- 27 May, 2010 40 commits
-
-
Manfred Spraul authored
The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by:
Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
Imre Deak authored
Currently idr_remove_all will fail with a use after free error if idr::layers is bigger than 2, which on 32 bit systems corresponds to items more than 1024. This is due to stepping back too many levels during backtracking. For simplicity let's assume that IDR_BITS=1 -> we have 2 nodes at each level below the root node and each leaf node stores two IDs. (In reality for 32 bit systems IDR_BITS=5, with 32 nodes at each sub-root level and 32 IDs in each leaf node). The sequence of freeing the nodes at the moment is as follows: layer 1 -> a(7) 2 -> b(3) c(5) 3 -> d(1) e(2) f(4) g(6) Until step 4 things go fine, but then node c is freed, whereas node g should be freed first. Since node c contains the pointer to node g we'll have a use after free error at step 6. How many levels we step back after visiting the leaf nodes is currently determined by the msb of the id we are currently visiting: Step 1. node d with IDs 0,1 is freed, current ID is advanced to 2. msb of the current ID bit 1. This means we need to step back 1 level to node b and take the next sibling, node e. 2-3. node e with IDs 2,3 is freed, current ID is 4, msb is bit 2. This means we need to step back 2 levels to node a, freeing node b on the way. 4-5. node f with IDs 4,5 is freed, current ID is 6, msb is still bit 2. This means we again need to step back 2 levels to node a and free c on the way. 6. We should visit node g, but its pointer is not available as node c was freed. The fix changes how we determine the number of levels to step back. Instead of deducting this merely from the msb of the current ID, we should really check if advancing the ID causes an overflow to a bit position corresponding to a given layer. In the above example overflow from bit 0 to bit 1 should mean stepping back 1 level. Overflow from bit 1 to bit 2 should mean stepping back 2 levels and so on. The fix was tested with IDs up to 1 << 20, which corresponds to 4 layers on 32 bit systems. Signed-off-by:Imre Deak <imre.deak@nokia.com> Reviewed-by:
Tejun Heo <tj@kernel.org> Cc: Eric Paris <eparis@redhat.com> Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com> Cc: <stable@kernel.org> [2.6.34.1] Signed-off-by:
-
Lai Jiangshan authored
Since when CONFIG_HOTPLUG_CPU=n, get_online_cpus() do nothing, so we don't need cpu_hotplug_begin() either. This patch moves cpu_hotplug_begin()/cpu_hotplug_done() into the code block of CONFIG_HOTPLUG_CPU=y. Signed-off-by:
Lai Jiangshan <laijs@cn.fujitsu.com> Cc: Gautham R Shenoy <ego@in.ibm.com> Cc: Ingo Molnar <mingo@elte.hu> Signed-off-by:
-
Akinobu Mita authored
I used this module to test the series of modification to the cpu notifiers code. Example1: inject CPU offline error (-1 == -EPERM) # modprobe cpu-notifier-error-inject cpu_down_prepare_error=-1 # echo 0 > /sys/devices/system/cpu/cpu1/online bash: echo: write error: Operation not permitted Example2: inject CPU online error (-2 == -ENOENT) # modprobe cpu-notifier-error-inject cpu_up_prepare_error=-2 # echo 1 > /sys/devices/system/cpu/cpu1/online bash: echo: write error: No such file or directory [akpm@linux-foundation.org: fix Kconfig help text] Signed-off-by:
Akinobu Mita <akinobu.mita@gmail.com> Signed-off-by:
-
Akinobu Mita authored
By the previous modification, the cpu notifier can return encapsulate errno value. This converts the cpu notifiers for raid5. Signed-off-by:
-
Akinobu Mita authored
By the previous modification, the cpu notifier can return encapsulate errno value. This converts the cpu notifiers for s390. Signed-off-by:
-
Akinobu Mita authored
By the previous modification, the cpu notifier can return encapsulate errno value. This converts the cpu notifiers for ehca. Signed-off-by:
-
Akinobu Mita authored
By the previous modification, the cpu notifier can return encapsulate errno value. This converts the cpu notifiers for iucv. Signed-off-by:
-
Akinobu Mita authored
By the previous modification, the cpu notifier can return encapsulate errno value. This converts the cpu notifiers for slab. Signed-off-by:
Pekka Enberg <penberg@cs.helsinki.fi> Cc: Matt Mackall <mpm@selenic.com> Signed-off-by:
-
Akinobu Mita authored
By the previous modification, the cpu notifier can return encapsulate errno value. This converts the cpu notifiers for kernel/*.c Signed-off-by:
-
Akinobu Mita authored
By the previous modification, the cpu notifier can return encapsulate errno value. This converts the cpu notifiers for topology. Signed-off-by:
-
Akinobu Mita authored
By the previous modification, the cpu notifier can return encapsulate errno value. This converts the cpu notifiers for msr, cpuid, and therm_throt. Signed-off-by:
-
Akinobu Mita authored
This changes notifier_from_errno(0) to be NOTIFY_OK instead of NOTIFY_STOP_MASK | NOTIFY_OK. Currently, the notifiers which return encapsulated errno value have to do something like this: err = do_something(); // returns -errno if (err) return notifier_from_errno(err); else return NOTIFY_OK; This change makes the above code simple: err = do_something(); // returns -errno return return notifier_from_errno(err); Signed-off-by:
-
Akinobu Mita authored
Currently, onlining or offlining a CPU failure by one of the cpu notifiers error always cause -EINVAL error. (i.e. writing 0 or 1 to /sys/devices/system/cpu/cpuX/online gets EINVAL) To get better error reporting rather than always getting -EINVAL, This changes cpu_notify() to return -errno value with notifier_to_errno() and fix the callers. Now that cpu notifiers can return encapsulate errno value. Currently, all cpu hotplug notifiers return NOTIFY_OK, NOTIFY_BAD, or NOTIFY_DONE. So cpu_notify() can returns 0 or -EPERM with this change for now. (notifier_to_errno(NOTIFY_OK) == 0, notifier_to_errno(NOTIFY_DONE) == 0, notifier_to_errno(NOTIFY_BAD) == -EPERM) Forthcoming patches convert several cpu notifiers to return encapsulate errno value with notifier_from_errno(). Signed-off-by:
-
Akinobu Mita authored
No functional change. These are just wrappers of raw_cpu_notifier_call_chain. Signed-off-by:
-
Wu Fengguang authored
Extend KCORE_TEXT to cover the pages between _text and _stext, to allow examining some important page table pages. `readelf -a` output on x86_64 before and after patch: Type Offset VirtAddr PhysAddr before LOAD 0x00007fff8100c000 0xffffffff81009000 0x0000000000000000 after LOAD 0x00007fff81003000 0xffffffff81000000 0x0000000000000000 The newly covered pages are: 0xffffffff81000000 <startup_64> etc. 0xffffffff81001000 <init_level4_pgt> 0xffffffff81002000 <level3_ident_pgt> 0xffffffff81003000 <level3_kernel_pgt> 0xffffffff81004000 <level2_fixmap_pgt> 0xffffffff81005000 <level1_fixmap_pgt> 0xffffffff81006000 <level2_ident_pgt> 0xffffffff81007000 <level2_kernel_pgt> 0xffffffff81008000 <level2_spare_pgt> Before patch, /proc/kcore shows outdated contents for the above page table pages, for example: (gdb) p level3_ident_pgt $1 = {<text variable, no debug info>} 0xffffffff81002000 <level3_ident_pgt> (gdb) p/x *((pud_t *)&level3_ident_pgt)@512 $2 = {{pud = 0x1006063}, {pud = 0x0} <repeats 511 times>} while the real content is: root@hp /home/wfg# hexdump -s 0x1002000 -n 4096 /dev/mem 1002000 6063 0100 0000 0000 8067 0000 0000 0000 1002010 0000 0000 0000 0000 0000 0000 0000 0000 * 1003000 That is, on a x86_64 box with 2GB memory, we can see first-1GB / full-2GB identity mapping before/after patch: (gdb) p/x *((pud_t *)&level3_ident_pgt)@512 before $1 = {{pud = 0x1006063}, {pud = 0x0} <repeats 511 times>} after $1 = {{pud = 0x1006063}, {pud = 0x8067}, {pud = 0x0} <repeats 510 times>} Obviously the content before patch is wrong. Signed-off-by:Wu Fengguang <fengguang.wu@intel.com> Cc: Andi Kleen <andi@firstfloor.org> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Signed-off-by:
-
Amerigo Wang authored
A quick test shows these comments are obsolete, so just remove them. Signed-off-by:
WANG Cong <amwang@redhat.com> Cc: Alexey Dobriyan <adobriyan@gmail.com> Signed-off-by:
-
Dan Carpenter authored
I removed 3 unused assignments. The first two get reset on the first statement of their functions. For "err" in root.c we don't return an error and we don't use the variable again. Signed-off-by:
Dan Carpenter <error27@gmail.com> Cc: Oleg Nesterov <oleg@redhat.com> Acked-by:
Serge Hallyn <serue@us.ibm.com> Reviewed-by:
KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Signed-off-by:
-
Oleg Nesterov authored
No functional changes, just s/atomic_t count/int nr_threads/. With the recent changes this counter has a single user, get_nr_threads() And, none of its callers need the really accurate number of threads, not to mention each caller obviously races with fork/exit. It is only used to report this value to the user-space, except first_tid() uses it to avoid the unnecessary while_each_thread() loop in the unlikely case. It is a bit sad we need a word in struct signal_struct for this, perhaps we can change get_nr_threads() to approximate the number of threads using signal->live and kill ->nr_threads later. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by:
Oleg Nesterov <oleg@redhat.com> Cc: Alexey Dobriyan <adobriyan@gmail.com> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Acked-by:
Roland McGrath <roland@redhat.com> Signed-off-by:
-
Oleg Nesterov authored
No functional changes. keyctl_session_to_parent() is the only user of signal->count which needs the correct value. Change it to use thread_group_empty() instead, this must be strictly equivalent under tasklist, and imho looks better. Signed-off-by:
David Howells <dhowells@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Acked-by:
-
Oleg Nesterov authored
Trivial, use get_nr_threads() helper to read signal->count which we are going to change. Like other callers, proc_sched_show_task() doesn't need the exactly precise nr_threads. David said: : Note that get_nr_threads() isn't completely equivalent (it can return 0 : where proc_sched_show_task() will display a 1). But I don't think this : should be a problem. Signed-off-by:
-
Oleg Nesterov authored
Now that task->signal can't go away get_nr_threads() doesn't need ->siglock to read signal->count. Also, make it inline, move into sched.h, and convert 2 other proc users of signal->count to use this (now trivial) helper. Henceforth get_nr_threads() is the only valid user of signal->count, we are ready to turn it into "int nr_threads" or, perhaps, kill it. Signed-off-by:
-
Oleg Nesterov authored
check_unshare_flags(CLONE_SIGHAND) adds CLONE_THREAD to *flags_ptr if the task is multithreaded to ensure unshare_thread() will fail. Not only this is a bit strange way to return the error, this is absolutely meaningless. If signal->count > 1 then sighand->count must be also > 1, and unshare_sighand() will fail anyway. In fact, all CLONE_THREAD/SIGHAND/VM checks inside sys_unshare() do not look right. Fortunately this code doesn't really work anyway. Signed-off-by:
-
Oleg Nesterov authored
Move taskstats_tgid_free() from __exit_signal() to free_signal_struct(). This way signal->stats never points to nowhere and we can read ->stats lockless. Signed-off-by:
-
Oleg Nesterov authored
Kill the empty thread_group_cputime_free() helper. It was needed to free the per-cpu data which we no longer have. Signed-off-by:
-
Oleg Nesterov authored
Cleanup: - Add the boolean, group_dead = thread_group_leader(), for clarity. - Do not test/set sig == NULL to detect the all-dead case, use this boolean. - Pass this boolen to __unhash_process() and use it instead of another thread_group_leader() call which needs ->group_leader. This can be considered as microoptimization, but hopefully this also allows us do do other cleanups later. Signed-off-by:
-
Oleg Nesterov authored
Now that task->signal can't go away we can revert the horrible hack added by ad474cac ("fix for account_group_exec_runtime(), make sure ->signal can't be freed under rq->lock"). And we can do more cleanups sched_stats.h/posix-cpu-timers.c later. Signed-off-by:
-
Oleg Nesterov authored
When the last thread exits signal->tty is freed, but the pointer is not cleared and points to nowhere. This is OK. Nobody should use signal->tty lockless, and it is no longer possible to take ->siglock. However this looks wrong even if correct, and the nice OOPS is better than subtle and hard to find bugs. Change __exit_signal() to clear signal->tty under ->siglock. Note: __exit_signal() needs more cleanups. It should not check "sig != NULL" to detect the all-dead case and we have the same issues with signal->stats. Signed-off-by:
Peter Zijlstra <peterz@infradead.org> Acked-by:
-
Oleg Nesterov authored
We have a lot of problems with accessing task_struct->signal, it can "disappear" at any moment. Even current can't use its ->signal safely after exit_notify(). ->siglock helps, but it is not convenient, not always possible, and sometimes it makes sense to use task->signal even after this task has already dead. This patch adds the reference counter, sigcnt, into signal_struct. This reference is owned by task_struct and it is dropped in __put_task_struct(). Perhaps it makes sense to export get/put_signal_struct() later, but currently I don't see the immediate reason. Rename __cleanup_signal() to free_signal_struct() and unexport it. With the previous changes it does nothing except kmem_cache_free(). Change __exit_signal() to not clear/free ->signal, it will be freed when the last reference to any thread in the thread group goes away. Note: - when the last thead exits signal->tty can point to nowhere, see the next patch. - with or without this patch signal_struct->count should go away, or at least it should be "int nr_threads" for fs/proc. This will be addressed later. Signed-off-by:
-
Oleg Nesterov authored
tty_kref_put() has two callsites in copy_process() paths, 1. if copy_process() suceeds it is called before we copy signal->tty from parent 2. otherwise it is called from __cleanup_signal() under bad_fork_cleanup_signal: label In both cases tty_kref_put() is not right and unneeded because we don't have the balancing tty_kref_get(). Fortunately, this is harmless because this can only happen without CLONE_THREAD, and in this case signal->tty must be NULL. Remove tty_kref_put() from copy_process() and __cleanup_signal(), and change another caller of __cleanup_signal(), __exit_signal(), to call tty_kref_put() by hand. I hope this change makes sense by itself, but it is also needed to make ->signal refcountable. Signed-off-by:
Alan Cox <alan@linux.intel.com> Acked-by:
-
Oleg Nesterov authored
Preparation to make task->signal immutable, no functional changes. It doesn't matter which pointer we check under tasklist to ensure the task was not released, ->signal or ->sighand. But we are going to make ->signal refcountable, change the code to use ->sighand. Note: this code doesn't need this check and tasklist_lock at all, it should be converted to use lock_task_sighand(). And, the code under SIGNAL_STOP_STOPPED check looks wrong. Signed-off-by:
-
Oleg Nesterov authored
Preparation to make task->signal immutable, no functional changes. posix-cpu-timers.c checks task->signal != NULL to ensure this task is alive and didn't pass __exit_signal(). This is correct but we are going to change the lifetime rules for ->signal and never reset this pointer. Change the code to check ->sighand instead, it doesn't matter which pointer we check under tasklist, they both are cleared simultaneously. As Roland pointed out, some of these changes are not strictly needed and probably it makes sense to revert them later, when ->signal will be pinned to task_struct. But this patch tries to ensure the subsequent changes in fork/exit can't make any visible impact on posix cpu timers. Signed-off-by:
-
Oleg Nesterov authored
Change __exit_signal() to check thread_group_leader() instead of atomic_dec_and_test(&sig->count). This must be equivalent, the group leader must be released only after all other threads have exited and passed __exit_signal(). Henceforth sig->count is not actually used, except in fs/proc for get_nr_threads/etc. Signed-off-by:
-
Oleg Nesterov authored
de_thread() and __exit_signal() use signal_struct->count/notify_count for synchronization. We can simplify the code and use ->notify_count only. Instead of comparing these two counters, we can change de_thread() to set ->notify_count = nr_of_sub_threads, then change __exit_signal() to dec-and-test this counter and notify group_exit_task. Note that __exit_signal() checks "notify_count > 0" just for symmetry with exit_notify(), we could just check it is != 0. Signed-off-by:
-
Oleg Nesterov authored
Change zap_other_threads() to return the number of other sub-threads found on ->thread_group list. Other changes are cosmetic: - change the code to use while_each_thread() helper - remove the obsolete comment about SIGKILL/SIGSTOP Signed-off-by:
-
Oleg Nesterov authored
signal_struct->count in its current form must die. - it has no reasons to be atomic_t - it looks like a reference counter, but it is not - otoh, we really need to make task->signal refcountable, just look at the extremely ugly task_rq_unlock_wait() called from __exit_signals(). - we should change the lifetime rules for task->signal, it should be pinned to task_struct. We have a lot of code which can be simplified after that. - it is not needed! while the code is correct, any usage of this counter is artificial, except fs/proc uses it correctly to show the number of threads. This series removes the usage of sig->count from exit pathes. This patch: Now that Veaceslav changed copy_signal() to use zalloc(), exit_notify() can just check notify_count < 0 to ensure the execing sub-threads needs the notification from us. No need to do other checks, notify_count != 0 must always mean ->group_exit_task != NULL is waiting for us. Signed-off-by:
-
Oleg Nesterov authored
- move the cprm.mm_flags checks up, before we take mmap_sem - move down_write(mmap_sem) and ->core_state check from do_coredump() to coredump_wait() This simplifies the code and makes the locking symmetrical. Signed-off-by:
-
Oleg Nesterov authored
Given that do_coredump() calls put_cred() on exit path, it is a bit ugly to do put_cred() + "goto fail" twice, just add the new "fail_creds" label. Signed-off-by:
-
Oleg Nesterov authored
- kill "int dump_count", argv_split(argcp) accepts argcp == NULL. - move "int dump_count" under " if (ispipe)" branch, fail_dropcount can check ispipe. - move "char **helper_argv" as well, change the code to do argv_free() right after call_usermodehelper_fns(). - If call_usermodehelper_fns() fails goto close_fail label instead of closing the file by hand. Signed-off-by:
-
Oleg Nesterov authored
do_coredump() does a lot of file checks after it opens the file or calls usermode helper. But all of these checks are only needed in !ispipe case. Move this code into the "else" branch and kill the ugly repetitive ispipe checks. Signed-off-by:
-
