3;
108 break;
109 default:
110 if (put_user((u32)_KERNEL_CAPABILITY_VERSION, &header->version))
111 return -EFAULT;
112 return -EINVAL;
113 }
114
115 return 0;
116}
117
118/*
56 * For sys_getproccap() and sys_setproccap(), any of the three 119 * For sys_getproccap() and sys_setproccap(), any of the three
57 * capability set pointers may be NULL -- indicating that that set is 120 * capability set pointers may be NULL -- indicating that that set is
58 * uninteresting and/or not to be changed. 121 * uninteresting and/or not to be changed.
@@ -71,27 +134,13 @@ asmlinkage long sys_capget(cap_user_header_t header, cap_user_data_t dataptr)
71{ 134{
72 int ret = 0; 135 int ret = 0;
73 pid_t pid; 136 pid_t pid;
74 __u32 version;
75 struct task_struct *target; 137 struct task_struct *target;
76 unsigned tocopy; 138 unsigned tocopy;
77 kernel_cap_t pE, pI, pP; 139 kernel_cap_t pE, pI, pP;
78 140
79 if (get_user(version, &header->version)) 141 ret = cap_validate_magic(header, &tocopy);
80 return -EFAULT; 142 if (ret != 0)
81 143 return ret;
82 switch (version) {
83 case _LINUX_CAPABILITY_VERSION_1:
84 warn_legacy_capability_use();
85 tocopy = _LINUX_CAPABILITY_U32S_1;
86 break;
87 case _LINUX_CAPABILITY_VERSION_2:
88 tocopy = _LINUX_CAPABILITY_U32S_2;
89 break;
90 default:
91 if (put_user(_LINUX_CAPABILITY_VERSION, &header->version))
92 return -EFAULT;
93 return -EINVAL;
94 }
95 144
96 if (get_user(pid, &header->pid)) 145 if (get_user(pid, &header->pid))
97 return -EFAULT; 146 return -EFAULT;
@@ -118,7 +167,7 @@ out:
118 spin_unlock(&task_capability_lock); 167 spin_unlock(&task_capability_lock);
119 168
120 if (!ret) { 169 if (!ret) {
121 struct __user_cap_data_struct kdata[_LINUX_CAPABILITY_U32S]; 170 struct __user_cap_data_struct kdata[_KERNEL_CAPABILITY_U32S];
122 unsigned i; 171 unsigned i;
123 172
124 for (i = 0; i < tocopy; i++) { 173 for (i = 0; i < tocopy; i++) {
@@ -128,7 +177,7 @@ out:
128 } 177 }
129 178
130 /* 179 /*
131 * Note, in the case, tocopy < _LINUX_CAPABILITY_U32S, 180 * Note, in the case, tocopy < _KERNEL_CAPABILITY_U32S,
132 * we silently drop the upper capabilities here. This 181 * we silently drop the upper capabilities here. This
133 * has the effect of making older libcap 182 * has the effect of making older libcap
134 * implementations implicitly drop upper capability 183 * implementations implicitly drop upper capability
@@ -240,30 +289,16 @@ static inline int cap_set_all(kernel_cap_t *effective,
240 */ 289 */
241asmlinkage long sys_capset(cap_user_header_t header, const cap_user_data_t data) 290asmlinkage long sys_capset(cap_user_header_t header, const cap_user_data_t data)
242{ 291{
243 struct __user_cap_data_struct kdata[_LINUX_CAPABILITY_U32S]; 292 struct __user_cap_data_struct kdata[_KERNEL_CAPABILITY_U32S];
244 unsigned i, tocopy; 293 unsigned i, tocopy;
245 kernel_cap_t inheritable, permitted, effective; 294 kernel_cap_t inheritable, permitted, effective;
246 __u32 version;
247 struct task_struct *target; 295 struct task_struct *target;
248 int ret; 296 int ret;
249 pid_t pid; 297 pid_t pid;
250 298
251 if (get_user(version, &header->version)) 299 ret = cap_validate_magic(header, &tocopy);
252 return -EFAULT; 300 if (ret != 0)
253 301 return ret;
254 switch (version) {
255 case _LINUX_CAPABILITY_VERSION_1:
256 warn_legacy_capability_use();
257 tocopy = _LINUX_CAPABILITY_U32S_1;
258 break;
259 case _LINUX_CAPABILITY_VERSION_2:
260 tocopy = _LINUX_CAPABILITY_U32S_2;
261 break;
262 default:
263 if (put_user(_LINUX_CAPABILITY_VERSION, &header->version))
264 return -EFAULT;
265 return -EINVAL;
266 }
267 302
268 if (get_user(pid, &header->pid)) 303 if (get_user(pid, &header->pid))
269 return -EFAULT; 304 return -EFAULT;
@@ -281,7 +316,7 @@ asmlinkage long sys_capset(cap_user_header_t header, const cap_user_data_t data)
281 permitted.cap[i] = kdata[i].permitted; 316 permitted.cap[i] = kdata[i].permitted;
282 inheritable.cap[i] = kdata[i].inheritable; 317 inheritable.cap[i] = kdata[i].inheritable;
283 } 318 }
284 while (i < _LINUX_CAPABILITY_U32S) { 319 while (i < _KERNEL_CAPABILITY_U32S) {
285 effective.cap[i] = 0; 320 effective.cap[i] = 0;
286 permitted.cap[i] = 0; 321 permitted.cap[i] = 0;
287 inheritable.cap[i] = 0; 322 inheritable.cap[i] = 0;
diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index fbc6fc8949b4..15ac0e1e4f4d 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -2903,7 +2903,7 @@ int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys)
2903 cg = tsk->cgroups; 2903 cg = tsk->cgroups;
2904 parent = task_cgroup(tsk, subsys->subsys_id); 2904 parent = task_cgroup(tsk, subsys->subsys_id);
2905 2905
2906 snprintf(nodename, MAX_CGROUP_TYPE_NAMELEN, "node_%d", tsk->pid); 2906 snprintf(nodename, MAX_CGROUP_TYPE_NAMELEN, "%d", tsk->pid);
2907 2907
2908 /* Pin the hierarchy */ 2908 /* Pin the hierarchy */
2909 atomic_inc(&parent->root->sb->s_active); 2909 atomic_inc(&parent->root->sb->s_active);
diff --git a/kernel/cpuset.c b/kernel/cpuset.c
index 86ea9e34e326..039baa4cd90c 100644
--- a/kernel/cpuset.c
+++ b/kernel/cpuset.c
@@ -797,8 +797,10 @@ static int update_cpumask(struct cpuset *cs, char *buf)
797 retval = cpulist_parse(buf, trialcs.cpus_allowed); 797 retval = cpulist_parse(buf, trialcs.cpus_allowed);
798 if (retval < 0) 798 if (retval < 0)
799 return retval; 799 return retval;
800
801 if (!cpus_subset(trialcs.cpus_allowed, cpu_online_map))
802 return -EINVAL;
800 } 803 }
801 cpus_and(trialcs.cpus_allowed, trialcs.cpus_allowed, cpu_online_map);
802 retval = validate_change(cs, &trialcs); 804 retval = validate_change(cs, &trialcs);
803 if (retval < 0) 805 if (retval < 0)
804 return retval; 806 return retval;
@@ -932,9 +934,11 @@ static int update_nodemask(struct cpuset *cs, char *buf)
932 retval = nodelist_parse(buf, trialcs.mems_allowed); 934 retval = nodelist_parse(buf, trialcs.mems_allowed);
933 if (retval < 0) 935 if (retval < 0)
934 goto done; 936 goto done;
937
938 if (!nodes_subset(trialcs.mems_allowed,
939 node_states[N_HIGH_MEMORY]))
940 return -EINVAL;
935 } 941 }
936 nodes_and(trialcs.mems_allowed, trialcs.mems_allowed,
937 node_states[N_HIGH_MEMORY]);
938 oldmem = cs->mems_allowed; 942 oldmem = cs->mems_allowed;
939 if (nodes_equal(oldmem, trialcs.mems_allowed)) { 943 if (nodes_equal(oldmem, trialcs.mems_allowed)) {
940 retval = 0; /* Too easy - nothing to do */ 944 retval = 0; /* Too easy - nothing to do */
diff --git a/kernel/exit.c b/kernel/exit.c
index 1510f78a0ffa..8f6185e69b69 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -126,6 +126,12 @@ static void __exit_signal(struct task_struct *tsk)
126 126
127 __unhash_process(tsk); 127 __unhash_process(tsk);
128 128
129 /*
130 * Do this under ->siglock, we can race with another thread
131 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
132 */
133 flush_sigqueue(&tsk->pending);
134
129 tsk->signal = NULL; 135 tsk->signal = NULL;
130 tsk->sighand = NULL; 136 tsk->sighand = NULL;
131 spin_unlock(&sighand->siglock); 137 spin_unlock(&sighand->siglock);
@@ -133,7 +139,6 @@ static void __exit_signal(struct task_struct *tsk)
133 139
134 __cleanup_sighand(sighand); 140 __cleanup_sighand(sighand);
135 clear_tsk_thread_flag(tsk,TIF_SIGPENDING); 141 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
136 flush_sigqueue(&tsk->pending);
137 if (sig) { 142 if (sig) {
138 flush_sigqueue(&sig->shared_pending); 143 flush_sigqueue(&sig->shared_pending);
139 taskstats_tgid_free(sig); 144 taskstats_tgid_free(sig);
diff --git a/kernel/fork.c b/kernel/fork.c
index 933e60ebccae..19908b26cf80 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -660,136 +660,6 @@ static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
660 return 0; 660 return 0;
661} 661}
662 662
663static int count_open_files(struct fdtable *fdt)
664{
665 int size = fdt->max_fds;
666 int i;
667
668 /* Find the last open fd */
669 for (i = size/(8*sizeof(long)); i > 0; ) {
670 if (fdt->open_fds->fds_bits[--i])
671 break;
672 }
673 i = (i+1) * 8 * sizeof(long);
674 return i;
675}
676
677static struct files_struct *alloc_files(void)
678{
679 struct files_struct *newf;
680 struct fdtable *fdt;
681
682 newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
683 if (!newf)
684 goto out;
685
686 atomic_set(&newf->count, 1);
687
688 spin_lock_init(&newf->file_lock);
689 newf->next_fd = 0;
690 fdt = &newf->fdtab;
691 fdt->max_fds = NR_OPEN_DEFAULT;
692 fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
693 fdt->open_fds = (fd_set *)&newf->open_fds_init;
694 fdt->fd = &newf->fd_array[0];
695 INIT_RCU_HEAD(&fdt->rcu);
696 fdt->next = NULL;
697 rcu_assign_pointer(newf->fdt, fdt);
698out:
699 return newf;
700}
701
702/*
703 * Allocate a new files structure and copy contents from the
704 * passed in files structure.
705 * errorp will be valid only when the returned files_struct is NULL.
706 */
707static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
708{
709 struct files_struct *newf;
710 struct file **old_fds, **new_fds;
711 int open_files, size, i;
712 struct fdtable *old_fdt, *new_fdt;
713
714 *errorp = -ENOMEM;
715 newf = alloc_files();
716 if (!newf)
717 goto out;
718
719 spin_lock(&oldf->file_lock);
720 old_fdt = files_fdtable(oldf);
721 new_fdt = files_fdtable(newf);
722 open_files = count_open_files(old_fdt);
723
724 /*
725 * Check whether we need to allocate a larger fd array and fd set.
726 * Note: we're not a clone task, so the open count won't change.
727 */
728 if (open_files > new_fdt->max_fds) {
729 new_fdt->max_fds = 0;
730 spin_unlock(&oldf->file_lock);
731 spin_lock(&newf->file_lock);
732 *errorp = expand_files(newf, open_files-1);
733 spin_unlock(&newf->file_lock);
734 if (*errorp < 0)
735 goto out_release;
736 new_fdt = files_fdtable(newf);
737 /*
738 * Reacquire the oldf lock and a pointer to its fd table
739 * who knows it may have a new bigger fd table. We need
740 * the latest pointer.
741 */
742 spin_lock(&oldf->file_lock);
743 old_fdt = files_fdtable(oldf);
744 }
745
746 old_fds = old_fdt->fd;
747 new_fds = new_fdt->fd;
748
749 memcpy(new_fdt->open_fds->fds_bits,
750 old_fdt->open_fds->fds_bits, open_files/8);
751 memcpy(new_fdt->close_on_exec->fds_bits,
752 old_fdt->close_on_exec->fds_bits, open_files/8);
753
754 for (i = open_files; i != 0; i--) {
755 struct file *f = *old_fds++;
756 if (f) {
757 get_file(f);
758 } else {
759 /*
760 * The fd may be claimed in the fd bitmap but not yet
761 * instantiated in the files array if a sibling thread
762 * is partway through open(). So make sure that this
763 * fd is available to the new process.
764 */
765 FD_CLR(open_files - i, new_fdt->open_fds);
766 }
767 rcu_assign_pointer(*new_fds++, f);
768 }
769 spin_unlock(&oldf->file_lock);
770
771 /* compute the remainder to be cleared */
772 size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
773
774 /* This is long word aligned thus could use a optimized version */
775 memset(new_fds, 0, size);
776
777 if (new_fdt->max_fds > open_files) {
778 int left = (new_fdt->max_fds-open_files)/8;
779 int start = open_files / (8 * sizeof(unsigned long));
780
781 memset(&new_fdt->open_fds->fds_bits[start], 0, left);
782 memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
783 }
784
785 return newf;
786
787out_release:
788 kmem_cache_free(files_cachep, newf);
789out:
790 return NULL;
791}
792
793static int copy_files(unsigned long clone_flags, struct task_struct * tsk) 663static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
794{ 664{
795 struct files_struct *oldf, *newf; 665 struct files_struct *oldf, *newf;
diff --git a/kernel/kgdb.c b/kernel/kgdb.c
index 14787de568b3..79e3c90113c2 100644
--- a/kernel/kgdb.c
+++ b/kernel/kgdb.c
@@ -52,6 +52,7 @@
52#include <asm/byteorder.h> 52#include <asm/byteorder.h>
53#include <asm/atomic.h> 53#include <asm/atomic.h>
54#include <asm/system.h> 54#include <asm/system.h>
55#include <asm/unaligned.h>
55 56
56static int kgdb_break_asap; 57static int kgdb_break_asap;
57 58
@@ -227,8 +228,6 @@ void __weak kgdb_disable_hw_debug(struct pt_regs *regs)
227 * GDB remote protocol parser: 228 * GDB remote protocol parser:
228 */ 229 */
229 230
230static const char hexchars[] = "0123456789abcdef";
231
232static int hex(char ch) 231static int hex(char ch)
233{ 232{
234 if ((ch >= 'a') && (ch <= 'f')) 233 if ((ch >= 'a') && (ch <= 'f'))
@@ -316,8 +315,8 @@ static void put_packet(char *buffer)
316 } 315 }
317 316
318 kgdb_io_ops->write_char('#'); 317 kgdb_io_ops->write_char('#');
319 kgdb_io_ops->write_char(hexchars[checksum >> 4]); 318 kgdb_io_ops->write_char(hex_asc_hi(checksum));
320 kgdb_io_ops->write_char(hexchars[checksum & 0xf]); 319 kgdb_io_ops->write_char(hex_asc_lo(checksum));
321 if (kgdb_io_ops->flush) 320 if (kgdb_io_ops->flush)
322 kgdb_io_ops->flush(); 321 kgdb_io_ops->flush();
323 322
@@ -478,8 +477,8 @@ static void error_packet(char *pkt, int error)
478{ 477{
479 error = -error; 478 error = -error;
480 pkt[0] = 'E'; 479 pkt[0] = 'E';
481 pkt[1] = hexchars[(error / 10)]; 480 pkt[1] = hex_asc[(error / 10)];
482 pkt[2] = hexchars[(error % 10)]; 481 pkt[2] = hex_asc[(error % 10)];
483 pkt[3] = '\0'; 482 pkt[3] = '\0';
484} 483}
485 484
@@ -510,10 +509,7 @@ static void int_to_threadref(unsigned char *id, int value)
510 scan = (unsigned char *)id; 509 scan = (unsigned char *)id;
511 while (i--) 510 while (i--)
512 *scan++ = 0; 511 *scan++ = 0;
513 *scan++ = (value >> 24) & 0xff; 512 put_unaligned_be32(value, scan);
514 *scan++ = (value >> 16) & 0xff;
515 *scan++ = (value >> 8) & 0xff;
516 *scan++ = (value & 0xff);
517} 513}
518 514
519static struct task_struct *getthread(struct pt_regs *regs, int tid) 515static struct task_struct *getthread(struct pt_regs *regs, int tid)
diff --git a/kernel/kprobes.c b/kernel/kprobes.c
index 1e0250cb9486..d4998f81e229 100644
--- a/kernel/kprobes.c
+++ b/kernel/kprobes.c
@@ -699,8 +699,9 @@ static int __register_kprobes(struct kprobe **kps, int num,
699 return -EINVAL; 699 return -EINVAL;
700 for (i = 0; i < num; i++) { 700 for (i = 0; i < num; i++) {
701 ret = __register_kprobe(kps[i], called_from); 701 ret = __register_kprobe(kps[i], called_from);
702 if (ret < 0 && i > 0) { 702 if (ret < 0) {
703 unregister_kprobes(kps, i); 703 if (i > 0)
704 unregister_kprobes(kps, i);
704 break; 705 break;
705 } 706 }
706 } 707 }
@@ -776,8 +777,9 @@ static int __register_jprobes(struct jprobe **jps, int num,
776 jp->kp.break_handler = longjmp_break_handler; 777 jp->kp.break_handler = longjmp_break_handler;
777 ret = __register_kprobe(&jp->kp, called_from); 778 ret = __register_kprobe(&jp->kp, called_from);
778 } 779 }
779 if (ret < 0 && i > 0) { 780 if (ret < 0) {
780 unregister_jprobes(jps, i); 781 if (i > 0)
782 unregister_jprobes(jps, i);
781 break; 783 break;
782 } 784 }
783 } 785 }
@@ -920,8 +922,9 @@ static int __register_kretprobes(struct kretprobe **rps, int num,
920 return -EINVAL; 922 return -EINVAL;
921 for (i = 0; i < num; i++) { 923 for (i = 0; i < num; i++) {
922 ret = __register_kretprobe(rps[i], called_from); 924 ret = __register_kretprobe(rps[i], called_from);
923 if (ret < 0 && i > 0) { 925 if (ret < 0) {
924 unregister_kretprobes(rps, i); 926 if (i > 0)
927 unregister_kretprobes(rps, i);
925 break; 928 break;
926 } 929 }
927 } 930 }
diff --git a/kernel/module.c b/kernel/module.c
index f5e9491ef7ac..5f80478b746d 100644
--- a/kernel/module.c
+++ b/kernel/module.c
@@ -1337,7 +1337,19 @@ out_unreg:
1337 kobject_put(&mod->mkobj.kobj); 1337 kobject_put(&mod->mkobj.kobj);
1338 return err; 1338 return err;
1339} 1339}
1340#endif 1340
1341static void mod_sysfs_fini(struct module *mod)
1342{
1343 kobject_put(&mod->mkobj.kobj);
1344}
1345
1346#else /* CONFIG_SYSFS */
1347
1348static void mod_sysfs_fini(struct module *mod)
1349{
1350}
1351
1352#endif /* CONFIG_SYSFS */
1341 1353
1342static void mod_kobject_remove(struct module *mod) 1354static void mod_kobject_remove(struct module *mod)
1343{ 1355{
@@ -1345,7 +1357,7 @@ static void mod_kobject_remove(struct module *mod)
1345 module_param_sysfs_remove(mod); 1357 module_param_sysfs_remove(mod);
1346 kobject_put(mod->mkobj.drivers_dir); 1358 kobject_put(mod->mkobj.drivers_dir);
1347 kobject_put(mod->holders_dir); 1359 kobject_put(mod->holders_dir);
1348 kobject_put(&mod->mkobj.kobj); 1360 mod_sysfs_fini(mod);
1349} 1361}
1350 1362
1351/* 1363/*
@@ -1780,7 +1792,7 @@ static struct module *load_module(void __user *umod,
1780 1792
1781 /* Sanity checks against insmoding binaries or wrong arch, 1793 /* Sanity checks against insmoding binaries or wrong arch,
1782 weird elf version */ 1794 weird elf version */
1783 if (memcmp(hdr->e_ident, ELFMAG, 4) != 0 1795 if (memcmp(hdr->e_ident, ELFMAG, SELFMAG) != 0
1784 || hdr->e_type != ET_REL 1796 || hdr->e_type != ET_REL
1785 || !elf_check_arch(hdr) 1797 || !elf_check_arch(hdr)
1786 || hdr->e_shentsize != sizeof(*sechdrs)) { 1798 || hdr->e_shentsize != sizeof(*sechdrs)) {
diff --git a/kernel/relay.c b/kernel/relay.c
index bc24dcdc570f..7de644cdec43 100644
--- a/kernel/relay.c
+++ b/kernel/relay.c
@@ -1191,7 +1191,7 @@ static ssize_t relay_file_splice_read(struct file *in,
1191 ret = 0; 1191 ret = 0;
1192 spliced = 0; 1192 spliced = 0;
1193 1193
1194 while (len) { 1194 while (len && !spliced) {
1195 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret); 1195 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1196 if (ret < 0) 1196 if (ret < 0)
1197 break; 1197 break;
diff --git a/kernel/sched.c b/kernel/sched.c
index cfa222a91539..eaf6751e7612 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -136,7 +136,7 @@ static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val)
136 136
137static inline int rt_policy(int policy) 137static inline int rt_policy(int policy)
138{ 138{
139 if (unlikely(policy == SCHED_FIFO) || unlikely(policy == SCHED_RR)) 139 if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR))
140 return 1; 140 return 1;
141 return 0; 141 return 0;
142} 142}
@@ -312,12 +312,15 @@ static DEFINE_SPINLOCK(task_group_lock);
312#endif 312#endif
313 313
314/* 314/*
315 * A weight of 0, 1 or ULONG_MAX can cause arithmetics problems. 315 * A weight of 0 or 1 can cause arithmetics problems.
316 * A weight of a cfs_rq is the sum of weights of which entities
317 * are queued on this cfs_rq, so a weight of a entity should not be
318 * too large, so as the shares value of a task group.
316 * (The default weight is 1024 - so there's no practical 319 * (The default weight is 1024 - so there's no practical
317 * limitation from this.) 320 * limitation from this.)
318 */ 321 */
319#define MIN_SHARES 2 322#define MIN_SHARES 2
320#define MAX_SHARES (ULONG_MAX - 1) 323#define MAX_SHARES (1UL << 18)
321 324
322static int init_task_group_load = INIT_TASK_GROUP_LOAD; 325static int init_task_group_load = INIT_TASK_GROUP_LOAD;
323#endif 326#endif
@@ -398,43 +401,6 @@ struct cfs_rq {
398 */ 401 */
399 struct list_head leaf_cfs_rq_list; 402 struct list_head leaf_cfs_rq_list;
400 struct task_group *tg; /* group that "owns" this runqueue */ 403 struct task_group *tg; /* group that "owns" this runqueue */
401
402#ifdef CONFIG_SMP
403 unsigned long task_weight;
404 unsigned long shares;
405 /*
406 * We need space to build a sched_domain wide view of the full task
407 * group tree, in order to avoid depending on dynamic memory allocation
408 * during the load balancing we place this in the per cpu task group
409 * hierarchy. This limits the load balancing to one instance per cpu,
410 * but more should not be needed anyway.
411 */
412 struct aggregate_struct {
413 /*
414 * load = weight(cpus) * f(tg)
415 *
416 * Where f(tg) is the recursive weight fraction assigned to
417 * this group.
418 */
419 unsigned long load;
420
421 /*
422 * part of the group weight distributed to this span.
423 */
424 unsigned long shares;
425
426 /*
427 * The sum of all runqueue weights within this span.
428 */
429 unsigned long rq_weight;
430
431 /*
432 * Weight contributed by tasks; this is the part we can
433 * influence by moving tasks around.
434 */
435 unsigned long task_weight;
436 } aggregate;
437#endif
438#endif 404#endif
439}; 405};
440 406
@@ -1368,17 +1334,19 @@ static void __resched_task(struct task_struct *p, int tif_bit)
1368 */ 1334 */
1369#define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) 1335#define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y))
1370 1336
1371/*
1372 * delta *= weight / lw
1373 */
1374static unsigned long 1337static unsigned long
1375calc_delta_mine(unsigned long delta_exec, unsigned long weight, 1338calc_delta_mine(unsigned long delta_exec, unsigned long weight,
1376 struct load_weight *lw) 1339 struct load_weight *lw)
1377{ 1340{
1378 u64 tmp; 1341 u64 tmp;
1379 1342
1380 if (!lw->inv_weight) 1343 if (!lw->inv_weight) {
1381 lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2)/(lw->weight+1); 1344 if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST))
1345 lw->inv_weight = 1;
1346 else
1347 lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2)
1348 / (lw->weight+1);
1349 }
1382 1350
1383 tmp = (u64)delta_exec * weight; 1351 tmp = (u64)delta_exec * weight;
1384 /* 1352 /*
@@ -1393,6 +1361,12 @@ calc_delta_mine(unsigned long delta_exec, unsigned long weight,
1393 return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); 1361 return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX);
1394} 1362}
1395 1363
1364static inline unsigned long
1365calc_delta_fair(unsigned long delta_exec, struct load_weight *lw)
1366{
1367 return calc_delta_mine(delta_exec, NICE_0_LOAD, lw);
1368}
1369
1396static inline void update_load_add(struct load_weight *lw, unsigned long inc) 1370static inline void update_load_add(struct load_weight *lw, unsigned long inc)
1397{ 1371{
1398 lw->weight += inc; 1372 lw->weight += inc;
@@ -1505,326 +1479,6 @@ static unsigned long source_load(int cpu, int type);
1505static unsigned long target_load(int cpu, int type); 1479static unsigned long target_load(int cpu, int type);
1506static unsigned long cpu_avg_load_per_task(int cpu); 1480static unsigned long cpu_avg_load_per_task(int cpu);
1507static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); 1481static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd);
1508
1509#ifdef CONFIG_FAIR_GROUP_SCHED
1510
1511/*
1512 * Group load balancing.
1513 *
1514 * We calculate a few balance domain wide aggregate numbers; load and weight.
1515 * Given the pictures below, and assuming each item has equal weight:
1516 *
1517 * root 1 - thread
1518 * / | \ A - group
1519 * A 1 B
1520 * /|\ / \
1521 * C 2 D 3 4
1522 * | |
1523 * 5 6
1524 *
1525 * load:
1526 * A and B get 1/3-rd of the total load. C and D get 1/3-rd of A's 1/3-rd,
1527 * which equals 1/9-th of the total load.
1528 *
1529 * shares:
1530 * The weight of this group on the selected cpus.
1531 *
1532 * rq_weight:
1533 * Direct sum of all the cpu's their rq weight, e.g. A would get 3 while
1534 * B would get 2.
1535 *
1536 * task_weight:
1537 * Part of the rq_weight contributed by tasks; all groups except B would
1538 * get 1, B gets 2.
1539 */
1540
1541static inline struct aggregate_struct *
1542aggregate(struct task_group *tg, struct sched_domain *sd)
1543{
1544 return &tg->cfs_rq[sd->first_cpu]->aggregate;
1545}
1546
1547typedef void (*aggregate_func)(struct task_group *, struct sched_domain *);
1548
1549/*
1550 * Iterate the full tree, calling @down when first entering a node and @up when
1551 * leaving it for the final time.
1552 */
1553static
1554void aggregate_walk_tree(aggregate_func down, aggregate_func up,
1555 struct sched_domain *sd)
1556{
1557 struct task_group *parent, *child;
1558
1559 rcu_read_lock();
1560 parent = &root_task_group;
1561down:
1562 (*down)(parent, sd);
1563 list_for_each_entry_rcu(child, &parent->children, siblings) {
1564 parent = child;
1565 goto down;
1566
1567up:
1568 continue;
1569 }
1570 (*up)(parent, sd);
1571
1572 child = parent;
1573 parent = parent->parent;
1574 if (parent)
1575 goto up;
1576 rcu_read_unlock();
1577}
1578
1579/*
1580 * Calculate the aggregate runqueue weight.
1581 */
1582static
1583void aggregate_group_weight(struct task_group *tg, struct sched_domain *sd)
1584{
1585 unsigned long rq_weight = 0;
1586 unsigned long task_weight = 0;
1587 int i;
1588
1589 for_each_cpu_mask(i, sd->span) {
1590 rq_weight += tg->cfs_rq[i]->load.weight;
1591 task_weight += tg->cfs_rq[i]->task_weight;
1592 }
1593
1594 aggregate(tg, sd)->rq_weight = rq_weight;
1595 aggregate(tg, sd)->task_weight = task_weight;
1596}
1597
1598/*
1599 * Compute the weight of this group on the given cpus.
1600 */
1601static
1602void aggregate_group_shares(struct task_group *tg, struct sched_domain *sd)
1603{
1604 unsigned long shares = 0;
1605 int i;
1606
1607 for_each_cpu_mask(i, sd->span)
1608 shares += tg->cfs_rq[i]->shares;
1609
1610 if ((!shares && aggregate(tg, sd)->rq_weight) || shares > tg->shares)
1611 shares = tg->shares;
1612
1613 aggregate(tg, sd)->shares = shares;
1614}
1615
1616/*
1617 * Compute the load fraction assigned to this group, relies on the aggregate
1618 * weight and this group's parent's load, i.e. top-down.
1619 */
1620static
1621void aggregate_group_load(struct task_group *tg, struct sched_domain *sd)
1622{
1623 unsigned long load;
1624
1625 if (!tg->parent) {
1626 int i;
1627
1628 load = 0;
1629 for_each_cpu_mask(i, sd->span)
1630 load += cpu_rq(i)->load.weight;
1631
1632 } else {
1633 load = aggregate(tg->parent, sd)->load;
1634
1635 /*
1636 * shares is our weight in the parent's rq so
1637 * shares/parent->rq_weight gives our fraction of the load
1638 */
1639 load *= aggregate(tg, sd)->shares;
1640 load /= aggregate(tg->parent, sd)->rq_weight + 1;
1641 }
1642
1643 aggregate(tg, sd)->load = load;
1644}
1645
1646static void __set_se_shares(struct sched_entity *se, unsigned long shares);
1647
1648/*
1649 * Calculate and set the cpu's group shares.
1650 */
1651static void
1652__update_group_shares_cpu(struct task_group *tg, struct sched_domain *sd,
1653 int tcpu)
1654{
1655 int boost = 0;
1656 unsigned long shares;
1657 unsigned long rq_weight;
1658
1659 if (!tg->se[tcpu])
1660 return;
1661
1662 rq_weight = tg->cfs_rq[tcpu]->load.weight;
1663
1664 /*
1665 * If there are currently no tasks on the cpu pretend there is one of
1666 * average load so that when a new task gets to run here it will not
1667 * get delayed by group starvation.
1668 */
1669 if (!rq_weight) {
1670 boost = 1;
1671 rq_weight = NICE_0_LOAD;
1672 }
1673
1674 /*
1675 * \Sum shares * rq_weight
1676 * shares = -----------------------
1677 * \Sum rq_weight
1678 *
1679 */
1680 shares = aggregate(tg, sd)->shares * rq_weight;
1681 shares /= aggregate(tg, sd)->rq_weight + 1;
1682
1683 /*
1684 * record the actual number of shares, not the boosted amount.
1685 */
1686 tg->cfs_rq[tcpu]->shares = boost ? 0 : shares;
1687
1688 if (shares < MIN_SHARES)
1689 shares = MIN_SHARES;
1690 else if (shares > MAX_SHARES)
1691 shares = MAX_SHARES;
1692
1693 __set_se_shares(tg->se[tcpu], shares);
1694}
1695
1696/*
1697 * Re-adjust the weights on the cpu the task came from and on the cpu the
1698 * task went to.
1699 */
1700static void
1701__move_group_shares(struct task_group *tg, struct sched_domain *sd,
1702 int scpu, int dcpu)
1703{
1704 unsigned long shares;
1705
1706 shares = tg->cfs_rq[scpu]->shares + tg->cfs_rq[dcpu]->shares;
1707
1708 __update_group_shares_cpu(tg, sd, scpu);
1709 __update_group_shares_cpu(tg, sd, dcpu);
1710
1711 /*
1712 * ensure we never loose shares due to rounding errors in the
1713 * above redistribution.
1714 */
1715 shares -= tg->cfs_rq[scpu]->shares + tg->cfs_rq[dcpu]->shares;
1716 if (shares)
1717 tg->cfs_rq[dcpu]->shares += shares;
1718}
1719
1720/*
1721 * Because changing a group's shares changes the weight of the super-group
1722 * we need to walk up the tree and change all shares until we hit the root.
1723 */
1724static void
1725move_group_shares(struct task_group *tg, struct sched_domain *sd,
1726 int scpu, int dcpu)
1727{
1728 while (tg) {
1729 __move_group_shares(tg, sd, scpu, dcpu);
1730 tg = tg->parent;
1731 }
1732}
1733
1734static
1735void aggregate_group_set_shares(struct task_group *tg, struct sched_domain *sd)
1736{
1737 unsigned long shares = aggregate(tg, sd)->shares;
1738 int i;
1739
1740 for_each_cpu_mask(i, sd->span) {
1741 struct rq *rq = cpu_rq(i);
1742 unsigned long flags;
1743
1744 spin_lock_irqsave(&rq->lock, flags);
1745 __update_group_shares_cpu(tg, sd, i);
1746 spin_unlock_irqrestore(&rq->lock, flags);
1747 }
1748
1749 aggregate_group_shares(tg, sd);
1750
1751 /*
1752 * ensure we never loose shares due to rounding errors in the
1753 * above redistribution.
1754 */
1755 shares -= aggregate(tg, sd)->shares;
1756 if (shares) {
1757 tg->cfs_rq[sd->first_cpu]->shares += shares;
1758 aggregate(tg, sd)->shares += shares;
1759 }
1760}
1761
1762/*
1763 * Calculate the accumulative weight and recursive load of each task group
1764 * while walking down the tree.
1765 */
1766static
1767void aggregate_get_down(struct task_group *tg, struct sched_domain *sd)
1768{
1769 aggregate_group_weight(tg, sd);
1770 aggregate_group_shares(tg, sd);
1771 aggregate_group_load(tg, sd);
1772}
1773
1774/*
1775 * Rebalance the cpu shares while walking back up the tree.
1776 */
1777static
1778void aggregate_get_up(struct task_group *tg, struct sched_domain *sd)
1779{
1780 aggregate_group_set_shares(tg, sd);
1781}
1782
1783static DEFINE_PER_CPU(spinlock_t, aggregate_lock);
1784
1785static void __init init_aggregate(void)
1786{
1787 int i;
1788
1789 for_each_possible_cpu(i)
1790 spin_lock_init(&per_cpu(aggregate_lock, i));
1791}
1792
1793static int get_aggregate(struct sched_domain *sd)
1794{
1795 if (!spin_trylock(&per_cpu(aggregate_lock, sd->first_cpu)))
1796 return 0;
1797
1798 aggregate_walk_tree(aggregate_get_down, aggregate_get_up, sd);
1799 return 1;
1800}
1801
1802static void put_aggregate(struct sched_domain *sd)
1803{
1804 spin_unlock(&per_cpu(aggregate_lock, sd->first_cpu));
1805}
1806
1807static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares)
1808{
1809 cfs_rq->shares = shares;
1810}
1811
1812#else
1813
1814static inline void init_aggregate(void)
1815{
1816}
1817
1818static inline int get_aggregate(struct sched_domain *sd)
1819{
1820 return 0;
1821}
1822
1823static inline void put_aggregate(struct sched_domain *sd)
1824{
1825}
1826#endif
1827
1828#else /* CONFIG_SMP */ 1482#else /* CONFIG_SMP */
1829 1483
1830#ifdef CONFIG_FAIR_GROUP_SCHED 1484#ifdef CONFIG_FAIR_GROUP_SCHED
@@ -1845,14 +1499,26 @@ static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares)
1845 1499
1846#define sched_class_highest (&rt_sched_class) 1500#define sched_class_highest (&rt_sched_class)
1847 1501
1848static void inc_nr_running(struct rq *rq) 1502static inline void inc_load(struct rq *rq, const struct task_struct *p)
1503{
1504 update_load_add(&rq->load, p->se.load.weight);
1505}
1506
1507static inline void dec_load(struct rq *rq, const struct task_struct *p)
1508{
1509 update_load_sub(&rq->load, p->se.load.weight);
1510}
1511
1512static void inc_nr_running(struct task_struct *p, struct rq *rq)
1849{ 1513{
1850 rq->nr_running++; 1514 rq->nr_running++;
1515 inc_load(rq, p);
1851} 1516}
1852 1517
1853static void dec_nr_running(struct rq *rq) 1518static void dec_nr_running(struct task_struct *p, struct rq *rq)
1854{ 1519{
1855 rq->nr_running--; 1520 rq->nr_running--;
1521 dec_load(rq, p);
1856} 1522}
1857 1523
1858static void set_load_weight(struct task_struct *p) 1524static void set_load_weight(struct task_struct *p)
@@ -1944,7 +1610,7 @@ static void activate_task(struct rq *rq, struct task_struct *p, int wakeup)
1944 rq->nr_uninterruptible--; 1610 rq->nr_uninterruptible--;
1945 1611
1946 enqueue_task(rq, p, wakeup); 1612 enqueue_task(rq, p, wakeup);
1947 inc_nr_running(rq); 1613 inc_nr_running(p, rq);
1948} 1614}
1949 1615
1950/* 1616/*
@@ -1956,7 +1622,7 @@ static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep)
1956 rq->nr_uninterruptible++; 1622 rq->nr_uninterruptible++;
1957 1623
1958 dequeue_task(rq, p, sleep); 1624 dequeue_task(rq, p, sleep);
1959 dec_nr_running(rq); 1625 dec_nr_running(p, rq);
1960} 1626}
1961 1627
1962/** 1628/**
@@ -2609,7 +2275,7 @@ void wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
2609 * management (if any): 2275 * management (if any):
2610 */ 2276 */
2611 p->sched_class->task_new(rq, p); 2277 p->sched_class->task_new(rq, p);
2612 inc_nr_running(rq); 2278 inc_nr_running(p, rq);
2613 } 2279 }
2614 check_preempt_curr(rq, p); 2280 check_preempt_curr(rq, p);
2615#ifdef CONFIG_SMP 2281#ifdef CONFIG_SMP
@@ -3600,12 +3266,9 @@ static int load_balance(int this_cpu, struct rq *this_rq,
3600 unsigned long imbalance; 3266 unsigned long imbalance;
3601 struct rq *busiest; 3267 struct rq *busiest;
3602 unsigned long flags; 3268 unsigned long flags;
3603 int unlock_aggregate;
3604 3269
3605 cpus_setall(*cpus); 3270 cpus_setall(*cpus);
3606 3271
3607 unlock_aggregate = get_aggregate(sd);
3608
3609 /* 3272 /*
3610 * When power savings policy is enabled for the parent domain, idle 3273 * When power savings policy is enabled for the parent domain, idle
3611 * sibling can pick up load irrespective of busy siblings. In this case, 3274 * sibling can pick up load irrespective of busy siblings. In this case,
@@ -3721,9 +3384,8 @@ redo:
3721 3384
3722 if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && 3385 if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
3723 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) 3386 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
3724 ld_moved = -1; 3387 return -1;
3725 3388 return ld_moved;
3726 goto out;
3727 3389
3728out_balanced: 3390out_balanced:
3729 schedstat_inc(sd, lb_balanced[idle]); 3391 schedstat_inc(sd, lb_balanced[idle]);
@@ -3738,13 +3400,8 @@ out_one_pinned:
3738 3400
3739 if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && 3401 if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER &&
3740 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) 3402 !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE))
3741 ld_moved = -1; 3403 return -1;
3742 else 3404 return 0;
3743 ld_moved = 0;
3744out:
3745 if (unlock_aggregate)
3746 put_aggregate(sd);
3747 return ld_moved;
3748} 3405}
3749 3406
3750/* 3407/*
@@ -4430,7 +4087,7 @@ static inline void schedule_debug(struct task_struct *prev)
4430 * schedule() atomically, we ignore that path for now. 4087 * schedule() atomically, we ignore that path for now.
4431 * Otherwise, whine if we are scheduling when we should not be. 4088 * Otherwise, whine if we are scheduling when we should not be.
4432 */ 4089 */
4433 if (unlikely(in_atomic_preempt_off()) && unlikely(!prev->exit_state)) 4090 if (unlikely(in_atomic_preempt_off() && !prev->exit_state))
4434 __schedule_bug(prev); 4091 __schedule_bug(prev);
4435 4092
4436 profile_hit(SCHED_PROFILING, __builtin_return_address(0)); 4093 profile_hit(SCHED_PROFILING, __builtin_return_address(0));
@@ -4510,12 +4167,10 @@ need_resched_nonpreemptible:
4510 clear_tsk_need_resched(prev); 4167 clear_tsk_need_resched(prev);
4511 4168
4512 if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { 4169 if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) {
4513 if (unlikely((prev->state & TASK_INTERRUPTIBLE) && 4170 if (unlikely(signal_pending_state(prev->state, prev)))
4514 signal_pending(prev))) {
4515 prev->state = TASK_RUNNING; 4171 prev->state = TASK_RUNNING;
4516 } else { 4172 else
4517 deactivate_task(rq, prev, 1); 4173 deactivate_task(rq, prev, 1);
4518 }
4519 switch_count = &prev->nvcsw; 4174 switch_count = &prev->nvcsw;
4520 } 4175 }
4521 4176
@@ -4931,8 +4586,10 @@ void set_user_nice(struct task_struct *p, long nice)
4931 goto out_unlock; 4586 goto out_unlock;
4932 } 4587 }
4933 on_rq = p->se.on_rq; 4588 on_rq = p->se.on_rq;
4934 if (on_rq) 4589 if (on_rq) {
4935 dequeue_task(rq, p, 0); 4590 dequeue_task(rq, p, 0);
4591 dec_load(rq, p);
4592 }
4936 4593
4937 p->static_prio = NICE_TO_PRIO(nice); 4594 p->static_prio = NICE_TO_PRIO(nice);
4938 set_load_weight(p); 4595 set_load_weight(p);
@@ -4942,6 +4599,7 @@ void set_user_nice(struct task_struct *p, long nice)
4942 4599
4943 if (on_rq) { 4600 if (on_rq) {
4944 enqueue_task(rq, p, 0); 4601 enqueue_task(rq, p, 0);
4602 inc_load(rq, p);
4945 /* 4603 /*
4946 * If the task increased its priority or is running and 4604 * If the task increased its priority or is running and
4947 * lowered its priority, then reschedule its CPU: 4605 * lowered its priority, then reschedule its CPU:
@@ -7316,7 +6974,6 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7316 SD_INIT(sd, ALLNODES); 6974 SD_INIT(sd, ALLNODES);
7317 set_domain_attribute(sd, attr); 6975 set_domain_attribute(sd, attr);
7318 sd->span = *cpu_map; 6976 sd->span = *cpu_map;
7319 sd->first_cpu = first_cpu(sd->span);
7320 cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); 6977 cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask);
7321 p = sd; 6978 p = sd;
7322 sd_allnodes = 1; 6979 sd_allnodes = 1;
@@ -7327,7 +6984,6 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7327 SD_INIT(sd, NODE); 6984 SD_INIT(sd, NODE);
7328 set_domain_attribute(sd, attr); 6985 set_domain_attribute(sd, attr);
7329 sched_domain_node_span(cpu_to_node(i), &sd->span); 6986 sched_domain_node_span(cpu_to_node(i), &sd->span);
7330 sd->first_cpu = first_cpu(sd->span);
7331 sd->parent = p; 6987 sd->parent = p;
7332 if (p) 6988 if (p)
7333 p->child = sd; 6989 p->child = sd;
@@ -7339,7 +6995,6 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7339 SD_INIT(sd, CPU); 6995 SD_INIT(sd, CPU);
7340 set_domain_attribute(sd, attr); 6996 set_domain_attribute(sd, attr);
7341 sd->span = *nodemask; 6997 sd->span = *nodemask;
7342 sd->first_cpu = first_cpu(sd->span);
7343 sd->parent = p; 6998 sd->parent = p;
7344 if (p) 6999 if (p)
7345 p->child = sd; 7000 p->child = sd;
@@ -7351,7 +7006,6 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7351 SD_INIT(sd, MC); 7006 SD_INIT(sd, MC);
7352 set_domain_attribute(sd, attr); 7007 set_domain_attribute(sd, attr);
7353 sd->span = cpu_coregroup_map(i); 7008 sd->span = cpu_coregroup_map(i);
7354 sd->first_cpu = first_cpu(sd->span);
7355 cpus_and(sd->span, sd->span, *cpu_map); 7009 cpus_and(sd->span, sd->span, *cpu_map);
7356 sd->parent = p; 7010 sd->parent = p;
7357 p->child = sd; 7011 p->child = sd;
@@ -7364,7 +7018,6 @@ static int __build_sched_domains(const cpumask_t *cpu_map,
7364 SD_INIT(sd, SIBLING); 7018 SD_INIT(sd, SIBLING);
7365 set_domain_attribute(sd, attr); 7019 set_domain_attribute(sd, attr);
7366 sd->span = per_cpu(cpu_sibling_map, i); 7020 sd->span = per_cpu(cpu_sibling_map, i);
7367 sd->first_cpu = first_cpu(sd->span);
7368 cpus_and(sd->span, sd->span, *cpu_map); 7021 cpus_and(sd->span, sd->span, *cpu_map);
7369 sd->parent = p; 7022 sd->parent = p;
7370 p->child = sd; 7023 p->child = sd;
@@ -7568,8 +7221,8 @@ static int build_sched_domains(const cpumask_t *cpu_map)
7568 7221
7569static cpumask_t *doms_cur; /* current sched domains */ 7222static cpumask_t *doms_cur; /* current sched domains */
7570static int ndoms_cur; /* number of sched domains in 'doms_cur' */ 7223static int ndoms_cur; /* number of sched domains in 'doms_cur' */
7571static struct sched_domain_attr *dattr_cur; /* attribues of custom domains 7224static struct sched_domain_attr *dattr_cur;
7572 in 'doms_cur' */ 7225 /* attribues of custom domains in 'doms_cur' */
7573 7226
7574/* 7227/*
7575 * Special case: If a kmalloc of a doms_cur partition (array of 7228 * Special case: If a kmalloc of a doms_cur partition (array of
@@ -8034,7 +7687,6 @@ void __init sched_init(void)
8034 } 7687 }
8035 7688
8036#ifdef CONFIG_SMP 7689#ifdef CONFIG_SMP
8037 init_aggregate();
8038 init_defrootdomain(); 7690 init_defrootdomain();
8039#endif 7691#endif
8040 7692
@@ -8599,11 +8251,14 @@ void sched_move_task(struct task_struct *tsk)
8599#endif 8251#endif
8600 8252
8601#ifdef CONFIG_FAIR_GROUP_SCHED 8253#ifdef CONFIG_FAIR_GROUP_SCHED
8602static void __set_se_shares(struct sched_entity *se, unsigned long shares) 8254static void set_se_shares(struct sched_entity *se, unsigned long shares)
8603{ 8255{
8604 struct cfs_rq *cfs_rq = se->cfs_rq; 8256 struct cfs_rq *cfs_rq = se->cfs_rq;
8257 struct rq *rq = cfs_rq->rq;
8605 int on_rq; 8258 int on_rq;
8606 8259
8260 spin_lock_irq(&rq->lock);
8261
8607 on_rq = se->on_rq; 8262 on_rq = se->on_rq;
8608 if (on_rq) 8263 if (on_rq)
8609 dequeue_entity(cfs_rq, se, 0); 8264 dequeue_entity(cfs_rq, se, 0);
@@ -8613,17 +8268,8 @@ static void __set_se_shares(struct sched_entity *se, unsigned long shares)
8613 8268
8614 if (on_rq) 8269 if (on_rq)
8615 enqueue_entity(cfs_rq, se, 0); 8270 enqueue_entity(cfs_rq, se, 0);
8616}
8617 8271
8618static void set_se_shares(struct sched_entity *se, unsigned long shares) 8272 spin_unlock_irq(&rq->lock);
8619{
8620 struct cfs_rq *cfs_rq = se->cfs_rq;
8621 struct rq *rq = cfs_rq->rq;
8622 unsigned long flags;
8623
8624 spin_lock_irqsave(&rq->lock, flags);
8625 __set_se_shares(se, shares);
8626 spin_unlock_irqrestore(&rq->lock, flags);
8627} 8273}
8628 8274
8629static DEFINE_MUTEX(shares_mutex); 8275static DEFINE_MUTEX(shares_mutex);
@@ -8662,13 +8308,8 @@ int sched_group_set_shares(struct task_group *tg, unsigned long shares)
8662 * w/o tripping rebalance_share or load_balance_fair. 8308 * w/o tripping rebalance_share or load_balance_fair.
8663 */ 8309 */
8664 tg->shares = shares; 8310 tg->shares = shares;
8665 for_each_possible_cpu(i) { 8311 for_each_possible_cpu(i)
8666 /*
8667 * force a rebalance
8668 */
8669 cfs_rq_set_shares(tg->cfs_rq[i], 0);
8670 set_se_shares(tg->se[i], shares); 8312 set_se_shares(tg->se[i], shares);
8671 }
8672 8313
8673 /* 8314 /*
8674 * Enable load balance activity on this group, by inserting it back on 8315 * Enable load balance activity on this group, by inserting it back on
diff --git a/kernel/sched_clock.c b/kernel/sched_clock.c
index 9c597e37f7de..ce05271219ab 100644
--- a/kernel/sched_clock.c
+++ b/kernel/sched_clock.c
@@ -59,22 +59,26 @@ static inline struct sched_clock_data *cpu_sdc(int cpu)
59 return &per_cpu(sched_clock_data, cpu); 59 return &per_cpu(sched_clock_data, cpu);
60} 60}
61 61
62static __read_mostly int sched_clock_running;
63
62void sched_clock_init(void) 64void sched_clock_init(void)
63{ 65{
64 u64 ktime_now = ktime_to_ns(ktime_get()); 66 u64 ktime_now = ktime_to_ns(ktime_get());
65 u64 now = 0; 67 unsigned long now_jiffies = jiffies;
66 int cpu; 68 int cpu;
67 69
68 for_each_possible_cpu(cpu) { 70 for_each_possible_cpu(cpu) {
69 struct sched_clock_data *scd = cpu_sdc(cpu); 71 struct sched_clock_data *scd = cpu_sdc(cpu);
70 72
71 scd->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED; 73 scd->lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
72 scd->prev_jiffies = jiffies; 74 scd->prev_jiffies = now_jiffies;
73 scd->prev_raw = now; 75 scd->prev_raw = 0;
74 scd->tick_raw = now; 76 scd->tick_raw = 0;
75 scd->tick_gtod = ktime_now; 77 scd->tick_gtod = ktime_now;
76 scd->clock = ktime_now; 78 scd->clock = ktime_now;
77 } 79 }
80
81 sched_clock_running = 1;
78} 82}
79 83
80/* 84/*
@@ -136,6 +140,9 @@ u64 sched_clock_cpu(int cpu)
136 struct sched_clock_data *scd = cpu_sdc(cpu); 140 struct sched_clock_data *scd = cpu_sdc(cpu);
137 u64 now, clock; 141 u64 now, clock;
138 142
143 if (unlikely(!sched_clock_running))
144 return 0ull;
145
139 WARN_ON_ONCE(!irqs_disabled()); 146 WARN_ON_ONCE(!irqs_disabled());
140 now = sched_clock(); 147 now = sched_clock();
141 148
@@ -174,6 +181,9 @@ void sched_clock_tick(void)
174 struct sched_clock_data *scd = this_scd(); 181 struct sched_clock_data *scd = this_scd();
175 u64 now, now_gtod; 182 u64 now, now_gtod;
176 183
184 if (unlikely(!sched_clock_running))
185 return;
186
177 WARN_ON_ONCE(!irqs_disabled()); 187 WARN_ON_ONCE(!irqs_disabled());
178 188
179 now = sched_clock(); 189 now = sched_clock();
diff --git a/kernel/sched_debug.c b/kernel/sched_debug.c
index 5f06118fbc31..8bb713040ac9 100644
--- a/kernel/sched_debug.c
+++ b/kernel/sched_debug.c
@@ -167,11 +167,6 @@ void print_cfs_rq(struct seq_file *m, int cpu, struct cfs_rq *cfs_rq)
167#endif 167#endif
168 SEQ_printf(m, " .%-30s: %ld\n", "nr_spread_over", 168 SEQ_printf(m, " .%-30s: %ld\n", "nr_spread_over",
169 cfs_rq->nr_spread_over); 169 cfs_rq->nr_spread_over);
170#ifdef CONFIG_FAIR_GROUP_SCHED
171#ifdef CONFIG_SMP
172 SEQ_printf(m, " .%-30s: %lu\n", "shares", cfs_rq->shares);
173#endif
174#endif
175} 170}
176 171
177static void print_cpu(struct seq_file *m, int cpu) 172static void print_cpu(struct seq_file *m, int cpu)
diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
index e24ecd39c4b8..08ae848b71d4 100644
--- a/kernel/sched_fair.c
+++ b/kernel/sched_fair.c
@@ -334,34 +334,6 @@ int sched_nr_latency_handler(struct ctl_table *table, int write,
334#endif 334#endif
335 335
336/* 336/*
337 * delta *= w / rw
338 */
339static inline unsigned long
340calc_delta_weight(unsigned long delta, struct sched_entity *se)
341{
342 for_each_sched_entity(se) {
343 delta = calc_delta_mine(delta,
344 se->load.weight, &cfs_rq_of(se)->load);
345 }
346
347 return delta;
348}
349
350/*
351 * delta *= rw / w
352 */
353static inline unsigned long
354calc_delta_fair(unsigned long delta, struct sched_entity *se)
355{
356 for_each_sched_entity(se) {
357 delta = calc_delta_mine(delta,
358 cfs_rq_of(se)->load.weight, &se->load);
359 }
360
361 return delta;
362}
363
364/*
365 * The idea is to set a period in which each task runs once. 337 * The idea is to set a period in which each task runs once.
366 * 338 *
367 * When there are too many tasks (sysctl_sched_nr_latency) we have to stretch 339 * When there are too many tasks (sysctl_sched_nr_latency) we have to stretch
@@ -390,54 +362,47 @@ static u64 __sched_period(unsigned long nr_running)
390 */ 362 */
391static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se) 363static u64 sched_slice(struct cfs_rq *cfs_rq, struct sched_entity *se)
392{ 364{
393 return calc_delta_weight(__sched_period(cfs_rq->nr_running), se); 365 u64 slice = __sched_period(cfs_rq->nr_running);
366
367 for_each_sched_entity(se) {
368 cfs_rq = cfs_rq_of(se);
369
370 slice *= se->load.weight;
371 do_div(slice, cfs_rq->load.weight);
372 }
373
374
375 return slice;
394} 376}
395 377
396/* 378/*
397 * We calculate the vruntime slice of a to be inserted task 379 * We calculate the vruntime slice of a to be inserted task
398 * 380 *
399 * vs = s*rw/w = p 381 * vs = s/w = p/rw
400 */ 382 */
401static u64 sched_vslice_add(struct cfs_rq *cfs_rq, struct sched_entity *se) 383static u64 sched_vslice_add(struct cfs_rq *cfs_rq, struct sched_entity *se)
402{ 384{
403 unsigned long nr_running = cfs_rq->nr_running; 385 unsigned long nr_running = cfs_rq->nr_running;
386 unsigned long weight;
387 u64 vslice;
404 388
405 if (!se->on_rq) 389 if (!se->on_rq)
406 nr_running++; 390 nr_running++;
407 391
408 return __sched_period(nr_running); 392 vslice = __sched_period(nr_running);
409}
410
411/*
412 * The goal of calc_delta_asym() is to be asymmetrically around NICE_0_LOAD, in
413 * that it favours >=0 over <0.
414 *
415 * -20 |
416 * |
417 * 0 --------+-------
418 * .'
419 * 19 .'
420 *
421 */
422static unsigned long
423calc_delta_asym(unsigned long delta, struct sched_entity *se)
424{
425 struct load_weight lw = {
426 .weight = NICE_0_LOAD,
427 .inv_weight = 1UL << (WMULT_SHIFT-NICE_0_SHIFT)
428 };
429 393
430 for_each_sched_entity(se) { 394 for_each_sched_entity(se) {
431 struct load_weight *se_lw = &se->load; 395 cfs_rq = cfs_rq_of(se);
432 396
433 if (se->load.weight < NICE_0_LOAD) 397 weight = cfs_rq->load.weight;
434 se_lw = &lw; 398 if (!se->on_rq)
399 weight += se->load.weight;
435 400
436 delta = calc_delta_mine(delta, 401 vslice *= NICE_0_LOAD;
437 cfs_rq_of(se)->load.weight, se_lw); 402 do_div(vslice, weight);
438 } 403 }
439 404
440 return delta; 405 return vslice;
441} 406}
442 407
443/* 408/*
@@ -454,7 +419,11 @@ __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr,
454 419
455 curr->sum_exec_runtime += delta_exec; 420 curr->sum_exec_runtime += delta_exec;
456 schedstat_add(cfs_rq, exec_clock, delta_exec); 421 schedstat_add(cfs_rq, exec_clock, delta_exec);
457 delta_exec_weighted = calc_delta_fair(delta_exec, curr); 422 delta_exec_weighted = delta_exec;
423 if (unlikely(curr->load.weight != NICE_0_LOAD)) {
424 delta_exec_weighted = calc_delta_fair(delta_exec_weighted,
425 &curr->load);
426 }
458 curr->vruntime += delta_exec_weighted; 427 curr->vruntime += delta_exec_weighted;
459} 428}
460 429
@@ -541,27 +510,10 @@ update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
541 * Scheduling class queueing methods: 510 * Scheduling class queueing methods:
542 */ 511 */
543 512
544#if defined CONFIG_SMP && defined CONFIG_FAIR_GROUP_SCHED
545static void
546add_cfs_task_weight(struct cfs_rq *cfs_rq, unsigned long weight)
547{
548 cfs_rq->task_weight += weight;
549}
550#else
551static inline void
552add_cfs_task_weight(struct cfs_rq *cfs_rq, unsigned long weight)
553{
554}
555#endif
556
557static void 513static void
558account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se) 514account_entity_enqueue(struct cfs_rq *cfs_rq, struct sched_entity *se)
559{ 515{
560 update_load_add(&cfs_rq->load, se->load.weight); 516 update_load_add(&cfs_rq->load, se->load.weight);
561 if (!parent_entity(se))
562 inc_cpu_load(rq_of(cfs_rq), se->load.weight);
563 if (entity_is_task(se))
564 add_cfs_task_weight(cfs_rq, se->load.weight);
565 cfs_rq->nr_running++; 517 cfs_rq->nr_running++;
566 se->on_rq = 1; 518 se->on_rq = 1;
567 list_add(&se->group_node, &cfs_rq->tasks); 519 list_add(&se->group_node, &cfs_rq->tasks);
@@ -571,10 +523,6 @@ static void
571account_entity_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se) 523account_entity_dequeue(struct cfs_rq *cfs_rq, struct sched_entity *se)
572{ 524{
573 update_load_sub(&cfs_rq->load, se->load.weight); 525 update_load_sub(&cfs_rq->load, se->load.weight);
574 if (!parent_entity(se))
575 dec_cpu_load(rq_of(cfs_rq), se->load.weight);
576 if (entity_is_task(se))
577 add_cfs_task_weight(cfs_rq, -se->load.weight);
578 cfs_rq->nr_running--; 526 cfs_rq->nr_running--;
579 se->on_rq = 0; 527 se->on_rq = 0;
580 list_del_init(&se->group_node); 528 list_del_init(&se->group_node);
@@ -661,17 +609,8 @@ place_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int initial)
661 609
662 if (!initial) { 610 if (!initial) {
663 /* sleeps upto a single latency don't count. */ 611 /* sleeps upto a single latency don't count. */
664 if (sched_feat(NEW_FAIR_SLEEPERS)) { 612 if (sched_feat(NEW_FAIR_SLEEPERS))
665 unsigned long thresh = sysctl_sched_latency; 613 vruntime -= sysctl_sched_latency;
666
667 /*
668 * convert the sleeper threshold into virtual time
669 */
670 if (sched_feat(NORMALIZED_SLEEPER))
671 thresh = calc_delta_fair(thresh, se);
672
673 vruntime -= thresh;
674 }
675 614
676 /* ensure we never gain time by being placed backwards. */ 615 /* ensure we never gain time by being placed backwards. */
677 vruntime = max_vruntime(se->vruntime, vruntime); 616 vruntime = max_vruntime(se->vruntime, vruntime);
@@ -1057,16 +996,27 @@ wake_affine(struct rq *rq, struct sched_domain *this_sd, struct rq *this_rq,
1057 struct task_struct *curr = this_rq->curr; 996 struct task_struct *curr = this_rq->curr;
1058 unsigned long tl = this_load; 997 unsigned long tl = this_load;
1059 unsigned long tl_per_task; 998 unsigned long tl_per_task;
999 int balanced;
1060 1000
1061 if (!(this_sd->flags & SD_WAKE_AFFINE)) 1001 if (!(this_sd->flags & SD_WAKE_AFFINE) || !sched_feat(AFFINE_WAKEUPS))
1062 return 0; 1002 return 0;
1063 1003
1064 /* 1004 /*
1005 * If sync wakeup then subtract the (maximum possible)
1006 * effect of the currently running task from the load
1007 * of the current CPU:
1008 */
1009 if (sync)
1010 tl -= current->se.load.weight;
1011
1012 balanced = 100*(tl + p->se.load.weight) <= imbalance*load;
1013
1014 /*
1065 * If the currently running task will sleep within 1015 * If the currently running task will sleep within
1066 * a reasonable amount of time then attract this newly 1016 * a reasonable amount of time then attract this newly
1067 * woken task: 1017 * woken task:
1068 */ 1018 */
1069 if (sync && curr->sched_class == &fair_sched_class) { 1019 if (sync && balanced && curr->sched_class == &fair_sched_class) {
1070 if (curr->se.avg_overlap < sysctl_sched_migration_cost && 1020 if (curr->se.avg_overlap < sysctl_sched_migration_cost &&
1071 p->se.avg_overlap < sysctl_sched_migration_cost) 1021 p->se.avg_overlap < sysctl_sched_migration_cost)
1072 return 1; 1022 return 1;
@@ -1075,16 +1025,8 @@ wake_affine(struct rq *rq, struct sched_domain *this_sd, struct rq *this_rq,
1075 schedstat_inc(p, se.nr_wakeups_affine_attempts); 1025 schedstat_inc(p, se.nr_wakeups_affine_attempts);
1076 tl_per_task = cpu_avg_load_per_task(this_cpu); 1026 tl_per_task = cpu_avg_load_per_task(this_cpu);
1077 1027
1078 /*
1079 * If sync wakeup then subtract the (maximum possible)
1080 * effect of the currently running task from the load
1081 * of the current CPU:
1082 */
1083 if (sync)
1084 tl -= current->se.load.weight;
1085
1086 if ((tl <= load && tl + target_load(prev_cpu, idx) <= tl_per_task) || 1028 if ((tl <= load && tl + target_load(prev_cpu, idx) <= tl_per_task) ||
1087 100*(tl + p->se.load.weight) <= imbalance*load) { 1029 balanced) {
1088 /* 1030 /*
1089 * This domain has SD_WAKE_AFFINE and 1031 * This domain has SD_WAKE_AFFINE and
1090 * p is cache cold in this domain, and 1032 * p is cache cold in this domain, and
@@ -1169,10 +1111,11 @@ static unsigned long wakeup_gran(struct sched_entity *se)
1169 unsigned long gran = sysctl_sched_wakeup_granularity; 1111 unsigned long gran = sysctl_sched_wakeup_granularity;
1170 1112
1171 /* 1113 /*
1172 * More easily preempt - nice tasks, while not making it harder for 1114 * More easily preempt - nice tasks, while not making
1173 * + nice tasks. 1115 * it harder for + nice tasks.
1174 */ 1116 */
1175 gran = calc_delta_asym(sysctl_sched_wakeup_granularity, se); 1117 if (unlikely(se->load.weight > NICE_0_LOAD))
1118 gran = calc_delta_fair(gran, &se->load);
1176 1119
1177 return gran; 1120 return gran;
1178} 1121}
@@ -1366,90 +1309,75 @@ static struct task_struct *load_balance_next_fair(void *arg)
1366 return __load_balance_iterator(cfs_rq, cfs_rq->balance_iterator); 1309 return __load_balance_iterator(cfs_rq, cfs_rq->balance_iterator);
1367} 1310}
1368 1311
1369static unsigned long 1312#ifdef CONFIG_FAIR_GROUP_SCHED
1370__load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, 1313static int cfs_rq_best_prio(struct cfs_rq *cfs_rq)
1371 unsigned long max_load_move, struct sched_domain *sd,
1372 enum cpu_idle_type idle, int *all_pinned, int *this_best_prio,
1373 struct cfs_rq *cfs_rq)
1374{ 1314{
1375 struct rq_iterator cfs_rq_iterator; 1315 struct sched_entity *curr;
1316 struct task_struct *p;
1376 1317
1377 cfs_rq_iterator.start = load_balance_start_fair; 1318 if (!cfs_rq->nr_running || !first_fair(cfs_rq))
1378 cfs_rq_iterator.next = load_balance_next_fair; 1319 return MAX_PRIO;
1379 cfs_rq_iterator.arg = cfs_rq; 1320
1321 curr = cfs_rq->curr;
1322 if (!curr)
1323 curr = __pick_next_entity(cfs_rq);
1324
1325 p = task_of(curr);
1380 1326
1381 return balance_tasks(this_rq, this_cpu, busiest, 1327 return p->prio;
1382 max_load_move, sd, idle, all_pinned,
1383 this_best_prio, &cfs_rq_iterator);
1384} 1328}
1329#endif
1385 1330
1386#ifdef CONFIG_FAIR_GROUP_SCHED
1387static unsigned long 1331static unsigned long
1388load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, 1332load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest,
1389 unsigned long max_load_move, 1333 unsigned long max_load_move,
1390 struct sched_domain *sd, enum cpu_idle_type idle, 1334 struct sched_domain *sd, enum cpu_idle_type idle,
1391 int *all_pinned, int *this_best_prio) 1335 int *all_pinned, int *this_best_prio)
1392{ 1336{
1337 struct cfs_rq *busy_cfs_rq;
1393 long rem_load_move = max_load_move; 1338 long rem_load_move = max_load_move;
1394 int busiest_cpu = cpu_of(busiest); 1339 struct rq_iterator cfs_rq_iterator;
1395 struct task_group *tg;
1396
1397 rcu_read_lock();
1398 list_for_each_entry(tg, &task_groups, list) {
1399 long imbalance;
1400 unsigned long this_weight, busiest_weight;
1401 long rem_load, max_load, moved_load;
1402
1403 /*
1404 * empty group
1405 */
1406 if (!aggregate(tg, sd)->task_weight)
1407 continue;
1408
1409 rem_load = rem_load_move * aggregate(tg, sd)->rq_weight;
1410 rem_load /= aggregate(tg, sd)->load + 1;
1411
1412 this_weight = tg->cfs_rq[this_cpu]->task_weight;
1413 busiest_weight = tg->cfs_rq[busiest_cpu]->task_weight;
1414 1340
1415 imbalance = (busiest_weight - this_weight) / 2; 1341 cfs_rq_iterator.start = load_balance_start_fair;
1342 cfs_rq_iterator.next = load_balance_next_fair;
1416 1343
1417 if (imbalance < 0) 1344 for_each_leaf_cfs_rq(busiest, busy_cfs_rq) {
1418 imbalance = busiest_weight; 1345#ifdef CONFIG_FAIR_GROUP_SCHED
1346 struct cfs_rq *this_cfs_rq;
1347 long imbalance;
1348 unsigned long maxload;
1419 1349
1420 max_load = max(rem_load, imbalance); 1350 this_cfs_rq = cpu_cfs_rq(busy_cfs_rq, this_cpu);
1421 moved_load = __load_balance_fair(this_rq, this_cpu, busiest,
1422 max_load, sd, idle, all_pinned, this_best_prio,
1423 tg->cfs_rq[busiest_cpu]);
1424 1351
1425 if (!moved_load) 1352 imbalance = busy_cfs_rq->load.weight - this_cfs_rq->load.weight;
1353 /* Don't pull if this_cfs_rq has more load than busy_cfs_rq */
1354 if (imbalance <= 0)
1426 continue; 1355 continue;
1427 1356
1428 move_group_shares(tg, sd, busiest_cpu, this_cpu); 1357 /* Don't pull more than imbalance/2 */
1358 imbalance /= 2;
1359 maxload = min(rem_load_move, imbalance);
1429 1360
1430 moved_load *= aggregate(tg, sd)->load; 1361 *this_best_prio = cfs_rq_best_prio(this_cfs_rq);
1431 moved_load /= aggregate(tg, sd)->rq_weight + 1; 1362#else
1363# define maxload rem_load_move
1364#endif
1365 /*
1366 * pass busy_cfs_rq argument into
1367 * load_balance_[start|next]_fair iterators
1368 */
1369 cfs_rq_iterator.arg = busy_cfs_rq;
1370 rem_load_move -= balance_tasks(this_rq, this_cpu, busiest,
1371 maxload, sd, idle, all_pinned,
1372 this_best_prio,
1373 &cfs_rq_iterator);
1432 1374
1433 rem_load_move -= moved_load; 1375 if (rem_load_move <= 0)
1434 if (rem_load_move < 0)
1435 break; 1376 break;
1436 } 1377 }
1437 rcu_read_unlock();
1438 1378
1439 return max_load_move - rem_load_move; 1379 return max_load_move - rem_load_move;
1440} 1380}
1441#else
1442static unsigned long
1443load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest,
1444 unsigned long max_load_move,
1445 struct sched_domain *sd, enum cpu_idle_type idle,
1446 int *all_pinned, int *this_best_prio)
1447{
1448 return __load_balance_fair(this_rq, this_cpu, busiest,
1449 max_load_move, sd, idle, all_pinned,
1450 this_best_prio, &busiest->cfs);
1451}
1452#endif
1453 1381
1454static int 1382static int
1455move_one_task_fair(struct rq *this_rq, int this_cpu, struct rq *busiest, 1383move_one_task_fair(struct rq *this_rq, int this_cpu, struct rq *busiest,
diff --git a/kernel/sched_rt.c b/kernel/sched_rt.c
index 060e87b0cb1c..3432d573205d 100644
--- a/kernel/sched_rt.c
+++ b/kernel/sched_rt.c
@@ -513,8 +513,6 @@ static void enqueue_task_rt(struct rq *rq, struct task_struct *p, int wakeup)
513 */ 513 */
514 for_each_sched_rt_entity(rt_se) 514 for_each_sched_rt_entity(rt_se)
515 enqueue_rt_entity(rt_se); 515 enqueue_rt_entity(rt_se);
516
517 inc_cpu_load(rq, p->se.load.weight);
518} 516}
519 517
520static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep) 518static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep)
@@ -534,8 +532,6 @@ static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep)
534 if (rt_rq && rt_rq->rt_nr_running) 532 if (rt_rq && rt_rq->rt_nr_running)
535 enqueue_rt_entity(rt_se); 533 enqueue_rt_entity(rt_se);
536 } 534 }
537
538 dec_cpu_load(rq, p->se.load.weight);
539} 535}
540 536
541/* 537/*
diff --git a/kernel/sched_stats.h b/kernel/sched_stats.h
index 5bae2e0c3ff2..a38878e0e49d 100644
--- a/kernel/sched_stats.h
+++ b/kernel/sched_stats.h
@@ -67,6 +67,7 @@ static int show_schedstat(struct seq_file *seq, void *v)
67 preempt_enable(); 67 preempt_enable();
68#endif 68#endif
69 } 69 }
70 kfree(mask_str);
70 return 0; 71 return 0;
71} 72}
72 73
diff --git a/kernel/signal.c b/kernel/signal.c
index 72bb4f51f963..6c0958e52ea7 100644
--- a/kernel/signal.c
+++ b/kernel/signal.c
@@ -231,6 +231,40 @@ void flush_signals(struct task_struct *t)
231 spin_unlock_irqrestore(&t->sighand->siglock, flags); 231 spin_unlock_irqrestore(&t->sighand->siglock, flags);
232} 232}
233 233
234static void __flush_itimer_signals(struct sigpending *pending)
235{
236 sigset_t signal, retain;
237 struct sigqueue *q, *n;
238
239 signal = pending->signal;
240 sigemptyset(&retain);
241
242 list_for_each_entry_safe(q, n, &pending->list, list) {
243 int sig = q->info.si_signo;
244
245 if (likely(q->info.si_code != SI_TIMER)) {
246 sigaddset(&retain, sig);
247 } else {
248 sigdelset(&signal, sig);
249 list_del_init(&q->list);
250 __sigqueue_free(q);
251 }
252 }
253
254 sigorsets(&pending->signal, &signal, &retain);
255}
256
257void flush_itimer_signals(void)
258{
259 struct task_struct *tsk = current;
260 unsigned long flags;
261
262 spin_lock_irqsave(&tsk->sighand->siglock, flags);
263 __flush_itimer_signals(&tsk->pending);
264 __flush_itimer_signals(&tsk->signal->shared_pending);
265 spin_unlock_irqrestore(&tsk->sighand->siglock, flags);
266}
267
234void ignore_signals(struct task_struct *t) 268void ignore_signals(struct task_struct *t)
235{ 269{
236 int i; 270 int i;
@@ -1240,17 +1274,22 @@ void sigqueue_free(struct sigqueue *q)
1240 1274
1241 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC)); 1275 BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
1242 /* 1276 /*
1243 * If the signal is still pending remove it from the 1277 * We must hold ->siglock while testing q->list
1244 * pending queue. We must hold ->siglock while testing 1278 * to serialize with collect_signal() or with
1245 * q->list to serialize with collect_signal(). 1279 * __exit_signal()->flush_sigqueue().
1246 */ 1280 */
1247 spin_lock_irqsave(lock, flags); 1281 spin_lock_irqsave(lock, flags);
1282 q->flags &= ~SIGQUEUE_PREALLOC;
1283 /*
1284 * If it is queued it will be freed when dequeued,
1285 * like the "regular" sigqueue.
1286 */
1248 if (!list_empty(&q->list)) 1287 if (!list_empty(&q->list))
1249 list_del_init(&q->list); 1288 q = NULL;
1250 spin_unlock_irqrestore(lock, flags); 1289 spin_unlock_irqrestore(lock, flags);
1251 1290
1252 q->flags &= ~SIGQUEUE_PREALLOC; 1291 if (q)
1253 __sigqueue_free(q); 1292 __sigqueue_free(q);
1254} 1293}
1255 1294
1256int send_sigqueue(struct sigqueue *q, struct task_struct *t, int group) 1295int send_sigqueue(struct sigqueue *q, struct task_struct *t, int group)
diff --git a/kernel/stop_machine.c b/kernel/stop_machine.c
index 0101aeef7ed7..b7350bbfb076 100644
--- a/kernel/stop_machine.c
+++ b/kernel/stop_machine.c
@@ -62,8 +62,7 @@ static int stopmachine(void *cpu)
62 * help our sisters onto their CPUs. */ 62 * help our sisters onto their CPUs. */
63 if (!prepared && !irqs_disabled) 63 if (!prepared && !irqs_disabled)
64 yield(); 64 yield();
65 else 65 cpu_relax();
66 cpu_relax();
67 } 66 }
68 67
69 /* Ack: we are exiting. */ 68 /* Ack: we are exiting. */
@@ -106,8 +105,10 @@ static int stop_machine(void)
106 } 105 }
107 106
108 /* Wait for them all to come to life. */ 107 /* Wait for them all to come to life. */
109 while (atomic_read(&stopmachine_thread_ack) != stopmachine_num_threads) 108 while (atomic_read(&stopmachine_thread_ack) != stopmachine_num_threads) {
110 yield(); 109 yield();
110 cpu_relax();
111 }
111 112
112 /* If some failed, kill them all. */ 113 /* If some failed, kill them all. */
113 if (ret < 0) { 114 if (ret < 0) {
diff --git a/kernel/sys.c b/kernel/sys.c
index 895d2d4c9493..14e97282eb6c 100644
--- a/kernel/sys.c
+++ b/kernel/sys.c
@@ -1652,7 +1652,7 @@ asmlinkage long sys_umask(int mask)
1652asmlinkage long sys_prctl(int option, unsigned long arg2, unsigned long arg3, 1652asmlinkage long sys_prctl(int option, unsigned long arg2, unsigned long arg3,
1653 unsigned long arg4, unsigned long arg5) 1653 unsigned long arg4, unsigned long arg5)
1654{ 1654{
1655 long uninitialized_var(error); 1655 long error = 0;
1656 1656
1657 if (security_task_prctl(option, arg2, arg3, arg4, arg5, &error)) 1657 if (security_task_prctl(option, arg2, arg3, arg4, arg5, &error))
1658 return error; 1658 return error;
@@ -1701,9 +1701,7 @@ asmlinkage long sys_prctl(int option, unsigned long arg2, unsigned long arg3,
1701 error = PR_TIMING_STATISTICAL; 1701 error = PR_TIMING_STATISTICAL;
1702 break; 1702 break;
1703 case PR_SET_TIMING: 1703 case PR_SET_TIMING:
1704 if (arg2 == PR_TIMING_STATISTICAL) 1704 if (arg2 != PR_TIMING_STATISTICAL)
1705 error = 0;
1706 else
1707 error = -EINVAL; 1705 error = -EINVAL;
1708 break; 1706 break;
1709 1707
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index d7ffdc59816a..29116652dca8 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -81,6 +81,7 @@ extern int compat_log;
81extern int maps_protect; 81extern int maps_protect;
82extern int sysctl_stat_interval; 82extern int sysctl_stat_interval;
83extern int latencytop_enabled; 83extern int latencytop_enabled;
84extern int sysctl_nr_open_min, sysctl_nr_open_max;
84 85
85/* Constants used for minimum and maximum */ 86/* Constants used for minimum and maximum */
86#if defined(CONFIG_DETECT_SOFTLOCKUP) || defined(CONFIG_HIGHMEM) 87#if defined(CONFIG_DETECT_SOFTLOCKUP) || defined(CONFIG_HIGHMEM)
@@ -1190,7 +1191,9 @@ static struct ctl_table fs_table[] = {
1190 .data = &sysctl_nr_open, 1191 .data = &sysctl_nr_open,
1191 .maxlen = sizeof(int), 1192 .maxlen = sizeof(int),
1192 .mode = 0644, 1193 .mode = 0644,
1193 .proc_handler = &proc_dointvec, 1194 .proc_handler = &proc_dointvec_minmax,
1195 .extra1 = &sysctl_nr_open_min,
1196 .extra2 = &sysctl_nr_open_max,
1194 }, 1197 },
1195 { 1198 {
1196 .ctl_name = FS_DENTRY, 1199 .ctl_name = FS_DENTRY,
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-12 15:22:09 -0500