Linus Torvalds [Wed, 1 May 2013 14:21:43 +0000 (07:21 -0700)]
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/signal
Pull compat cleanup from Al Viro:
"Mostly about syscall wrappers this time; there will be another pile
with patches in the same general area from various people, but I'd
rather push those after both that and vfs.git pile are in."
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/signal:
syscalls.h: slightly reduce the jungles of macros
get rid of union semop in sys_semctl(2) arguments
make do_mremap() static
sparc: no need to sign-extend in sync_file_range() wrapper
ppc compat wrappers for add_key(2) and request_key(2) are pointless
x86: trim sys_ia32.h
x86: sys32_kill and sys32_mprotect are pointless
get rid of compat_sys_semctl() and friends in case of ARCH_WANT_OLD_COMPAT_IPC
merge compat sys_ipc instances
consolidate compat lookup_dcookie()
convert vmsplice to COMPAT_SYSCALL_DEFINE
switch getrusage() to COMPAT_SYSCALL_DEFINE
switch epoll_pwait to COMPAT_SYSCALL_DEFINE
convert sendfile{,64} to COMPAT_SYSCALL_DEFINE
switch signalfd{,4}() to COMPAT_SYSCALL_DEFINE
make SYSCALL_DEFINE<n>-generated wrappers do asmlinkage_protect
make HAVE_SYSCALL_WRAPPERS unconditional
consolidate cond_syscall and SYSCALL_ALIAS declarations
teach SYSCALL_DEFINE<n> how to deal with long long/unsigned long long
get rid of duplicate logics in __SC_....[1-6] definitions
Linus Torvalds [Wed, 1 May 2013 00:37:43 +0000 (17:37 -0700)]
Merge branch 'akpm' (incoming from Andrew)
Merge third batch of fixes from Andrew Morton:
"Most of the rest. I still have two large patchsets against AIO and
IPC, but they're a bit stuck behind other trees and I'm about to
vanish for six days.
- random fixlets
- inotify
- more of the MM queue
- show_stack() cleanups
- DMI update
- kthread/workqueue things
- compat cleanups
- epoll udpates
- binfmt updates
- nilfs2
- hfs
- hfsplus
- ptrace
- kmod
- coredump
- kexec
- rbtree
- pids
- pidns
- pps
- semaphore tweaks
- some w1 patches
- relay updates
- core Kconfig changes
- sysrq tweaks"
* emailed patches from Andrew Morton <[email protected]>: (109 commits)
Documentation/sysrq: fix inconstistent help message of sysrq key
ethernet/emac/sysrq: fix inconstistent help message of sysrq key
sparc/sysrq: fix inconstistent help message of sysrq key
powerpc/xmon/sysrq: fix inconstistent help message of sysrq key
ARM/etm/sysrq: fix inconstistent help message of sysrq key
power/sysrq: fix inconstistent help message of sysrq key
kgdb/sysrq: fix inconstistent help message of sysrq key
lib/decompress.c: fix initconst
notifier-error-inject: fix module names in Kconfig
kernel/sys.c: make prctl(PR_SET_MM) generally available
UAPI: remove empty Kbuild files
menuconfig: print more info for symbol without prompts
init/Kconfig: re-order CONFIG_EXPERT options to fix menuconfig display
kconfig menu: move Virtualization drivers near other virtualization options
Kconfig: consolidate CONFIG_DEBUG_STRICT_USER_COPY_CHECKS
relay: use macro PAGE_ALIGN instead of FIX_SIZE
kernel/relay.c: move FIX_SIZE macro into relay.c
kernel/relay.c: remove unused function argument actor
drivers/w1/slaves/w1_ds2760.c: fix the error handling in w1_ds2760_add_slave()
drivers/w1/slaves/w1_ds2781.c: fix the error handling in w1_ds2781_add_slave()
...
Linus Torvalds [Wed, 1 May 2013 00:16:17 +0000 (17:16 -0700)]
Merge tag 'md-3.10' of git://neil.brown.name/md
Pull md fixes from NeilBrown:
"A mixed bag of little fixes. No real new functionality here. Several
patches are tagged for -stable."
* tag 'md-3.10' of git://neil.brown.name/md:
MD: ignore discard request for hard disks of hybid raid1/raid10 array
md: bad block list should default to disabled.
md: raid1/raid10 md devices leak memory when stopping
DM RAID: Add message/status support for changing sync action
MD: Export 'md_reap_sync_thread' function
md: don't update metadata when stopping a read-only array.
md: Allow devices to be re-added to a read-only array.
md/raid10: Allow skipping recovery when clean arrays are assembled
MD: Fix typos in MD documentation
md/raid5: avoid an extra write when writing to a known-bad-block.
md/raid5: Change or of some order to improve efficiency.
md: use set_bit_le and clear_bit_le
md: HOT_DISK_REMOVE shouldn't make a read-auto device active.
md: use common code for all calls to ->hot_remove_disk()
md: never update metadata when array is read-only.
This functionality is needed by any application or package that needs to
reconstruct Linux processes, that is, to start them in any way other than
by means of an "execve()" from an executable file. This includes:
1. Restoring processes from a checkpoint-file (by all potential
user-level checkpointing packages, not only CRIU's).
2. Restarting processes on another node after process migration.
3. Starting duplicated copies of a running process (for reliability
and high-availablity).
4. Starting a process from an executable format that is not supported
by Linux, thus requiring a "manual execve" by a user-level utility.
5. Similarly, starting a process from a networked and/or crypted
executable that, for confidentiality, licensing or other reasons,
may not be written to the local file-systems.
The code that does that was already included in the Linux kernel by the
CRIU group, in the form of "prctl(PR_SET_MM)", but prior to this was
enclosed within their private "#ifdef CONFIG_CHECKPOINT_RESTORE", which is
normally disabled. The patch removes those ifdefs.
menuconfig: print more info for symbol without prompts
When we search a config symbol, if it has no prompt the position of this
symbol in the Kconfig file and it's dependencies are not printed. This
can be inconvenient, especially when it's set to n and we want to find out
why.
the following is an example:
before:
Symbol: GENERIC_SMP_IDLE_THREAD [=y]
Type : boolean
Selected by: X86 [=y]
after:
Symbol: GENERIC_SMP_IDLE_THREAD [=y]
Type : boolean
Defined at arch/Kconfig:213
Selected by: X86 [=y]
Mike Frysinger [Tue, 30 Apr 2013 22:28:45 +0000 (15:28 -0700)]
init/Kconfig: re-order CONFIG_EXPERT options to fix menuconfig display
The kconfig language requires that dependent options all follow the
menuconfig symbol in order to be collapsed below it. Recently some hidden
options were added below the EXPERT menuconfig, but did not depend on
EXPERT (because hidden options can't). This broke the display. So
re-order all these options, and while we're here stick the PCI quirks
under the EXPERT menu (since it isn't sitting with any related options).
Before this commit, we get:
[*] Configure standard kernel features (expert users) --->
[ ] Sysctl syscall support
[*] Load all symbols for debugging/ksymoops
...
[ ] Embedded system
Now we get the older (and correct) behavior:
[*] Configure standard kernel features (expert users) --->
[ ] Embedded system
And if you go into the expert menu you get the expert options:
[ ] Sysctl syscall support
[*] Load all symbols for debugging/ksymoops
...
Randy Dunlap [Tue, 30 Apr 2013 22:28:44 +0000 (15:28 -0700)]
kconfig menu: move Virtualization drivers near other virtualization options
Make virtualization drivers be logically grouped together (physically
near each other) in the kconfig menu by moving "Virtualization drivers"
to be near "Virtio drivers", Microsort Hyper-V, and Xen driver support.
This is just a user-friendly, visual search change.
The help text for this config is duplicated across the x86, parisc, and
s390 Kconfig.debug files. Arnd Bergman noted that the help text was
slightly misleading and should be fixed to state that enabling this
option isn't a problem when using pre 4.4 gcc.
To simplify the rewording, consolidate the text into lib/Kconfig.debug
and modify it there to be more explicit about when you should say N to
this config.
Also, make the text a bit more generic by stating that this option
enables compile time checks so we can cover architectures which emit
warnings vs. ones which emit errors. The details of how an
architecture decided to implement the checks isn't as important as the
concept of compile time checking of copy_from_user() calls.
While we're doing this, remove all the copy_from_user_overflow() code
that's duplicated many times and place it into lib/ so that any
architecture supporting this option can get the function for free.
drivers/w1/slaves/w1_ds2760.c: fix the error handling in w1_ds2760_add_slave()
Use platform_device_put() instead of platform_device_unregister() if
platform_device_add() fail, and platform_device_del() should be used in
the error handling case after platform_device_add() success.
drivers/w1/slaves/w1_ds2781.c: fix the error handling in w1_ds2781_add_slave()
Use platform_device_put() instead of platform_device_unregister() if
platform_device_add() fail, and platform_device_del() should be used in
the error handling case after platform_device_add() success.
drivers/w1/slaves/w1_ds2780.c: fix the error handling in w1_ds2780_add_slave()
Use platform_device_put() instead of platform_device_unregister() if
platform_device_add() fail, and platform_device_del() should be used in
the error handling case after platform_device_add() success.
drivers/w1/slaves/w1_bq27000.c: fix the error handling in w1_bq27000_add_slave()
Use platform_device_put() instead of platform_device_unregister() if
platform_device_add() fails, and also add the return value check of
platform_device_add_data().
drivers/pps/kc.c:37:1: sparse: symbol 'pps_kc_hardpps_lock' was not declared. Should it be static?
drivers/pps/kc.c:39:19: sparse: symbol 'pps_kc_hardpps_dev' was not declared. Should it be static?
drivers/pps/kc.c:40:5: sparse: symbol 'pps_kc_hardpps_mode' was not declared. Should it be static?
pps: hide more configuration symbols behind CONFIG_PPS
Make CONFIG_PPS_DEBUG and CONFIG_NTP_PPS be hidden if CONFIG_PPS is not
selected, so that we are not prompted for these configuration options if
CONFIG_PPS is not set.
Michal Belczyk [Tue, 30 Apr 2013 22:28:28 +0000 (15:28 -0700)]
nbd: increase default and max request sizes
Raise the default max request size for nbd to 128KB (from 127KB) to get it
4KB aligned. This patch also allows the max request size to be increased
(via /sys/block/nbd<x>/queue/max_sectors_kb) to 32MB.
The patch makes nbd network traffic more efficient by:
- reducing request fragmentation (4KB alignment)
- reducing the number of requests (fewer round trips, less network overhead)
Especially in high latency networks, larger request size can make a dramatic
kernel/pid.c: improve flow of a loop inside alloc_pidmap.
find_next_offset() searches for an available "cleaned bit" in the
respective pid bitmap (page), so returns the offset if found, otherwise
it returns a value equals to BITS_PER_PAGE.
For example, suppose find_next_offset didn't find any available bit, so
there's no purpose to call mk_pid (Wasteful Cpu Cycles).
Therefore, I found it could be better to call mk_pid after the checking
(offset < BITS_PER_PAGE) returned sucessfully! Another point: If (offset
< BITS_PER_PAGE) results in a "failure", then mk_pid would be called
again afterwards.
exec: do not abuse ->cred_guard_mutex in threadgroup_lock()
threadgroup_lock() takes signal->cred_guard_mutex to ensure that
thread_group_leader() is stable. This doesn't look nice, the scope of
this lock in do_execve() is huge.
And as Dave pointed out this can lead to deadlock, we have the
following dependencies:
Change de_thread() to take threadgroup_change_begin() around the
switch-the-leader code and change threadgroup_lock() to avoid
->cred_guard_mutex.
Note that de_thread() can't sleep with ->group_rwsem held, this can
obviously deadlock with the exiting leader if the writer is active, so it
does threadgroup_change_end() before schedule().
David Rientjes [Tue, 30 Apr 2013 22:28:18 +0000 (15:28 -0700)]
fs, proc: truncate /proc/pid/comm writes to first TASK_COMM_LEN bytes
Currently, a write to a procfs file will return the number of bytes
successfully written. If the actual string is longer than this, the
remainder of the string will not be be written and userspace will
complete the operation by issuing additional write()s.
Hence
$ echo -n "abcdefghijklmnopqrs" > /proc/self/comm
results in
$ cat /proc/$$/comm
pqrs
since the final four bytes were written with a second write() since
TASK_COMM_LEN == 16. This is obviously an undesired result and not
equivalent to prctl(PR_SET_NAME). The implementation should not need to
know the definition of TASK_COMM_LEN.
This patch truncates the string to the first TASK_COMM_LEN bytes and
returns the bytes written as the length of the string written so the
second write() is suppressed.
coredump: change wait_for_dump_helpers() to use wait_event_interruptible()
wait_for_dump_helpers() calls wake_up/kill_fasync from inside the
wait_event-like loop. This is not needed and in fact this is not
strictly correct, we can/should do this only once after we change
pipe->writers. We could even check if it becomes zero.
Change this code to use use wait_event_interruptible(), this can also
help to make this wait freezable.
With this patch we check pipe->readers without pipe_lock(), this is
fine. Once we see pipe->readers == 1 we know that the handler
decremented the counter, this is all we need.
Change dump_write(), dump_seek() and do_coredump() to check
signal_pending() and abort if it is true. dump_seek() does this only
before f_op->llseek(), otherwise it relies on dump_write().
We need this change to ensure that the coredump won't delay suspend, and
to ensure it reacts to SIGKILL "quickly enough", a core dump can take a
lot of time. In particular this can help oom-killer.
We add the new trivial helper, dump_interrupted() to add the comments and
to simplify the potential freezer changes. Perhaps it will have more
callers.
Ideally it should do try_to_freeze() but then we need the unpleasant
changes in dump_write() and wait_for_dump_helpers(). It is not trivial to
change dump_write() to restart if f_op->write() fails because of
freezing(). We need to handle the short writes, we need to clear
TIF_SIGPENDING (and we can't rely on recalc_sigpending() unless we change
it to check PF_DUMPCORE). And if the buggy f_op->write() sets
TIF_SIGPENDING we can not distinguish this case from the race with
freeze_task() + __thaw_task().
So we simply accept the fact that the freezer can truncate a core-dump but
at least you can reliably suspend. Hopefully we can tolerate this
unlikely case and the necessary complications doesn't worth a trouble.
But if we decide to make the coredumping freezable later we can do this on
top of this change.
coredump: sanitize the setting of signal->group_exit_code
Now that the coredumping process can be SIGKILL'ed, the setting of
->group_exit_code in do_coredump() can race with complete_signal() and
SIGKILL or 0x80 can be "lost", or wait(status) can report status ==
SIGKILL | 0x80.
But the main problem is that it is not clear to me what should we do if
binfmt->core_dump() succeeds but SIGKILL was sent, that is why this patch
comes as a separate change.
This patch adds 0x80 if ->core_dump() succeeds and the process was not
killed. But perhaps we can (should?) re-set ->group_exit_code changed by
SIGKILL back to "siginfo->si_signo |= 0x80" in case when core_dumped == T.
coredump: ensure that SIGKILL always kills the dumping thread
prepare_signal() blesses SIGKILL sent to the dumping process but this
signal can be "lost" anyway. The problems is, complete_signal() sees
SIGNAL_GROUP_EXIT and skips the "kill them all" logic. And even if the
dumping process is single-threaded (so the target is always "correct"),
the group-wide SIGKILL is not recorded in task->pending and thus
__fatal_signal_pending() won't be true. A multi-threaded case has even
more problems.
And even ignoring all technical details, SIGNAL_GROUP_EXIT doesn't look
right to me. This coredumping process is not exiting yet, it can do a lot
of work dumping the core.
With this patch the dumping process doesn't have SIGNAL_GROUP_EXIT, we set
signal->group_exit_task instead. This makes signal_group_exit() true and
thus this should equally close the races with exit/exec/stop but allows to
kill the dumping thread reliably.
Notes:
- It is not clear what should we do with ->group_exit_code
if the dumper was killed, see the next change.
- we need more (hopefully straightforward) changes to ensure
that SIGKILL actually interrupts the coredump. Basically we
need to check __fatal_signal_pending() in dump_write() and
dump_seek().
coredump: only SIGKILL should interrupt the coredumping task
There are 2 well known and ancient problems with coredump/signals, and a
lot of related bug reports:
- do_coredump() clears TIF_SIGPENDING but of course this can't help
if, say, SIGCHLD comes after that.
In this case the coredump can fail unexpectedly. See for example
wait_for_dump_helper()->signal_pending() check but there are other
reasons.
- At the same time, dumping a huge core on the slow media can take a
lot of time/resources and there is no way to kill the coredumping
task reliably. In particular this is not oom_kill-friendly.
This patch tries to fix the 1st problem, and makes the preparation for the
next changes.
We add the new SIGNAL_GROUP_COREDUMP flag set by zap_threads() to indicate
that this process dumps the core. prepare_signal() checks this flag and
nacks any signal except SIGKILL.
Note that this check tries to be conservative, in the long term we should
probably treat the SIGNAL_GROUP_EXIT case equally but this needs more
discussion. See marc.info/?l=linux-kernel&m=120508897917439
Notes:
- recalc_sigpending() doesn't check SIGNAL_GROUP_COREDUMP.
The patch assumes that dump_write/etc paths should never
call it, but we can change it as well.
- There is another source of TIF_SIGPENDING, freezer. This
will be addressed separately.
Lucas De Marchi [Tue, 30 Apr 2013 22:28:09 +0000 (15:28 -0700)]
kmod: remove call_usermodehelper_fns()
This function suffers from not being able to determine if the cleanup is
called in case it returns -ENOMEM. Nobody is using it anymore, so let's
remove it.
Lucas De Marchi [Tue, 30 Apr 2013 22:28:07 +0000 (15:28 -0700)]
usermodehelper: split remaining calls to call_usermodehelper_fns()
These are the only users of call_usermodehelper_fns(). This function
suffers from not being able to determine if the cleanup is called. Even
if in this places the cleanup pointer is NULL, convert them to use the
separate call_usermodehelper_setup() + call_usermodehelper_exec()
functions so we can remove the _fns variant.
Lucas De Marchi [Tue, 30 Apr 2013 22:28:05 +0000 (15:28 -0700)]
KEYS: split call to call_usermodehelper_fns()
Use call_usermodehelper_setup() + call_usermodehelper_exec() instead of
calling call_usermodehelper_fns(). In case there's an OOM in this last
function the cleanup function may not be called - in this case we would
miss a call to key_put().
Lucas De Marchi [Tue, 30 Apr 2013 22:28:03 +0000 (15:28 -0700)]
kmod: split call to call_usermodehelper_fns()
Use call_usermodehelper_setup() + call_usermodehelper_exec() instead of
calling call_usermodehelper_fns(). In case the latter returns -ENOMEM the
cleanup function may had not been called - in this case we would not free
argv and module_name.
Lucas De Marchi [Tue, 30 Apr 2013 22:28:02 +0000 (15:28 -0700)]
usermodehelper: export call_usermodehelper_exec() and call_usermodehelper_setup()
call_usermodehelper_setup() + call_usermodehelper_exec() need to be
called instead of call_usermodehelper_fns() when the cleanup function
needs to be called even when an ENOMEM error occurs. In this case using
call_usermodehelper_fns() the user can't distinguish if the cleanup
function was called or not.
* Dump signals from process-wide and per-thread queues with
different sizes of buffers.
* Check error paths for buffers with restricted permissions. A part of
buffer or a whole buffer is for read-only.
* Try to get nonexistent signal.
ptrace: add ability to retrieve signals without removing from a queue (v4)
This patch adds a new ptrace request PTRACE_PEEKSIGINFO.
This request is used to retrieve information about pending signals
starting with the specified sequence number. Siginfo_t structures are
copied from the child into the buffer starting at "data".
The argument "addr" is a pointer to struct ptrace_peeksiginfo_args.
struct ptrace_peeksiginfo_args {
u64 off; /* from which siginfo to start */
u32 flags;
s32 nr; /* how may siginfos to take */
};
"nr" has type "s32", because ptrace() returns "long", which has 32 bits on
i386 and a negative values is used for errors.
Currently here is only one flag PTRACE_PEEKSIGINFO_SHARED for dumping
signals from process-wide queue. If this flag is not set, signals are
read from a per-thread queue.
The request PTRACE_PEEKSIGINFO returns a number of dumped signals. If a
signal with the specified sequence number doesn't exist, ptrace returns
zero. The request returns an error, if no signal has been dumped.
Errors:
EINVAL - one or more specified flags are not supported or nr is negative
EFAULT - buf or addr is outside your accessible address space.
A result siginfo contains a kernel part of si_code which usually striped,
but it's required for queuing the same siginfo back during restore of
pending signals.
This functionality is required for checkpointing pending signals. Pedro
Alves suggested using it in "gdb" to peek at pending signals. gdb already
uses PTRACE_GETSIGINFO to get the siginfo for the signal which was already
dequeued. This functionality allows gdb to look at the pending signals
which were not reported yet.
The prototype of this code was developed by Oleg Nesterov.
fs/hfsplus/bfind.c: In function 'hfs_find_1st_rec_by_cnid':
(1) include/uapi/linux/swab.h:60:2: warning: 'search_cnid' may be used uninitialized in this function [-Wmaybe-uninitialized]
(2) include/uapi/linux/swab.h:60:2: warning: 'cur_cnid' may be used uninitialized in this function [-Wmaybe-uninitialized]
hfs_find_init() may fail with ENOMEM, but there are places, where the
returned value is not checked. The consequences can be very unpleasant,
e.g. kfree uninitialized pointer and inappropriate mutex unlocking.
The patch adds checks for errors in hfs_find_init().
Found by Linux Driver Verification project (linuxtesting.org).
nilfs2: fix issue with flush kernel thread after remount in RO mode because of driver's internal error or metadata corruption
The NILFS2 driver remounts itself in RO mode in the case of discovering
metadata corruption (for example, discovering a broken bmap). But
usually, this takes place when there have been file system operations
before remounting in RO mode.
Thereby, NILFS2 driver can be in RO mode with presence of dirty pages in
modified inodes' address spaces. It results in flush kernel thread's
infinite trying to flush dirty pages in RO mode. As a result, it is
possible to see such side effects as: (1) flush kernel thread occupies
50% - 99% of CPU time; (2) system can't be shutdowned without manual
power switch off.
SYMPTOMS:
(1) System log contains error message: "Remounting filesystem read-only".
(2) The flush kernel thread occupies 50% - 99% of CPU time.
(3) The system can't be shutdowned without manual power switch off.
REPRODUCTION PATH:
(1) Create volume group with name "unencrypted" by means of vgcreate utility.
(2) Run script (prepared by Anthony Doggett <[email protected]>):
This patch implements checking mount state of NILFS2 driver in
nilfs_writepage(), nilfs_writepages() and nilfs_mdt_write_page()
methods. If it is detected the RO mount state then all dirty pages are
simply discarded with warning messages is written in system log.
Move the calls to memcpy_fromio() up into the loop in
dmi_scan_machine(), and move the signature checks back down into
dmi_decode(). We need to check at 16-byte intervals but keep a 32-byte
buffer for an SMBIOS entry, so shift the buffer after each iteration.
Merge smbios_present() into dmi_present(), so we look for an SMBIOS
signature at the beginning of the given buffer and then for a DMI
signature at an offset of 16 bytes.
binfmt_elf: PIE: make PF_RANDOMIZE check comment more accurate
The comment I originally added in commit a3defbe5c337 ("binfmt_elf: fix
PIE execution with randomization disabled") is not really 100% accurate
-- sysctl is not the only way how PF_RANDOMIZE could be forcibly unset
in runtime.
Another option of course is direct modification of personality flags
(i.e. running through setarch wrapper).
fs: make binfmt support for #! scripts modular and removable
Add a new configuration option CONFIG_BINFMT_SCRIPT to configure support
for interpreted scripts starting with "#!"; allow compiling out that
support, or building it as a module. Embedded systems running exclusively
compiled binaries could leave this support out, and systems that don't
need scripts before mounting the root filesystem can build this as a
module.
Eric Wong [Tue, 30 Apr 2013 22:27:40 +0000 (15:27 -0700)]
epoll: lock ep->mtx in ep_free to silence lockdep
Technically we do not need to hold ep->mtx during ep_free since we are
certain there are no other users of ep at that point. However, lockdep
complains with a "suspicious rcu_dereference_check() usage!" message; so
lock the mutex before ep_remove to silence the warning.
Eric Wong [Tue, 30 Apr 2013 22:27:38 +0000 (15:27 -0700)]
epoll: trim epitem by one cache line
It is common for epoll users to have thousands of epitems, so saving a
cache line on every allocation leads to large memory savings.
Since epitem allocations are cache-aligned, reducing sizeof(struct
epitem) from 136 bytes to 128 bytes will allow it to squeeze under a
cache line boundary on x86_64.
Via /sys/kernel/slab/eventpoll_epi, I see the following changes on my
x86_64 Core2 Duo (which has 64-byte cache alignment):
object_size : 192 => 128
objs_per_slab: 21 => 32
Also, add a BUILD_BUG_ON() to check for future accidental breakage.
Fan Du [Tue, 30 Apr 2013 22:27:27 +0000 (15:27 -0700)]
include/linux/fs.h: disable preempt when acquire i_size_seqcount write lock
Two rt tasks bind to one CPU core.
The higher priority rt task A preempts a lower priority rt task B which
has already taken the write seq lock, and then the higher priority rt
task A try to acquire read seq lock, it's doomed to lockup.
rt task A with lower priority: call write
i_size_write rt task B with higher priority: call sync, and preempt task A
write_seqcount_begin(&inode->i_size_seqcount); i_size_read
inode->i_size = i_size; read_seqcount_begin <-- lockup here...
So disable preempt when acquiring every i_size_seqcount *write* lock will
cure the problem.
csd_lock() uses assignment to data->flags rather than |=. That is not
buggy at present because only one bit (CSD_FLAG_LOCK) is defined in
call_single_data.flags.
But it will become buggy if we later add another flag, so fix it now.
writeback: set worker desc to identify writeback workers in task dumps
Writeback has been recently converted to use workqueue instead of its
private thread pool implementation. One negative side effect of this
conversion is that there's no easy to tell which backing device a
writeback work item was working on at the time of task dump, be it
sysrq-t, BUG, WARN or whatever, which, according to our writeback
brethren, is important in tracking down issues with a lot of mounted
file systems on a lot of different devices.
This patch restores that information using the new worker description
facility. bdi_writeback_workfn() calls set_work_desc() to identify
which bdi it's working on. The description is printed out together with
the worqueue name and worker function as in the following example dump.
workqueue: include workqueue info when printing debug dump of a worker task
One of the problems that arise when converting dedicated custom
threadpool to workqueue is that the shared worker pool used by workqueue
anonimizes each worker making it more difficult to identify what the
worker was doing on which target from the output of sysrq-t or debug
dump from oops, BUG() and friends.
This patch implements set_worker_desc() which can be called from any
workqueue work function to set its description. When the worker task is
dumped for whatever reason - sysrq-t, WARN, BUG, oops, lockdep assertion
and so on - the description will be printed out together with the
workqueue name and the worker function pointer.
The printing side is implemented by print_worker_info() which is called
from functions in task dump paths - sched_show_task() and
dump_stack_print_info(). print_worker_info() can be safely called on
any task in any state as long as the task struct itself is accessible.
It uses probe_*() functions to access worker fields. It may print
garbage if something went very wrong, but it wouldn't cause (another)
oops.
The description is currently limited to 24bytes including the
terminating \0. worker->desc_valid and workder->desc[] are added and
the 64 bytes marker which was already incorrect before adding the new
fields is moved to the correct position.
Here's an example dump with writeback updated to set the bdi name as
worker desc.
One of the problems that arise when converting dedicated custom threadpool
to workqueue is that the shared worker pool used by workqueue anonimizes
each worker making it more difficult to identify what the worker was doing
on which target from the output of sysrq-t or debug dump from oops, BUG()
and friends.
For example, after writeback is converted to use workqueue instead of
priviate thread pool, there's no easy to tell which backing device a
writeback work item was working on at the time of task dump, which,
according to our writeback brethren, is important in tracking down issues
with a lot of mounted file systems on a lot of different devices.
This patchset implements a way for a work function to mark its execution
instance so that task dump of the worker task includes information to
indicate what the work item was doing.
An example WARN dump would look like the following.
Implement probe_kthread_data() which returns kthread_data if accessible.
The function is equivalent to kthread_data() except that the specified
@task may not be a kthread or its vfork_done is already cleared rendering
struct kthread inaccessible. In the former case, probe_kthread_data() may
return any value. In the latter, NULL.
This will be used to safely print debug information without affecting
synchronization in the normal paths. Workqueue debug info printing on
dump_stack() and friends will make use of it.
dump_stack: unify debug information printed by show_regs()
show_regs() is inherently arch-dependent but it does make sense to print
generic debug information and some archs already do albeit in slightly
different forms. This patch introduces a generic function to print debug
information from show_regs() so that different archs print out the same
information and it's much easier to modify what's printed.
show_regs_print_info() prints out the same debug info as dump_stack()
does plus task and thread_info pointers.
* Already prints debug info. Replaced with show_regs_print_info().
The printed information is superset of what used to be there.
arm, arm64, avr32, mips, powerpc, sh32, tile, unicore32, x86
* s390 is special in that it used to print arch-specific information
along with generic debug info. Heiko and Martin think that the
arch-specific extra isn't worth keeping s390 specfic implementation.
Converted to use the generic version.
Note that now all archs print the debug info before actual register
dumps.
v3: CPU number dropped from show_regs_print_info() as
dump_stack_print_info() has been updated to print it. s390
specific implementation dropped as requested by s390 maintainers.
dump_stack: implement arch-specific hardware description in task dumps
x86 and ia64 can acquire extra hardware identification information
from DMI and print it along with task dumps; however, the usage isn't
consistent.
* x86 show_regs() collects vendor, product and board strings and print
them out with PID, comm and utsname. Some of the information is
printed again later in the same dump.
* warn_slowpath_common() explicitly accesses the DMI board and prints
it out with "Hardware name:" label. This applies to both x86 and
ia64 but is irrelevant on all other archs.
* ia64 doesn't show DMI information on other non-WARN dumps.
This patch introduces arch-specific hardware description used by
dump_stack(). It can be set by calling dump_stack_set_arch_desc()
during boot and, if exists, printed out in a separate line with
"Hardware name:" label.
dmi_set_dump_stack_arch_desc() is added which sets arch-specific
description from DMI data. It uses dmi_ids_string[] which is set from
dmi_present() used for DMI debug message. It is superset of the
information x86 show_regs() is using. The function is called from x86
and ia64 boot code right after dmi_scan_machine().
This makes the explicit DMI handling in warn_slowpath_common()
unnecessary. Removed.
show_regs() isn't yet converted to use generic debug information
printing and this patch doesn't remove the duplicate DMI handling in
x86 show_regs(). The next patch will unify show_regs() handling and
remove the duplication.
v2: Use the same string as the debug message from dmi_present() which
also contains BIOS information. Move hardware name into its own
line as warn_slowpath_common() did. This change was suggested by
Bjorn Helgaas.
dmi: morph dmi_dump_ids() into dmi_format_ids() which formats into a buffer
We're goning to use DMI identification for other purposes too. Morph
dmi_dump_ids() which is used to print DMI identification as a debug
message during boot into dmi_format_ids() which formats the same
information sans the leading "DMI:" tag into a string buffer.
dmi_present() is updated to format the information into dmi_ids_string[]
using the new function and print it with "DMI:" prefix.
dmi_ids_string[] will be used for another purpose by a future patch.
dump_stack: consolidate dump_stack() implementations and unify their behaviors
Both dump_stack() and show_stack() are currently implemented by each
architecture. show_stack(NULL, NULL) dumps the backtrace for the
current task as does dump_stack(). On some archs, dump_stack() prints
extra information - pid, utsname and so on - in addition to the
backtrace while the two are identical on other archs.
The usages in arch-independent code of the two functions indicate
show_stack(NULL, NULL) should print out bare backtrace while
dump_stack() is used for debugging purposes when something went wrong,
so it does make sense to print additional information on the task which
triggered dump_stack().
There's no reason to require archs to implement two separate but mostly
identical functions. It leads to unnecessary subtle information.
This patch expands the dummy fallback dump_stack() implementation in
lib/dump_stack.c such that it prints out debug information (taken from
x86) and invokes show_stack(NULL, NULL) and drops arch-specific
dump_stack() implementations in all archs except blackfin. Blackfin's
dump_stack() does something wonky that I don't understand.
Debug information can be printed separately by calling
dump_stack_print_info() so that arch-specific dump_stack()
implementation can still emit the same debug information. This is used
in blackfin.
This patch brings the following behavior changes.
* On some archs, an extra level in backtrace for show_stack() could be
printed. This is because the top frame was determined in
dump_stack() on those archs while generic dump_stack() can't do that
reliably. It can be compensated by inlining dump_stack() but not
sure whether that'd be necessary.
* Most archs didn't use to print debug info on dump_stack(). They do
now.
v2: CPU number added to the generic debug info as requested by s390
folks and dropped the s390 specific dump_stack(). This loses %ksp
from the debug message which the maintainers think isn't important
enough to keep the s390-specific dump_stack() implementation.
dump_stack_print_info() is moved to kernel/printk.c from
lib/dump_stack.c. Because linkage is per objecct file,
dump_stack_print_info() living in the same lib file as generic
dump_stack() means that archs which implement custom dump_stack()
- at this point, only blackfin - can't use dump_stack_print_info()
as that will bring in the generic version of dump_stack() too. v1
The v1 patch broke build on blackfin due to this issue. The build
breakage was reported by Fengguang Wu.
projects / linux.git / log