Move the debug_obj::object pointer into a union and add a pointer to the
last node in a batch. That allows to implement batch processing efficiently
by utilizing the stack property of hlist:
When the first object of a batch is added to the list, then the batch
pointer is set to the hlist node of the object itself. Any subsequent add
retrieves the pointer to the last node from the first object in the list
and uses that for storing the last node pointer in the newly added object.
Add the pointer to the data structure and ensure that all relevant pool
sizes are strictly batch sized. The actual batching implementation follows
in subsequent changes.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Zhen Lei <thunder.leizhen@huawei.com>
Link: https://lore.kernel.org/all/20241007164914.139204961@linutronix.de
Convert it to batch processing with intermediate helper functions. This
reduces the final changes for batch processing.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Zhen Lei <thunder.leizhen@huawei.com>
Link: https://lore.kernel.org/all/20241007164914.015906394@linutronix.de
__free_object() is uncomprehensibly complex. The same can be achieved by:
1) Adding the object to the per CPU pool
2) If that pool is full, move a batch of objects into the global pool
or if the global pool is full into the to free pool
This also prepares for batch processing.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Zhen Lei <thunder.leizhen@huawei.com>
Link: https://lore.kernel.org/all/20241007164913.955542307@linutronix.de
The current allocation scheme tries to allocate from the per CPU pool
first. If that fails it allocates one object from the global pool and then
refills the per CPU pool from the global pool.
That is in the way of switching the pool management to batch mode as the
global pool needs to be a strict stack of batches, which does not allow
to allocate single objects.
Rework the code to refill the per CPU pool first and then allocate the
object from the refilled batch. Also try to allocate from the to free pool
first to avoid freeing and reallocating objects.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Zhen Lei <thunder.leizhen@huawei.com>
Link: https://lore.kernel.org/all/20241007164913.893554162@linutronix.de
Having the accounting in the datastructure is better in terms of cache
lines and allows more optimizations later on.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Zhen Lei <thunder.leizhen@huawei.com>
Link: https://lore.kernel.org/all/20241007164913.831908427@linutronix.de
No point in having a separate data structure. Reuse struct obj_pool and
tidy up the code.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Zhen Lei <thunder.leizhen@huawei.com>
Link: https://lore.kernel.org/all/20241007164913.770595795@linutronix.de
There is no point to handle the statically allocated objects during early
boot in the actual pool list. This phase does not require accounting, so
all of the related complexity can be avoided.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Zhen Lei <thunder.leizhen@huawei.com>
Link: https://lore.kernel.org/all/20241007164913.708939081@linutronix.de
The contention on the global pool lock can be reduced by strict batch
processing where batches of objects are moved from one list head to another
instead of moving them object by object. This also reduces the cache
footprint because it avoids the list walk and dirties at maximum three
cache lines instead of potentially up to eighteen.
To prepare for that, move the hlist head and related counters into a
struct.
No functional change.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Zhen Lei <thunder.leizhen@huawei.com>
Link: https://lore.kernel.org/all/20241007164913.646171170@linutronix.de
The contention on the global pool_lock can be massive when the global pool
needs to be refilled and many CPUs try to handle this.
Address this by:
- splitting the refill from free list and allocation.
Refill from free list has no constraints vs. the context on RT, so
it can be tried outside of the RT specific preemptible() guard
- Let only one CPU handle the free list
- Let only one CPU do allocations unless the pool level is below
half of the minimum fill level.
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20240911083521.2257-4-thunder.leizhen@huawei.com-
Link: https://lore.kernel.org/all/20241007164913.582118421@linutronix.de
--
lib/debugobjects.c | 84 +++++++++++++++++++++++++++++++++++++----------------
1 file changed, 59 insertions(+), 25 deletions(-)
Freeing the per CPU pool of the unplugged CPU directly is suboptimal as the
objects can be reused in the real pool if there is room. Aside of that this
gets the accounting wrong.
Use the regular free path, which allows reuse and has the accounting correct.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Zhen Lei <thunder.leizhen@huawei.com>
Link: https://lore.kernel.org/all/20241007164913.263960570@linutronix.de
debug_objects_mem_init() is invoked from mm_core_init() before work queues
are available. If debug_objects_mem_init() destroys the kmem cache in the
error path it causes an Oops in __queue_work():
Oops: Oops: 0000 [#1] PREEMPT SMP PTI
RIP: 0010:__queue_work+0x35/0x6a0
queue_work_on+0x66/0x70
flush_all_cpus_locked+0xdf/0x1a0
__kmem_cache_shutdown+0x2f/0x340
kmem_cache_destroy+0x4e/0x150
mm_core_init+0x9e/0x120
start_kernel+0x298/0x800
x86_64_start_reservations+0x18/0x30
x86_64_start_kernel+0xc5/0xe0
common_startup_64+0x12c/0x138
Further the object cache pointer is used in various places to check for
early boot operation. It is exposed before the replacments for the static
boot time objects are allocated and the self test operates on it.
This can be avoided by:
1) Running the self test with the static boot objects
2) Exposing it only after the replacement objects have been added to
the pool.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20241007164913.137021337@linutronix.de
The statically allocated objects are all located in obj_static_pool[],
the whole memory of obj_static_pool[] will be reclaimed later. Therefore,
there is no need to split the remaining statically nodes in list obj_pool
into isolated ones, no one will use them anymore. Just write
INIT_HLIST_HEAD(&obj_pool) is enough. Since hlist_move_list() directly
discards the old list, even this can be omitted.
Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20240911083521.2257-2-thunder.leizhen@huawei.com
Link: https://lore.kernel.org/all/20241007164913.009849239@linutronix.de
fill_pool() checks locklessly at the beginning whether the pool has to be
refilled. After that it checks locklessly in a loop whether the free list
contains objects and repeats the refill check.
If both conditions are true, it acquires the pool lock and tries to move
objects from the free list to the pool repeating the same checks again.
There are two redundant issues with that:
1) The repeated check for the fill condition
2) The loop processing
The repeated check is pointless as it was just established that fill is
required. The condition has to be re-evaluated under the lock anyway.
The loop processing is not required either because there is practically
zero chance that a repeated attempt will succeed if the checks under the
lock terminate the moving of objects.
Remove the redundant check and replace the loop with a simple if condition.
[ tglx: Massaged change log ]
Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20240904133944.2124-4-thunder.leizhen@huawei.com
fill_pool() uses 'obj_pool_min_free' to decide whether objects should be
handed back to the kmem cache. But 'obj_pool_min_free' records the lowest
historical value of the number of objects in the object pool and not the
minimum number of objects which should be kept in the pool.
Use 'debug_objects_pool_min_level' instead, which holds the minimum number
which was scaled to the number of CPUs at boot time.
[ tglx: Massage change log ]
Fixes: d26bf5056f ("debugobjects: Reduce number of pool_lock acquisitions in fill_pool()")
Fixes: 36c4ead6f6 ("debugobjects: Add global free list and the counter")
Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/all/20240904133944.2124-3-thunder.leizhen@huawei.com
1. Both debug_objects_pool_min_level and debug_objects_pool_size are
read-only after initialization, change attribute '__read_mostly' to
'__ro_after_init', and remove '__data_racy'.
2. Many global variables are read in the debug_stats_show() function, but
didn't mask KCSAN's detection. Add '__data_racy' for them.
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Zhen Lei <thunder.leizhen@huawei.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20240904133944.2124-2-thunder.leizhen@huawei.com
KCSAN has identified a potential data race in debugobjects, where the
global variable debug_objects_maxchain is accessed for both reading and
writing simultaneously in separate and parallel data paths. This results in
the following splat printed by KCSAN:
BUG: KCSAN: data-race in debug_check_no_obj_freed / debug_object_activate
write to 0xffffffff847ccfc8 of 4 bytes by task 734 on cpu 41:
debug_object_activate (lib/debugobjects.c:199 lib/debugobjects.c:564 lib/debugobjects.c:710)
call_rcu (kernel/rcu/rcu.h:227 kernel/rcu/tree.c:2719 kernel/rcu/tree.c:2838)
security_inode_free (security/security.c:1626)
__destroy_inode (./include/linux/fsnotify.h:222 fs/inode.c:287)
...
read to 0xffffffff847ccfc8 of 4 bytes by task 384 on cpu 31:
debug_check_no_obj_freed (lib/debugobjects.c:1000 lib/debugobjects.c:1019)
kfree (mm/slub.c:2081 mm/slub.c:4280 mm/slub.c:4390)
percpu_ref_exit (lib/percpu-refcount.c:147)
css_free_rwork_fn (kernel/cgroup/cgroup.c:5357)
...
value changed: 0x00000070 -> 0x00000071
The data race is actually harmless as this is just used for debugfs
statistics, as all other debug variables.
Annotate all debug variables as racy explicitly, since these variables
are known to be racy and harmless.
Signed-off-by: Breno Leitao <leitao@debian.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/r/20240611091813.1189860-1-leitao@debian.org
After release of the hashbucket lock the tracking object can be modified or
freed by a concurrent thread. Using it in such a case is error prone, even
for printing the object state:
1. T1 tries to deactivate destroyed object, debugobjects detects it,
hash bucket lock is released.
2. T2 preempts T1 and frees the tracking object.
3. The freed tracking object is allocated and initialized for a
different to be tracked kernel object.
4. T1 resumes and reports error for wrong kernel object.
Create a local copy of the tracking object before releasing the hash bucket
lock and use the local copy for reporting and fixups to prevent this.
Signed-off-by: Andrzej Hajda <andrzej.hajda@intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/r/20231025-debugobjects_fix-v3-1-2bc3bf7084c2@intel.com
__read_mostly predates __ro_after_init. Many variables which are marked
__read_mostly should have been __ro_after_init from day 1.
Also, mark some stuff as "const" and "__init" while I'm at it.
[akpm@linux-foundation.org: revert sysctl_nr_open_min, sysctl_nr_open_max changes due to arm warning]
[akpm@linux-foundation.org: coding-style cleanups]
Link: https://lkml.kernel.org/r/4f6bb9c0-abba-4ee4-a7aa-89265e886817@p183
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
syzbot is reporting a lockdep warning in fill_pool() because the allocation
from debugobjects is using GFP_ATOMIC, which is (__GFP_HIGH | __GFP_KSWAPD_RECLAIM)
and therefore tries to wake up kswapd, which acquires kswapd_wait::lock.
Since fill_pool() might be called with arbitrary locks held, fill_pool()
should not assume that acquiring kswapd_wait::lock is safe.
Use __GFP_HIGH instead and remove __GFP_NORETRY as it is pointless for
!__GFP_DIRECT_RECLAIM allocation.
Fixes: 3ac7fe5a4a ("infrastructure to debug (dynamic) objects")
Reported-by: syzbot <syzbot+fe0c72f0ccbb93786380@syzkaller.appspotmail.com>
Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/r/6577e1fa-b6ee-f2be-2414-a2b51b1c5e30@I-love.SAKURA.ne.jp
Closes: https://syzkaller.appspot.com/bug?extid=fe0c72f0ccbb93786380
There is an explicit wait-type violation in debug_object_fill_pool()
for PREEMPT_RT=n kernels which allows them to more easily fill the
object pool and reduce the chance of allocation failures.
Lockdep's wait-type checks are designed to check the PREEMPT_RT
locking rules even for PREEMPT_RT=n kernels and object to this, so
create a lockdep annotation to allow this to stand.
Specifically, create a 'lock' type that overrides the inner wait-type
while it is held -- allowing one to temporarily raise it, such that
the violation is hidden.
Reported-by: Vlastimil Babka <vbabka@suse.cz>
Reported-by: Qi Zheng <zhengqi.arch@bytedance.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: Qi Zheng <zhengqi.arch@bytedance.com>
Link: https://lkml.kernel.org/r/20230429100614.GA1489784@hirez.programming.kicks-ass.net
The recent fix to ensure atomicity of lookup and allocation inadvertently
broke the pool refill mechanism.
Prior to that change debug_objects_activate() and debug_objecs_assert_init()
invoked debug_objecs_init() to set up the tracking object for statically
initialized objects. That's not longer the case and debug_objecs_init() is
now the only place which does pool refills.
Depending on the number of statically initialized objects this can be
enough to actually deplete the pool, which was observed by Ido via a
debugobjects OOM warning.
Restore the old behaviour by adding explicit refill opportunities to
debug_objects_activate() and debug_objecs_assert_init().
Fixes: 63a759694e ("debugobject: Prevent init race with static objects")
Reported-by: Ido Schimmel <idosch@nvidia.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Ido Schimmel <idosch@nvidia.com>
Link: https://lore.kernel.org/r/871qk05a9d.ffs@tglx
Statically initialized objects are usually not initialized via the init()
function of the subsystem. They are special cased and the subsystem
provides a function to validate whether an object which is not yet tracked
by debugobjects is statically initialized. This means the object is started
to be tracked on first use, e.g. activation.
This works perfectly fine, unless there are two concurrent operations on
that object. Schspa decoded the problem:
T0 T1
debug_object_assert_init(addr)
lock_hash_bucket()
obj = lookup_object(addr);
if (!obj) {
unlock_hash_bucket();
- > preemption
lock_subsytem_object(addr);
activate_object(addr)
lock_hash_bucket();
obj = lookup_object(addr);
if (!obj) {
unlock_hash_bucket();
if (is_static_object(addr))
init_and_track(addr);
lock_hash_bucket();
obj = lookup_object(addr);
obj->state = ACTIVATED;
unlock_hash_bucket();
subsys function modifies content of addr,
so static object detection does
not longer work.
unlock_subsytem_object(addr);
if (is_static_object(addr)) <- Fails
debugobject emits a warning and invokes the fixup function which
reinitializes the already active object in the worst case.
This race exists forever, but was never observed until mod_timer() got a
debug_object_assert_init() added which is outside of the timer base lock
held section right at the beginning of the function to cover the lockless
early exit points too.
Rework the code so that the lookup, the static object check and the
tracking object association happens atomically under the hash bucket
lock. This prevents the issue completely as all callers are serialized on
the hash bucket lock and therefore cannot observe inconsistent state.
Fixes: 3ac7fe5a4a ("infrastructure to debug (dynamic) objects")
Reported-by: syzbot+5093ba19745994288b53@syzkaller.appspotmail.com
Debugged-by: Schspa Shi <schspa@gmail.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Stephen Boyd <swboyd@chromium.org>
Link: https://syzkaller.appspot.com/bug?id=22c8a5938eab640d1c6bcc0e3dc7be519d878462
Link: https://lore.kernel.org/lkml/20230303161906.831686-1-schspa@gmail.com
Link: https://lore.kernel.org/r/87zg7dzgao.ffs@tglx
- A ptrace API cleanup series from Sergey Shtylyov
- Fixes and cleanups for kexec from ye xingchen
- nilfs2 updates from Ryusuke Konishi
- squashfs feature work from Xiaoming Ni: permit configuration of the
filesystem's compression concurrency from the mount command line.
- A series from Akinobu Mita which addresses bound checking errors when
writing to debugfs files.
- A series from Yang Yingliang to address rapido memory leaks
- A series from Zheng Yejian to address possible overflow errors in
encode_comp_t().
- And a whole shower of singleton patches all over the place.
-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQTTMBEPP41GrTpTJgfdBJ7gKXxAjgUCY5efRgAKCRDdBJ7gKXxA
jgvdAP0al6oFDtaSsshIdNhrzcMwfjt6PfVxxHdLmNhF1hX2dwD/SVluS1bPSP7y
0sZp7Ustu3YTb8aFkMl96Y9m9mY1Nwg=
=ga5B
-----END PGP SIGNATURE-----
Merge tag 'mm-nonmm-stable-2022-12-12' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm
Pull non-MM updates from Andrew Morton:
- A ptrace API cleanup series from Sergey Shtylyov
- Fixes and cleanups for kexec from ye xingchen
- nilfs2 updates from Ryusuke Konishi
- squashfs feature work from Xiaoming Ni: permit configuration of the
filesystem's compression concurrency from the mount command line
- A series from Akinobu Mita which addresses bound checking errors when
writing to debugfs files
- A series from Yang Yingliang to address rapidio memory leaks
- A series from Zheng Yejian to address possible overflow errors in
encode_comp_t()
- And a whole shower of singleton patches all over the place
* tag 'mm-nonmm-stable-2022-12-12' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm: (79 commits)
ipc: fix memory leak in init_mqueue_fs()
hfsplus: fix bug causing custom uid and gid being unable to be assigned with mount
rapidio: devices: fix missing put_device in mport_cdev_open
kcov: fix spelling typos in comments
hfs: Fix OOB Write in hfs_asc2mac
hfs: fix OOB Read in __hfs_brec_find
relay: fix type mismatch when allocating memory in relay_create_buf()
ocfs2: always read both high and low parts of dinode link count
io-mapping: move some code within the include guarded section
kernel: kcsan: kcsan_test: build without structleak plugin
mailmap: update email for Iskren Chernev
eventfd: change int to __u64 in eventfd_signal() ifndef CONFIG_EVENTFD
rapidio: fix possible UAF when kfifo_alloc() fails
relay: use strscpy() is more robust and safer
cpumask: limit visibility of FORCE_NR_CPUS
acct: fix potential integer overflow in encode_comp_t()
acct: fix accuracy loss for input value of encode_comp_t()
linux/init.h: include <linux/build_bug.h> and <linux/stringify.h>
rapidio: rio: fix possible name leak in rio_register_mport()
rapidio: fix possible name leaks when rio_add_device() fails
...
Delayed kobject debugging (CONFIG_DEBUG_KOBJECT_RELEASE) prints the kobject
pointer that's being released in kobject_release() before scheduling a
randomly delayed work to do the actual release work.
If the caller of kobject_put() frees the kobject upon return then this will
typically emit a debugobject warning about freeing an active timer.
Usually the release function is the function that does the kfree() of the
struct containing the kobject.
For example the following print is seen
kobject: 'queue' (ffff888114236190): kobject_release, parent 0000000000000000 (delayed 1000)
------------[ cut here ]------------
ODEBUG: free active (active state 0) object type: timer_list hint: kobject_delayed_cleanup+0x0/0x390
but the kobject printk cannot be matched with the debug object printk
because it could be any number of kobjects that was released around that
time. The random delay for the work doesn't help either.
Print the address of the object being tracked to help to figure out which
kobject is the problem here. Note that this does not use %px here to match
the other %p usage in debugobject debugging. Due to %p usage it is required
to disable pointer hashing to correlate the two pointer printks.
Signed-off-by: Stephen Boyd <swboyd@chromium.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Guenter Roeck <linux@roeck-us.net>
Link: https://lore.kernel.org/r/20220519202201.2348343-1-swboyd@chromium.org
1. Var debug_objects_allocated tracks valid kmem_cache_alloc calls, so
track it in debug_objects_replace_static_objects. Do similar things in
object_cpu_offline.
2. In debug_objects_mem_init, there is no need to call function
cpuhp_setup_state_nocalls when debug_objects_enabled = 0 (out of
memory).
Link: https://lkml.kernel.org/r/20220611130634.99741-1-wuchi.zero@gmail.com
Fixes: 634d61f45d ("debugobjects: Percpu pool lookahead freeing/allocation")
Fixes: c4b73aabd0 ("debugobjects: Track number of kmem_cache_alloc/kmem_cache_free done")
Signed-off-by: wuchi <wuchi.zero@gmail.com>
Reviewed-by: Waiman Long <longman@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Kees Cook <keescook@chromium.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
On PREEMPT_RT enabled kernels it is not possible to refill the object pool
from atomic context (preemption or interrupts disabled) as the allocator
might acquire 'sleeping' spinlocks.
Guard the invocation of fill_pool() accordingly.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Link: https://lore.kernel.org/r/87sfzehdnl.ffs@tglx
If a CPU is offlined the debug objects per CPU pool is not cleaned up. If
the CPU is never onlined again then the objects in the pool are wasted.
Add a CPU hotplug callback which is invoked after the CPU is dead to free
the pool.
[ tglx: Massaged changelog and added comment about remote access safety ]
Signed-off-by: Zqiang <qiang.zhang@windriver.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Link: https://lore.kernel.org/r/20200908062709.11441-1-qiang.zhang@windriver.com
The debugobject core could be slightly harder to corrupt if the
debug_obj_descr would be a pointer to const memory.
Depending on the architecture, const data structures are placed into
read-only memory and thus are harder to corrupt or hijack.
This descriptor is used to fix up stuff like timers and workqueues when
core kernel data structures are busted, so moving the descriptors to
read-only memory will make debugobjects more resilient to something going
wrong and then corrupting the function pointers inside struct
debug_obj_descr.
Signed-off-by: Stephen Boyd <swboyd@chromium.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Kees Cook <keescook@chromium.org>
Link: https://lore.kernel.org/r/20200815004027.2046113-2-swboyd@chromium.org
The counters obj_pool_free, and obj_nr_tofree, and the flag obj_freeing are
read locklessly outside the pool_lock critical sections. If read with plain
accesses, this would result in data races.
This is addressed as follows:
* reads outside critical sections become READ_ONCE()s (pairing with
WRITE_ONCE()s added);
* writes become WRITE_ONCE()s (pairing with READ_ONCE()s added); since
writes happen inside critical sections, only the write and not the read
of RMWs needs to be atomic, thus WRITE_ONCE(var, var +/- X) is
sufficient.
The data races were reported by KCSAN:
BUG: KCSAN: data-race in __free_object / fill_pool
write to 0xffffffff8beb04f8 of 4 bytes by interrupt on cpu 1:
__free_object+0x1ee/0x8e0 lib/debugobjects.c:404
__debug_check_no_obj_freed+0x199/0x330 lib/debugobjects.c:969
debug_check_no_obj_freed+0x3c/0x44 lib/debugobjects.c:994
slab_free_hook mm/slub.c:1422 [inline]
read to 0xffffffff8beb04f8 of 4 bytes by task 1 on cpu 2:
fill_pool+0x3d/0x520 lib/debugobjects.c:135
__debug_object_init+0x3c/0x810 lib/debugobjects.c:536
debug_object_init lib/debugobjects.c:591 [inline]
debug_object_activate+0x228/0x320 lib/debugobjects.c:677
debug_rcu_head_queue kernel/rcu/rcu.h:176 [inline]
BUG: KCSAN: data-race in __debug_object_init / fill_pool
read to 0xffffffff8beb04f8 of 4 bytes by task 10 on cpu 6:
fill_pool+0x3d/0x520 lib/debugobjects.c:135
__debug_object_init+0x3c/0x810 lib/debugobjects.c:536
debug_object_init_on_stack+0x39/0x50 lib/debugobjects.c:606
init_timer_on_stack_key kernel/time/timer.c:742 [inline]
write to 0xffffffff8beb04f8 of 4 bytes by task 1 on cpu 3:
alloc_object lib/debugobjects.c:258 [inline]
__debug_object_init+0x717/0x810 lib/debugobjects.c:544
debug_object_init lib/debugobjects.c:591 [inline]
debug_object_activate+0x228/0x320 lib/debugobjects.c:677
debug_rcu_head_queue kernel/rcu/rcu.h:176 [inline]
BUG: KCSAN: data-race in free_obj_work / free_object
read to 0xffffffff9140c190 of 4 bytes by task 10 on cpu 6:
free_object+0x4b/0xd0 lib/debugobjects.c:426
debug_object_free+0x190/0x210 lib/debugobjects.c:824
destroy_timer_on_stack kernel/time/timer.c:749 [inline]
write to 0xffffffff9140c190 of 4 bytes by task 93 on cpu 1:
free_obj_work+0x24f/0x480 lib/debugobjects.c:313
process_one_work+0x454/0x8d0 kernel/workqueue.c:2264
worker_thread+0x9a/0x780 kernel/workqueue.c:2410
Reported-by: Qian Cai <cai@lca.pw>
Signed-off-by: Marco Elver <elver@google.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/r/20200116185529.11026-1-elver@google.com
The db->lock is a raw spinlock and so the lock hold time is supposed
to be short. This will not be the case when printk() is being involved
in some of the critical sections. In order to avoid the long hold time,
in case some messages need to be printed, the debug_object_is_on_stack()
and debug_print_object() calls are now moved out of those critical
sections.
Signed-off-by: Waiman Long <longman@redhat.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Yang Shi <yang.shi@linux.alibaba.com>
Cc: "Joel Fernandes (Google)" <joel@joelfernandes.org>
Cc: Qian Cai <cai@gmx.us>
Cc: Zhong Jiang <zhongjiang@huawei.com>
Link: https://lkml.kernel.org/r/20190520141450.7575-6-longman@redhat.com
After a system bootup and 3 parallel kernel builds, a partial output
of the debug objects stats file was:
pool_free :5101
pool_pcp_free :4181
pool_min_free :220
pool_used :104172
pool_max_used :171920
on_free_list :0
objs_allocated:39268280
objs_freed :39160031
More than 39 millions debug objects had since been allocated and then
freed. The pool_max_used, however, was only about 172k. So this is a
lot of extra overhead in freeing and allocating objects from slabs. It
may also causes the slabs to be more fragmented and harder to reclaim.
Make the freeing of excess debug objects less aggressive by freeing them at
a maximum frequency of 10Hz and about 1k objects at each round of freeing.
With that change applied, the partial output of the debug objects stats
file after similar actions became:
pool_free :5901
pool_pcp_free :3742
pool_min_free :1022
pool_used :104805
pool_max_used :168081
on_free_list :0
objs_allocated:5796864
objs_freed :5687182
Signed-off-by: Waiman Long <longman@redhat.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Yang Shi <yang.shi@linux.alibaba.com>
Cc: "Joel Fernandes (Google)" <joel@joelfernandes.org>
Cc: Qian Cai <cai@gmx.us>
Cc: Zhong Jiang <zhongjiang@huawei.com>
Link: https://lkml.kernel.org/r/20190520141450.7575-5-longman@redhat.com
In fill_pool(), the pool_lock is acquired and then released once per debug
object. If many objects are to be filled, the constant lock and unlock
operations are extra overhead.
To reduce the overhead, batch them up and do an allocation of 4 objects per
lock/unlock sequence.
Signed-off-by: Waiman Long <longman@redhat.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Yang Shi <yang.shi@linux.alibaba.com>
Cc: "Joel Fernandes (Google)" <joel@joelfernandes.org>
Cc: Qian Cai <cai@gmx.us>
Cc: Zhong Jiang <zhongjiang@huawei.com>
Link: https://lkml.kernel.org/r/20190520141450.7575-4-longman@redhat.com
Most workloads will allocate a bunch of memory objects, work on them
and then freeing all or most of them. So just having a percpu free pool
may not reduce the pool_lock contention significantly if large number
of objects are being used.
To help those situations, we are now doing lookahead allocation and
freeing of the debug objects into and out of the percpu free pool. This
will hopefully reduce the number of times the pool_lock needs to be
taken and hence its contention level.
Signed-off-by: Waiman Long <longman@redhat.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Yang Shi <yang.shi@linux.alibaba.com>
Cc: "Joel Fernandes (Google)" <joel@joelfernandes.org>
Cc: Qian Cai <cai@gmx.us>
Cc: Zhong Jiang <zhongjiang@huawei.com>
Link: https://lkml.kernel.org/r/20190520141450.7575-3-longman@redhat.com
When a multi-threaded workload does a lot of small memory object
allocations and deallocations, it may cause the allocation and freeing of
many debug objects. This will make the global pool_lock a bottleneck in the
performance of the workload. Since interrupts are disabled when acquiring
the pool_lock, it may even cause hard lockups to happen.
To reduce contention of the global pool_lock, add a percpu debug object
free pool that can be used to buffer some of the debug object allocation
and freeing requests without acquiring the pool_lock. Each CPU will now
have a percpu free pool that can hold up to a maximum of 64 debug
objects. Allocation and freeing requests will go to the percpu free pool
first. If that fails, the pool_lock will be taken and the global free pool
will be used.
The presence or absence of obj_cache is used as a marker to see if the
percpu cache should be used.
Signed-off-by: Waiman Long <longman@redhat.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Yang Shi <yang.shi@linux.alibaba.com>
Cc: "Joel Fernandes (Google)" <joel@joelfernandes.org>
Cc: Qian Cai <cai@gmx.us>
Cc: Zhong Jiang <zhongjiang@huawei.com>
Link: https://lkml.kernel.org/r/20190520141450.7575-2-longman@redhat.com
When calling debugfs functions, there is no need to ever check the
return value. The function can work or not, but the code logic should
never do something different based on this.
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Qian Cai <cai@gmx.us>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Waiman Long <longman@redhat.com>
Cc: "Joel Fernandes (Google)" <joel@joelfernandes.org>
Cc: Zhong Jiang <zhongjiang@huawei.com>
Link: https://lkml.kernel.org/r/20190612153513.GA21082@kroah.com
The current value of the early boot static pool size, 1024 is not big
enough for systems with large number of CPUs with timer or/and workqueue
objects selected. As the results, systems have 60+ CPUs with both timer
and workqueue objects enabled could trigger "ODEBUG: Out of memory.
ODEBUG disabled".
Some debug objects are allocated during the early boot. Enabling some
options like timers or workqueue objects may increase the size required
significantly with large number of CPUs. For example,
CONFIG_DEBUG_OBJECTS_TIMERS:
No. CPUs x 2 (worker pool) objects:
start_kernel
workqueue_init_early
init_worker_pool
init_timer_key
debug_object_init
plus No. CPUs objects (CONFIG_HIGH_RES_TIMERS):
sched_init
hrtick_rq_init
hrtimer_init
CONFIG_DEBUG_OBJECTS_WORK:
No. CPUs objects:
vmalloc_init
__init_work
plus No. CPUs x 6 (workqueue) objects:
workqueue_init_early
alloc_workqueue
__alloc_workqueue_key
alloc_and_link_pwqs
init_pwq
Also, plus No. CPUs objects:
perf_event_init
__init_srcu_struct
init_srcu_struct_fields
init_srcu_struct_nodes
__init_work
However, none of the things are actually used or required before
debug_objects_mem_init() is invoked, so just move the call right before
vmalloc_init().
According to tglx, "the reason why the call is at this place in
start_kernel() is historical. It's because back in the days when
debugobjects were added the memory allocator was enabled way later than
today."
Link: http://lkml.kernel.org/r/20181126102407.1836-1-cai@gmx.us
Signed-off-by: Qian Cai <cai@gmx.us>
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Waiman Long <longman@redhat.com>
Cc: Yang Shi <yang.shi@linux.alibaba.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
CONFIG_DEBUG_OBJECTS_RCU_HEAD does not play well with kmemleak due to
recursive calls.
fill_pool
kmemleak_ignore
make_black_object
put_object
__call_rcu (kernel/rcu/tree.c)
debug_rcu_head_queue
debug_object_activate
debug_object_init
fill_pool
kmemleak_ignore
make_black_object
...
So add SLAB_NOLEAKTRACE to kmem_cache_create() to not register newly
allocated debug objects at all.
Link: http://lkml.kernel.org/r/20181126165343.2339-1-cai@gmx.us
Signed-off-by: Qian Cai <cai@gmx.us>
Suggested-by: Catalin Marinas <catalin.marinas@arm.com>
Acked-by: Waiman Long <longman@redhat.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Yang Shi <yang.shi@linux.alibaba.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
kmem_cache_destroy() has a built in NULL pointer check, so the one at the
call can be removed.
Signed-off-by: Zhong Jiang <zhongjiang@huawei.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: <longman@redhat.com>
Cc: <arnd@arndb.de>
Cc: <yang.shi@linux.alibaba.com>
Link: https://lkml.kernel.org/r/1533054298-35824-1-git-send-email-zhongjiang@huawei.com
While debugging an issue debugobject tracking warned about an annotation
issue of an object on stack. It turned out that the issue was due to the
object in concern being on a different stack which was due to another
issue.
Thomas suggested to print the pointers and the location of the stack for
the currently running task. This helped to figure out that the object was
on the wrong stack.
As this is general useful information for debugging similar issues, make
the error message more informative by printing the pointers.
[ tglx: Massaged changelog ]
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Waiman Long <longman@redhat.com>
Acked-by: Yang Shi <yang.shi@linux.alibaba.com>
Cc: kernel-team@android.com
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: astrachan@google.com
Link: https://lkml.kernel.org/r/20180723212531.202328-1-joel@joelfernandes.org
