struct cgroup is already an RCU-safe type in the verifier. We can
therefore update bpf_cgroup_acquire() to be KF_RCU | KF_RET_NULL, and
subsequently remove bpf_cgroup_kptr_get(). This patch does the first of
these by updating bpf_cgroup_acquire() to be KF_RCU | KF_RET_NULL, and
also updates selftests accordingly.
Signed-off-by: David Vernet <void@manifault.com>
Link: https://lore.kernel.org/r/20230411041633.179404-1-void@manifault.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
It is found that attaching a task to the top_cpuset does not currently
ignore CPUs allocated to subpartitions in cpuset_attach_task(). So the
code is changed to fix that.
Signed-off-by: Waiman Long <longman@redhat.com>
Reviewed-by: Michal Koutný <mkoutny@suse.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
In the case of CLONE_INTO_CGROUP, not all cpusets are ready to accept
new tasks. It is too late to check that in cpuset_fork(). So we need
to add the cpuset_can_fork() and cpuset_cancel_fork() methods to
pre-check it before we can allow attachment to a different cpuset.
We also need to set the attach_in_progress flag to alert other code
that a new task is going to be added to the cpuset.
Fixes: ef2c41cf38 ("clone3: allow spawning processes into cgroups")
Suggested-by: Michal Koutný <mkoutny@suse.com>
Signed-off-by: Waiman Long <longman@redhat.com>
Cc: stable@vger.kernel.org # v5.7+
Signed-off-by: Tejun Heo <tj@kernel.org>
By default, the clone(2) syscall spawn a child process into the same
cgroup as its parent. With the use of the CLONE_INTO_CGROUP flag
introduced by commit ef2c41cf38 ("clone3: allow spawning processes
into cgroups"), the child will be spawned into a different cgroup which
is somewhat similar to writing the child's tid into "cgroup.threads".
The current cpuset_fork() method does not properly handle the
CLONE_INTO_CGROUP case where the cpuset of the child may be different
from that of its parent. Update the cpuset_fork() method to treat the
CLONE_INTO_CGROUP case similar to cpuset_attach().
Since the newly cloned task has not been running yet, its actual
memory usage isn't known. So it is not necessary to make change to mm
in cpuset_fork().
Fixes: ef2c41cf38 ("clone3: allow spawning processes into cgroups")
Reported-by: Giuseppe Scrivano <gscrivan@redhat.com>
Signed-off-by: Waiman Long <longman@redhat.com>
Cc: stable@vger.kernel.org # v5.7+
Signed-off-by: Tejun Heo <tj@kernel.org>
After a successful cpuset_can_attach() call which increments the
attach_in_progress flag, either cpuset_cancel_attach() or cpuset_attach()
will be called later. In cpuset_attach(), tasks in cpuset_attach_wq,
if present, will be woken up at the end. That is not the case in
cpuset_cancel_attach(). So missed wakeup is possible if the attach
operation is somehow cancelled. Fix that by doing the wakeup in
cpuset_cancel_attach() as well.
Fixes: e44193d39e ("cpuset: let hotplug propagation work wait for task attaching")
Signed-off-by: Waiman Long <longman@redhat.com>
Reviewed-by: Michal Koutný <mkoutny@suse.com>
Cc: stable@vger.kernel.org # v3.11+
Signed-off-by: Tejun Heo <tj@kernel.org>
Drop the log_size>0 and log_buf!=NULL condition when log_level>0. This
allows users to request log_true_size of a full log without providing
actual (even if small) log buffer. Verifier log handling code was mostly
ready to handle NULL log->ubuf, so only few small changes were necessary
to prevent NULL log->ubuf from causing problems.
Note, that if user provided NULL log_buf with log_level>0 we don't
consider this a log truncation, and thus won't return -ENOSPC.
We also enforce that either (log_buf==NULL && log_size==0) or
(log_buf!=NULL && log_size>0).
Suggested-by: Lorenz Bauer <lmb@isovalent.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/bpf/20230406234205.323208-15-andrii@kernel.org
Simplify internal verifier log API down to bpf_vlog_init() and
bpf_vlog_finalize(). The former handles input arguments validation in
one place and makes it easier to change it. The latter subsumes -ENOSPC
(truncation) and -EFAULT handling and simplifies both caller's code
(bpf_check() and btf_parse()).
For btf_parse(), this patch also makes sure that verifier log
finalization happens even if there is some error condition during BTF
verification process prior to normal finalization step.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/bpf/20230406234205.323208-14-andrii@kernel.org
Add output-only log_true_size and btf_log_true_size field to
BPF_PROG_LOAD and BPF_BTF_LOAD commands, respectively. It will return
the size of log buffer necessary to fit in all the log contents at
specified log_level. This is very useful for BPF loader libraries like
libbpf to be able to size log buffer correctly, but could be used by
users directly, if necessary, as well.
This patch plumbs all this through the code, taking into account actual
bpf_attr size provided by user to determine if these new fields are
expected by users. And if they are, set them from kernel on return.
We refactory btf_parse() function to accommodate this, moving attr and
uattr handling inside it. The rest is very straightforward code, which
is split from the logging accounting changes in the previous patch to
make it simpler to review logic vs UAPI changes.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/bpf/20230406234205.323208-13-andrii@kernel.org
Change how we do accounting in BPF_LOG_FIXED mode and adopt log->end_pos
as *logical* log position. This means that we can go beyond physical log
buffer size now and be able to tell what log buffer size should be to
fit entire log contents without -ENOSPC.
To do this for BPF_LOG_FIXED mode, we need to remove a short-circuiting
logic of not vsnprintf()'ing further log content once we filled up
user-provided buffer, which is done by bpf_verifier_log_needed() checks.
We modify these checks to always keep going if log->level is non-zero
(i.e., log is requested), even if log->ubuf was NULL'ed out due to
copying data to user-space, or if entire log buffer is physically full.
We adopt bpf_verifier_vlog() routine to work correctly with
log->ubuf == NULL condition, performing log formatting into temporary
kernel buffer, doing all the necessary accounting, but just avoiding
copying data out if buffer is full or NULL'ed out.
With these changes, it's now possible to do this sort of determination of
log contents size in both BPF_LOG_FIXED and default rolling log mode.
We need to keep in mind bpf_vlog_reset(), though, which shrinks log
contents after successful verification of a particular code path. This
log reset means that log->end_pos isn't always increasing, so to return
back to users what should be the log buffer size to fit all log content
without causing -ENOSPC even in the presence of log resetting, we need
to keep maximum over "lifetime" of logging. We do this accounting in
bpf_vlog_update_len_max() helper.
A related and subtle aspect is that with this logical log->end_pos even in
BPF_LOG_FIXED mode we could temporary "overflow" buffer, but then reset
it back with bpf_vlog_reset() to a position inside user-supplied
log_buf. In such situation we still want to properly maintain
terminating zero. We will eventually return -ENOSPC even if final log
buffer is small (we detect this through log->len_max check). This
behavior is simpler to reason about and is consistent with current
behavior of verifier log. Handling of this required a small addition to
bpf_vlog_reset() logic to avoid doing put_user() beyond physical log
buffer dimensions.
Another issue to keep in mind is that we limit log buffer size to 32-bit
value and keep such log length as u32, but theoretically verifier could
produce huge log stretching beyond 4GB. Instead of keeping (and later
returning) 64-bit log length, we cap it at UINT_MAX. Current UAPI makes
it impossible to specify log buffer size bigger than 4GB anyways, so we
don't really loose anything here and keep everything consistently 32-bit
in UAPI. This property will be utilized in next patch.
Doing the same determination of maximum log buffer for rolling mode is
trivial, as log->end_pos and log->start_pos are already logical
positions, so there is nothing new there.
These changes do incidentally fix one small issue with previous logging
logic. Previously, if use provided log buffer of size N, and actual log
output was exactly N-1 bytes + terminating \0, kernel logic coun't
distinguish this condition from log truncation scenario which would end
up with truncated log contents of N-1 bytes + terminating \0 as well.
But now with log->end_pos being logical position that could go beyond
actual log buffer size, we can distinguish these two conditions, which
we do in this patch. This plays nicely with returning log_size_actual
(implemented in UAPI in the next patch), as we can now guarantee that if
user takes such log_size_actual and provides log buffer of that exact
size, they will not get -ENOSPC in return.
All in all, all these changes do conceptually unify fixed and rolling
log modes much better, and allow a nice feature requested by users:
knowing what should be the size of the buffer to avoid -ENOSPC.
We'll plumb this through the UAPI and the code in the next patch.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/bpf/20230406234205.323208-12-andrii@kernel.org
Currently, if user-supplied log buffer to collect BPF verifier log turns
out to be too small to contain full log, bpf() syscall returns -ENOSPC,
fails BPF program verification/load, and preserves first N-1 bytes of
the verifier log (where N is the size of user-supplied buffer).
This is problematic in a bunch of common scenarios, especially when
working with real-world BPF programs that tend to be pretty complex as
far as verification goes and require big log buffers. Typically, it's
when debugging tricky cases at log level 2 (verbose). Also, when BPF program
is successfully validated, log level 2 is the only way to actually see
verifier state progression and all the important details.
Even with log level 1, it's possible to get -ENOSPC even if the final
verifier log fits in log buffer, if there is a code path that's deep
enough to fill up entire log, even if normally it would be reset later
on (there is a logic to chop off successfully validated portions of BPF
verifier log).
In short, it's not always possible to pre-size log buffer. Also, what's
worse, in practice, the end of the log most often is way more important
than the beginning, but verifier stops emitting log as soon as initial
log buffer is filled up.
This patch switches BPF verifier log behavior to effectively behave as
rotating log. That is, if user-supplied log buffer turns out to be too
short, verifier will keep overwriting previously written log,
effectively treating user's log buffer as a ring buffer. -ENOSPC is
still going to be returned at the end, to notify user that log contents
was truncated, but the important last N bytes of the log would be
returned, which might be all that user really needs. This consistent
-ENOSPC behavior, regardless of rotating or fixed log behavior, allows
to prevent backwards compatibility breakage. The only user-visible
change is which portion of verifier log user ends up seeing *if buffer
is too small*. Given contents of verifier log itself is not an ABI,
there is no breakage due to this behavior change. Specialized tools that
rely on specific contents of verifier log in -ENOSPC scenario are
expected to be easily adapted to accommodate old and new behaviors.
Importantly, though, to preserve good user experience and not require
every user-space application to adopt to this new behavior, before
exiting to user-space verifier will rotate log (in place) to make it
start at the very beginning of user buffer as a continuous
zero-terminated string. The contents will be a chopped off N-1 last
bytes of full verifier log, of course.
Given beginning of log is sometimes important as well, we add
BPF_LOG_FIXED (which equals 8) flag to force old behavior, which allows
tools like veristat to request first part of verifier log, if necessary.
BPF_LOG_FIXED flag is also a simple and straightforward way to check if
BPF verifier supports rotating behavior.
On the implementation side, conceptually, it's all simple. We maintain
64-bit logical start and end positions. If we need to truncate the log,
start position will be adjusted accordingly to lag end position by
N bytes. We then use those logical positions to calculate their matching
actual positions in user buffer and handle wrap around the end of the
buffer properly. Finally, right before returning from bpf_check(), we
rotate user log buffer contents in-place as necessary, to make log
contents contiguous. See comments in relevant functions for details.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/bpf/20230406234205.323208-4-andrii@kernel.org
kernel/bpf/verifier.c file is large and growing larger all the time. So
it's good to start splitting off more or less self-contained parts into
separate files to keep source code size (somewhat) somewhat under
control.
This patch is a one step in this direction, moving some of BPF verifier log
routines into a separate kernel/bpf/log.c. Right now it's most low-level
and isolated routines to append data to log, reset log to previous
position, etc. Eventually we could probably move verifier state
printing logic here as well, but this patch doesn't attempt to do that
yet.
Subsequent patches will add more logic to verifier log management, so
having basics in a separate file will make sure verifier.c doesn't grow
more with new changes.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Lorenz Bauer <lmb@isovalent.com>
Link: https://lore.kernel.org/bpf/20230406234205.323208-2-andrii@kernel.org
Pull RCU fix from Paul McKenney:
"This fixes a pair of bugs in which an improbable but very real
sequence of events can cause kfree_rcu() to be a bit too quick about
freeing the memory passed to it.
It turns out that this pair of bugs is about two years old, and so
this is not a v6.3 regression. However: (1) It just started showing up
in the wild and (2) Its consequences are dire, so its fix needs to go
in sooner rather than later.
Testing is of course being upgraded, and the upgraded tests detect
this situation very quickly. But to the best of my knowledge right
now, the tests are not particularly urgent and will thus most likely
show up in the v6.5 merge window (the one after this coming one).
Kudos to Ziwei Dai and his group for tracking this one down the hard
way!"
* tag 'urgent-rcu.2023.04.07a' of git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu:
rcu/kvfree: Avoid freeing new kfree_rcu() memory after old grace period
Pull perf fixes from Borislav Petkov:
- Fix "same task" check when redirecting event output
- Do not wait unconditionally for RCU on the event migration path if
there are no events to migrate
* tag 'perf_urgent_for_v6.3_rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
perf/core: Fix the same task check in perf_event_set_output
perf: Optimize perf_pmu_migrate_context()
Pull dma-mapping fix from Christoph Hellwig:
- fix a braino in the swiotlb alignment check fix (Petr Tesarik)
* tag 'dma-mapping-6.3-2023-04-08' of git://git.infradead.org/users/hch/dma-mapping:
swiotlb: fix a braino in the alignment check fix
Pull tracing fixes from Steven Rostedt:
"A couple more minor fixes:
- Reset direct->addr back to its original value on error in updating
the direct trampoline code
- Make lastcmd_mutex static"
* tag 'trace-v6.3-rc5-2' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace:
tracing/synthetic: Make lastcmd_mutex static
ftrace: Fix issue that 'direct->addr' not restored in modify_ftrace_direct()
Pull MM fixes from Andrew Morton:
"28 hotfixes.
23 are cc:stable and the other five address issues which were
introduced during this merge cycle.
20 are for MM and the remainder are for other subsystems"
* tag 'mm-hotfixes-stable-2023-04-07-16-23' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm: (28 commits)
maple_tree: fix a potential concurrency bug in RCU mode
maple_tree: fix get wrong data_end in mtree_lookup_walk()
mm/swap: fix swap_info_struct race between swapoff and get_swap_pages()
nilfs2: fix sysfs interface lifetime
mm: take a page reference when removing device exclusive entries
mm: vmalloc: avoid warn_alloc noise caused by fatal signal
nilfs2: initialize "struct nilfs_binfo_dat"->bi_pad field
nilfs2: fix potential UAF of struct nilfs_sc_info in nilfs_segctor_thread()
zsmalloc: document freeable stats
zsmalloc: document new fullness grouping
fsdax: force clear dirty mark if CoW
mm/hugetlb: fix uffd wr-protection for CoW optimization path
mm: enable maple tree RCU mode by default
maple_tree: add RCU lock checking to rcu callback functions
maple_tree: add smp_rmb() to dead node detection
maple_tree: fix write memory barrier of nodes once dead for RCU mode
maple_tree: remove extra smp_wmb() from mas_dead_leaves()
maple_tree: fix freeing of nodes in rcu mode
maple_tree: detect dead nodes in mas_start()
maple_tree: be more cautious about dead nodes
...
Provide a kconfig option to allow arches to manipulate default
value of dma_default_coherent in Kconfig.
Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
dma_default_coherent was decleared unconditionally at kernel/dma/mapping.c
but only decleared when any of non-coherent options is enabled in
dma-map-ops.h.
Guard the declaration in mapping.c with non-coherent options and provide
a fallback definition.
Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Currently, the verifier does not handle '<const> <cond_op> <non_const>' well.
For example,
...
10: (79) r1 = *(u64 *)(r10 -16) ; R1_w=scalar() R10=fp0
11: (b7) r2 = 0 ; R2_w=0
12: (2d) if r2 > r1 goto pc+2
13: (b7) r0 = 0
14: (95) exit
15: (65) if r1 s> 0x1 goto pc+3
16: (0f) r0 += r1
...
At insn 12, verifier decides both true and false branch are possible, but
actually only false branch is possible.
Currently, the verifier already supports patterns '<non_const> <cond_op> <const>.
Add support for patterns '<const> <cond_op> <non_const>' in a similar way.
Also fix selftest 'verifier_bounds_mix_sign_unsign/bounds checks mixing signed and unsigned, variant 10'
due to this change.
Signed-off-by: Yonghong Song <yhs@fb.com>
Acked-by: Dave Marchevsky <davemarchevsky@fb.com>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20230406164505.1046801-1-yhs@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Currently, for BPF_JEQ/BPF_JNE insn, verifier determines
whether the branch is taken or not only if both operands
are constants. Therefore, for the following code snippet,
0: (85) call bpf_ktime_get_ns#5 ; R0_w=scalar()
1: (a5) if r0 < 0x3 goto pc+2 ; R0_w=scalar(umin=3)
2: (b7) r2 = 2 ; R2_w=2
3: (1d) if r0 == r2 goto pc+2 6
At insn 3, since r0 is not a constant, verifier assumes both branch
can be taken which may lead inproper verification failure.
Add comparing umin/umax value and the constant. If the umin value
is greater than the constant, or umax value is smaller than the constant,
for JEQ the branch must be not-taken, and for JNE the branch must be taken.
The jmp32 mode JEQ/JNE branch taken checking is also handled similarly.
The following lists the veristat result w.r.t. changed number
of processes insns during verification:
File Program Insns (A) Insns (B) Insns (DIFF)
----------------------------------------------------- ---------------------------------------------------- --------- --------- ---------------
test_cls_redirect.bpf.linked3.o cls_redirect 64980 73472 +8492 (+13.07%)
test_seg6_loop.bpf.linked3.o __add_egr_x 12425 12423 -2 (-0.02%)
test_tcp_hdr_options.bpf.linked3.o estab 2634 2558 -76 (-2.89%)
test_parse_tcp_hdr_opt.bpf.linked3.o xdp_ingress_v6 1421 1420 -1 (-0.07%)
test_parse_tcp_hdr_opt_dynptr.bpf.linked3.o xdp_ingress_v6 1238 1237 -1 (-0.08%)
test_tc_dtime.bpf.linked3.o egress_fwdns_prio100 414 411 -3 (-0.72%)
Mostly a small improvement but test_cls_redirect.bpf.linked3.o has a 13% regression.
I checked with verifier log and found it this is due to pruning.
For some JEQ/JNE branches impacted by this patch,
one branch is explored and the other has state equivalence and
pruned.
Signed-off-by: Yonghong Song <yhs@fb.com>
Acked-by: Dave Marchevsky <davemarchevsky@fb.com>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20230406164455.1045294-1-yhs@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Memory passed to kvfree_rcu() that is to be freed is tracked by a
per-CPU kfree_rcu_cpu structure, which in turn contains pointers
to kvfree_rcu_bulk_data structures that contain pointers to memory
that has not yet been handed to RCU, along with an kfree_rcu_cpu_work
structure that tracks the memory that has already been handed to RCU.
These structures track three categories of memory: (1) Memory for
kfree(), (2) Memory for kvfree(), and (3) Memory for both that arrived
during an OOM episode. The first two categories are tracked in a
cache-friendly manner involving a dynamically allocated page of pointers
(the aforementioned kvfree_rcu_bulk_data structures), while the third
uses a simple (but decidedly cache-unfriendly) linked list through the
rcu_head structures in each block of memory.
On a given CPU, these three categories are handled as a unit, with that
CPU's kfree_rcu_cpu_work structure having one pointer for each of the
three categories. Clearly, new memory for a given category cannot be
placed in the corresponding kfree_rcu_cpu_work structure until any old
memory has had its grace period elapse and thus has been removed. And
the kfree_rcu_monitor() function does in fact check for this.
Except that the kfree_rcu_monitor() function checks these pointers one
at a time. This means that if the previous kfree_rcu() memory passed
to RCU had only category 1 and the current one has only category 2, the
kfree_rcu_monitor() function will send that current category-2 memory
along immediately. This can result in memory being freed too soon,
that is, out from under unsuspecting RCU readers.
To see this, consider the following sequence of events, in which:
o Task A on CPU 0 calls rcu_read_lock(), then uses "from_cset",
then is preempted.
o CPU 1 calls kfree_rcu(cset, rcu_head) in order to free "from_cset"
after a later grace period. Except that "from_cset" is freed
right after the previous grace period ended, so that "from_cset"
is immediately freed. Task A resumes and references "from_cset"'s
member, after which nothing good happens.
In full detail:
CPU 0 CPU 1
---------------------- ----------------------
count_memcg_event_mm()
|rcu_read_lock() <---
|mem_cgroup_from_task()
|// css_set_ptr is the "from_cset" mentioned on CPU 1
|css_set_ptr = rcu_dereference((task)->cgroups)
|// Hard irq comes, current task is scheduled out.
cgroup_attach_task()
|cgroup_migrate()
|cgroup_migrate_execute()
|css_set_move_task(task, from_cset, to_cset, true)
|cgroup_move_task(task, to_cset)
|rcu_assign_pointer(.., to_cset)
|...
|cgroup_migrate_finish()
|put_css_set_locked(from_cset)
|from_cset->refcount return 0
|kfree_rcu(cset, rcu_head) // free from_cset after new gp
|add_ptr_to_bulk_krc_lock()
|schedule_delayed_work(&krcp->monitor_work, ..)
kfree_rcu_monitor()
|krcp->bulk_head[0]'s work attached to krwp->bulk_head_free[]
|queue_rcu_work(system_wq, &krwp->rcu_work)
|if rwork->rcu.work is not in WORK_STRUCT_PENDING_BIT state,
|call_rcu(&rwork->rcu, rcu_work_rcufn) <--- request new gp
// There is a perious call_rcu(.., rcu_work_rcufn)
// gp end, rcu_work_rcufn() is called.
rcu_work_rcufn()
|__queue_work(.., rwork->wq, &rwork->work);
|kfree_rcu_work()
|krwp->bulk_head_free[0] bulk is freed before new gp end!!!
|The "from_cset" is freed before new gp end.
// the task resumes some time later.
|css_set_ptr->subsys[(subsys_id) <--- Caused kernel crash, because css_set_ptr is freed.
This commit therefore causes kfree_rcu_monitor() to refrain from moving
kfree_rcu() memory to the kfree_rcu_cpu_work structure until the RCU
grace period has completed for all three categories.
v2: Use helper function instead of inserted code block at kfree_rcu_monitor().
Fixes: 34c8817455 ("rcu: Support kfree_bulk() interface in kfree_rcu()")
Fixes: 5f3c8d6204 ("rcu/tree: Maintain separate array for vmalloc ptrs")
Reported-by: Mukesh Ojha <quic_mojha@quicinc.com>
Signed-off-by: Ziwei Dai <ziwei.dai@unisoc.com>
Reviewed-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Tested-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
During Numa scanning make sure only relevant vmas of the tasks are
scanned.
Before:
All the tasks of a process participate in scanning the vma even if they
do not access vma in it's lifespan.
Now:
Except cases of first few unconditional scans, if a process do
not touch vma (exluding false positive cases of PID collisions)
tasks no longer scan all vma
Logic used:
1) 6 bits of PID used to mark active bit in vma numab status during
fault to remember PIDs accessing vma. (Thanks Mel)
2) Subsequently in scan path, vma scanning is skipped if current PID
had not accessed vma.
3) First two times we do allow unconditional scan to preserve earlier
behaviour of scanning.
Acknowledgement to Bharata B Rao <bharata@amd.com> for initial patch to
store pid information and Peter Zijlstra <peterz@infradead.org> (Usage of
test and set bit)
Link: https://lkml.kernel.org/r/092f03105c7c1d3450f4636b1ea350407f07640e.1677672277.git.raghavendra.kt@amd.com
Signed-off-by: Raghavendra K T <raghavendra.kt@amd.com>
Suggested-by: Mel Gorman <mgorman@techsingularity.net>
Cc: David Hildenbrand <david@redhat.com>
Cc: Disha Talreja <dishaa.talreja@amd.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Mike Rapoport <rppt@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Pach series "sched/numa: Enhance vma scanning", v3.
The patchset proposes one of the enhancements to numa vma scanning
suggested by Mel. This is continuation of [3].
Reposting the rebased patchset to akpm mm-unstable tree (March 1)
Existing mechanism of scan period involves, scan period derived from
per-thread stats. Process Adaptive autoNUMA [1] proposed to gather NUMA
fault stats at per-process level to capture aplication behaviour better.
During that course of discussion, Mel proposed several ideas to enhance
current numa balancing. One of the suggestion was below
Track what threads access a VMA. The suggestion was to use an unsigned
long pid_mask and use the lower bits to tag approximately what threads
access a VMA. Skip VMAs that did not trap a fault. This would be
approximate because of PID collisions but would reduce scanning of areas
the thread is not interested in. The above suggestion intends not to
penalize threads that has no interest in the vma, thus reduce scanning
overhead.
V3 changes are mostly based on PeterZ comments (details below in changes)
Summary of patchset:
Current patchset implements:
1. Delay the vma scanning logic for newly created VMA's so that
additional overhead of scanning is not incurred for short lived tasks
(implementation by Mel)
2. Store the information of tasks accessing VMA in 2 windows. It is
regularly cleared in (4*sysctl_numa_balancing_scan_delay) interval.
The above time is derived from experimenting (Suggested by PeterZ) to
balance between frequent clearing vs obsolete access data
3. hash_32 used to encode task index accessing VMA information
4. VMA's acess information is used to skip scanning for the tasks
which had not accessed VMA
Changes since V2:
patch1:
- Renaming of structure, macro to function,
- Add explanation to heuristics
- Adding more details from result (PeterZ)
Patch2:
- Usage of test and set bit (PeterZ)
- Move storing access PID info to numa_migrate_prep()
- Add a note on fainess among tasks allowed to scan
(PeterZ)
Patch3:
- Maintain two windows of access PID information
(PeterZ supported implementation and Gave idea to extend
to N if needed)
Patch4:
- Apply hash_32 function to track VMA accessing PIDs (PeterZ)
Changes since RFC V1:
- Include Mel's vma scan delay patch
- Change the accessing pid store logic (Thanks Mel)
- Fencing structure / code to NUMA_BALANCING (David, Mel)
- Adding clearing access PID logic (Mel)
- Descriptive change log ( Mike Rapoport)
Things to ponder over:
==========================================
- Improvement to clearing accessing PIDs logic (discussed in-detail in
patch3 itself (Done in this patchset by implementing 2 window history)
- Current scan period is not changed in the patchset, so we do see
frequent tries to scan. Relaxing scan period dynamically could improve
results further.
[1] sched/numa: Process Adaptive autoNUMA
Link: https://lore.kernel.org/lkml/20220128052851.17162-1-bharata@amd.com/T/
[2] RFC V1 Link:
https://lore.kernel.org/all/cover.1673610485.git.raghavendra.kt@amd.com/
[3] V2 Link:
https://lore.kernel.org/lkml/cover.1675159422.git.raghavendra.kt@amd.com/
Results:
Summary: Huge autonuma cost reduction seen in mmtest. Kernbench improvement
is more than 5% and huge system time (80%+) improvement from mmtest autonuma.
(dbench had huge std deviation to post)
kernbench
===========
6.2.0-mmunstable-base 6.2.0-mmunstable-patched
Amean user-256 22002.51 ( 0.00%) 22649.95 * -2.94%*
Amean syst-256 10162.78 ( 0.00%) 8214.13 * 19.17%*
Amean elsp-256 160.74 ( 0.00%) 156.92 * 2.38%*
Duration User 66017.43 67959.84
Duration System 30503.15 24657.03
Duration Elapsed 504.61 493.12
6.2.0-mmunstable-base 6.2.0-mmunstable-patched
Ops NUMA alloc hit 1738835089.00 1738780310.00
Ops NUMA alloc local 1738834448.00 1738779711.00
Ops NUMA base-page range updates 477310.00 392566.00
Ops NUMA PTE updates 477310.00 392566.00
Ops NUMA hint faults 96817.00 87555.00
Ops NUMA hint local faults % 10150.00 2192.00
Ops NUMA hint local percent 10.48 2.50
Ops NUMA pages migrated 86660.00 85363.00
Ops AutoNUMA cost 489.07 442.14
autonumabench
===============
6.2.0-mmunstable-base 6.2.0-mmunstable-patched
Amean syst-NUMA01 399.50 ( 0.00%) 52.05 * 86.97%*
Amean syst-NUMA01_THREADLOCAL 0.21 ( 0.00%) 0.22 * -5.41%*
Amean syst-NUMA02 0.80 ( 0.00%) 0.78 * 2.68%*
Amean syst-NUMA02_SMT 0.65 ( 0.00%) 0.68 * -3.95%*
Amean elsp-NUMA01 313.26 ( 0.00%) 313.11 * 0.05%*
Amean elsp-NUMA01_THREADLOCAL 1.06 ( 0.00%) 1.08 * -1.76%*
Amean elsp-NUMA02 3.19 ( 0.00%) 3.24 * -1.52%*
Amean elsp-NUMA02_SMT 3.72 ( 0.00%) 3.61 * 2.92%*
Duration User 396433.47 324835.96
Duration System 2808.70 376.66
Duration Elapsed 2258.61 2258.12
6.2.0-mmunstable-base 6.2.0-mmunstable-patched
Ops NUMA alloc hit 59921806.00 49623489.00
Ops NUMA alloc miss 0.00 0.00
Ops NUMA interleave hit 0.00 0.00
Ops NUMA alloc local 59920880.00 49622594.00
Ops NUMA base-page range updates 152259275.00 50075.00
Ops NUMA PTE updates 152259275.00 50075.00
Ops NUMA PMD updates 0.00 0.00
Ops NUMA hint faults 154660352.00 39014.00
Ops NUMA hint local faults % 138550501.00 23139.00
Ops NUMA hint local percent 89.58 59.31
Ops NUMA pages migrated 8179067.00 14147.00
Ops AutoNUMA cost 774522.98 195.69
This patch (of 4):
Currently whenever a new task is created we wait for
sysctl_numa_balancing_scan_delay to avoid unnessary scanning overhead.
Extend the same logic to new or very short-lived VMAs.
[raghavendra.kt@amd.com: add initialization in vm_area_dup())]
Link: https://lkml.kernel.org/r/cover.1677672277.git.raghavendra.kt@amd.com
Link: https://lkml.kernel.org/r/7a6fbba87c8b51e67efd3e74285bb4cb311a16ca.1677672277.git.raghavendra.kt@amd.com
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Raghavendra K T <raghavendra.kt@amd.com>
Cc: Bharata B Rao <bharata@amd.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Disha Talreja <dishaa.talreja@amd.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
vma->lock being part of the vm_area_struct causes performance regression
during page faults because during contention its count and owner fields
are constantly updated and having other parts of vm_area_struct used
during page fault handling next to them causes constant cache line
bouncing. Fix that by moving the lock outside of the vm_area_struct.
All attempts to keep vma->lock inside vm_area_struct in a separate cache
line still produce performance regression especially on NUMA machines.
Smallest regression was achieved when lock is placed in the fourth cache
line but that bloats vm_area_struct to 256 bytes.
Considering performance and memory impact, separate lock looks like the
best option. It increases memory footprint of each VMA but that can be
optimized later if the new size causes issues. Note that after this
change vma_init() does not allocate or initialize vma->lock anymore. A
number of drivers allocate a pseudo VMA on the stack but they never use
the VMA's lock, therefore it does not need to be allocated. The future
drivers which might need the VMA lock should use
vm_area_alloc()/vm_area_free() to allocate the VMA.
Link: https://lkml.kernel.org/r/20230227173632.3292573-34-surenb@google.com
Signed-off-by: Suren Baghdasaryan <surenb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
call_rcu() can take a long time when callback offloading is enabled. Its
use in the vm_area_free can cause regressions in the exit path when
multiple VMAs are being freed.
Because exit_mmap() is called only after the last mm user drops its
refcount, the page fault handlers can't be racing with it. Any other
possible user like oom-reaper or process_mrelease are already synchronized
using mmap_lock. Therefore exit_mmap() can free VMAs directly, without
the use of call_rcu().
Expose __vm_area_free() and use it from exit_mmap() to avoid possible
call_rcu() floods and performance regressions caused by it.
Link: https://lkml.kernel.org/r/20230227173632.3292573-33-surenb@google.com
Signed-off-by: Suren Baghdasaryan <surenb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Introduce per-VMA locking. The lock implementation relies on a per-vma
and per-mm sequence counters to note exclusive locking:
- read lock - (implemented by vma_start_read) requires the vma
(vm_lock_seq) and mm (mm_lock_seq) sequence counters to differ.
If they match then there must be a vma exclusive lock held somewhere.
- read unlock - (implemented by vma_end_read) is a trivial vma->lock
unlock.
- write lock - (vma_start_write) requires the mmap_lock to be held
exclusively and the current mm counter is assigned to the vma counter.
This will allow multiple vmas to be locked under a single mmap_lock
write lock (e.g. during vma merging). The vma counter is modified
under exclusive vma lock.
- write unlock - (vma_end_write_all) is a batch release of all vma
locks held. It doesn't pair with a specific vma_start_write! It is
done before exclusive mmap_lock is released by incrementing mm
sequence counter (mm_lock_seq).
- write downgrade - if the mmap_lock is downgraded to the read lock, all
vma write locks are released as well (effectivelly same as write
unlock).
Link: https://lkml.kernel.org/r/20230227173632.3292573-13-surenb@google.com
Signed-off-by: Suren Baghdasaryan <surenb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
