| CVE |
Vendors |
Products |
Updated |
CVSS v3.1 |
| In the Linux kernel, the following vulnerability has been resolved:
i3c: master: svc: Prevent IRQ storm from false SLVSTART on NPCM845
On NPCM845, when a target on the I3C bus gets stuck holding SDA low,
the controller reports a false Master Request (MR) in-band interrupt
event. The driver handles this by emitting a STOP condition to restore
the bus.
However, the hardware quirk SVC_I3C_QUIRK_FALSE_SLVSTART indicates that
emitting a STOP condition may spuriously set the SLVSTART interrupt
status bit. In the Master Request case, this creates a feedback loop:
the STOP triggers a new SLVSTART event, the IRQ handler fires again,
the controller still reports an MR type, another STOP is emitted, and
the cycle repeats indefinitely, resulting in an IRQ storm that can lock
up the CPU.
Clear the SLVSTART status bit explicitly after emitting the STOP in the
Master Request IBI handler when the SVC_I3C_QUIRK_FALSE_SLVSTART quirk
is set. This breaks the feedback loop without affecting normal SLVSTART
processing, which is already guarded in the top-level IRQ handler by
checking that MSTATUS is in SLVREQ state. |
| In the Linux kernel, the following vulnerability has been resolved:
serial: 8250_of: clear stuck empty-FIFO RX-timeout on LPC32xx
The NXP LPC32xx UART (PORT_LPC3220) can latch an RX character-timeout
interrupt while the RX FIFO is empty: IIR reports UART_IIR_RX_TIMEOUT
(0x0c) but LSR.DR is clear. A character timeout is only cleared by
reading RHR, but serial8250_rx_chars() reads RHR only when LSR.DR is
set, so nothing ever clears the condition. The interrupt is
level-triggered and re-fires immediately, so on a single-core ARM926
the resulting interrupt storm livelocks the CPU.
It is reproducible when userspace repeatedly opens the front-panel port
(ttyS1): serial8250_do_set_termios() re-enables interrupts on unlock and
the handler then spins forever with iir=0xcc lsr=0x60 ier=0x05, tripping
the soft-lockup detector in serial8250_handle_irq_locked().
LPC32xx has no dedicated 8250 glue driver, it's driven by the generic
8250_of. Add a hardware specific handle_irq for PORT_LPC3220, wired up
in of_platform_serial_setup() the same way fsl8250_handle_irq is
installed. The handler follows dw8250_handle_irq(): on an RX timeout
with an empty FIFO (LSR.DR and LSR.BI clear) it does one throwaway RHR
read to clear the condition, then calls serial8250_handle_irq_locked().
No real received data is ever discarded, and it is a no-op on healthy
UARTs which never report a timeout with DR clear.
This is the same class of bug already worked around in other 8250 drivers;
see commit 424d79183af0 ("serial: 8250_dw: Avoid "too much work" from bogus rx timeout interrupt")
which reports the identical iir=0xcc/lsr=0x60. See also
UART_RX_TIMEOUT_QUIRK in 8250_omap, and the note in 8250_bcm7271. |
| In the Linux kernel, the following vulnerability has been resolved:
md/raid1,raid10: fix deadlock in read error recovery path
raid1d and raid10d may resubmit a split md cloned bio while handling
a read error. In this case, resubmitting the bio can lead to a deadlock
if the array is suspended before md_handle_request() acquires an
active_io reference via percpu_ref_tryget_live().
Since the cloned bio already holds an active_io reference,
trying to acquire another reference via percpu_ref_tryget_live()
can lead to a deadlock while the array is suspended.
Fix this by using percpu_ref_get() for md cloned bios. |
| In the Linux kernel, the following vulnerability has been resolved:
NFS/localio: prevent direct reclaim recursion into NFS via nfs_writepages
LOCALIO is an NFS loopback mount optimization that avoids using the
network for READ, WRITE and COMMIT if the NFS client and server are
determined to be on the same system. But because LOCALIO is still
fundamentally "just NFS loopback mount" it is susceptible to recursion
deadlock via direct reclaim, e.g.: NFS LOCALIO down to XFS and then
back into NFS via nfs_writepages.
Fix LOCALIO's potential for direct reclaim deadlock by ensuring that
all its page cache allocations are done from GFP_NOFS context.
Thanks to Ben Coddington for pointing out commit ad22c7a043c2 ("xfs:
prevent stack overflows from page cache allocation"). |
| In the Linux kernel, the following vulnerability has been resolved:
mm/damon/core: fix damos_walk() vs kdamond_fn() exit race
When kdamond_fn() main loop is finished, the function cancels remaining
damos_walk() request and unset the damon_ctx->kdamond so that API callers
and API functions themselves can show the context is terminated.
damos_walk() adds the caller's request to the queue first. After that, it
shows if the kdamond of the damon_ctx is still running (damon_ctx->kdamond
is set). Only if the kdamond is running, damos_walk() starts waiting for
the kdamond's handling of the newly added request.
The damos_walk() requests registration and damon_ctx->kdamond unset are
protected by different mutexes, though. Hence, damos_walk() could race
with damon_ctx->kdamond unset, and result in deadlocks.
For example, let's suppose kdamond successfully finished the damow_walk()
request cancelling. Right after that, damos_walk() is called for the
context. It registers the new request, and shows the context is still
running, because damon_ctx->kdamond unset is not yet done. Hence the
damos_walk() caller starts waiting for the handling of the request.
However, the kdamond is already on the termination steps, so it never
handles the new request. As a result, the damos_walk() caller thread
infinitely waits.
Fix this by introducing another damon_ctx field, namely
walk_control_obsolete. It is protected by the
damon_ctx->walk_control_lock, which protects damos_walk() request
registration. Initialize (unset) it in kdamond_fn() before letting
damon_start() returns and set it just before the cancelling of the
remaining damos_walk() request is executed. damos_walk() reads the
obsolete field under the lock and avoids adding a new request.
After this change, only requests that are guaranteed to be handled or
cancelled are registered. Hence the after-registration DAMON context
termination check is no longer needed. Remove it together.
The issue is found by sashiko [1]. |
| In the Linux kernel, the following vulnerability has been resolved:
jbd2: bound shrinker scans by examined checkpoint buffers
The jbd2 shrinker currently accounts only checkpoint buffers that it
successfully releases against nr_to_scan. Busy buffers therefore do not
consume the scan budget.
If a checkpoint transaction contains mostly busy buffers, the shrinker
can scan its entire checkpoint list while holding journal->j_list_lock.
Large checkpoint lists can result in excessive lock hold times and leave
other CPUs spinning on j_list_lock, causing soft lockups or RCU stalls.
Pass nr_to_scan into journal_shrink_one_cp_list() and decrement it for
every buffer examined, including busy buffers. Pass NULL from checkpoint
cleanup paths so their existing full-list behavior is preserved.
This restores the scan-budget semantics that existed before
journal_shrink_one_cp_list() was changed to always scan a complete
checkpoint list. |
| In the Linux kernel, the following vulnerability has been resolved:
ntfs3: fix circular locking dependency in run_unpack_ex
Syzbot reported a circular locking dependency between wnd->rw_lock
(sbi->used.bitmap) and ni->file.run_lock.
The deadlock scenario:
1. ntfs_extend_mft() takes ni->file.run_lock then wnd->rw_lock.
2. run_unpack_ex() takes wnd->rw_lock then tries to acquire
ni->file.run_lock inside ntfs_refresh_zone().
This creates an AB-BA deadlock.
Fix this by using down_read_trylock() instead of down_read() when
acquiring run_lock in run_unpack_ex(). If the lock is contended,
skip ntfs_refresh_zone() - the MFT zone will be refreshed on the
next MFT operation. This breaks the circular dependency since we
never block waiting for run_lock while holding wnd->rw_lock. |
| tinytag is a Python library for reading audio file metadata. From 2.0.0 until 2.3.0, an M4A file containing a sub-atom with a size of zero inside a ---- atom can cause the metadata parser to stop advancing through the input. The custom-field atom traversal repeatedly processes the malformed location, causing an infinite loop and denying service to a program that reads the crafted file. This issue is fixed in version 2.3.0. |
| A missing circuit-breaker in the replicator in Apache CouchDB 3.5.2 and below on all platforms allows the owner of a replication source or target to cause a denial of service via creating an infinite HTTP redirect loop.
Users are recommended to upgrade to version 3.5.3, which fixes this issue. |
| In the Linux kernel, the following vulnerability has been resolved:
net: atl1: fix soft lockup on out-of-range cmb_tpd_next_to_clean read
Same issue as atl1c (see the first commit in this series, "net:
atl1c: fix soft lockup on out-of-range tpd_cons read"): the hardware
can report an out-of-range cmb_tpd_next_to_clean (seen as 0xffff)
while the PCIe link/MAC is resetting. An out-of-range value can
never be reached and the loop below would spin forever. Treat it as
"nothing new to clean" instead. |
| OPNsense is a FreeBSD based firewall and routing platform. Prior to version 26.1.11, Firewall Rules and NAT views place stored category names and alias descriptions into HTML title attributes in filter_rule.volt and nat_rule.volt without quote-safe encoding. The response layer uses ENT_NOQUOTES, so a double quote survives and closes the attribute, allowing an event-handler attribute to be added to the existing span. A user with Firewall Categories or Firewall Alias Edit privileges can store a category name or alias description; the category must be assigned to a filter or NAT rule, or the alias must be referenced by a rule source or destination, and the rule must be saved so the affected tooltip is rendered. Script then executes in the session of an administrator or other user who opens the affected Rules or NAT page. The script can access CSRF material and perform authenticated actions as the victim. This issue is fixed in version 26.1.11. |
| FreeRDP from 3.0.0 before 3.32.0 contains an infinite loop vulnerability in http_response_recv() that allows malicious gateways to exhaust client CPU via crafted Content-Type headers. Attackers controlling a gateway can return a nonzero Content-Length with a semicolon-containing unrecognized Content-Type, pinning a client thread at full CPU indefinitely. |
| quirc through 1.2 contains an excessive-iteration vulnerability in find_alignment_pattern() in lib/identify.c that allows attackers to hang quirc_end() with a crafted image. Attackers can arrange three finder patterns producing a near-degenerate perspective transform, causing billions of region_code() iterations that exhaust CPU in applications decoding attacker-supplied QR images. |
| In the Linux kernel, the following vulnerability has been resolved:
nvmet: avoid recursive nvmet-wq flush in nvmet_ctrl_free
nvmet_tcp_release_queue_work() runs on nvmet-wq and can drop the
final controller reference through nvmet_cq_put(). If that triggers
nvmet_ctrl_free(), the teardown path flushes ctrl->async_event_work on
the same nvmet-wq.
Call chain:
nvmet_tcp_schedule_release_queue()
kref_put(&queue->kref, nvmet_tcp_release_queue)
nvmet_tcp_release_queue()
queue_work(nvmet_wq, &queue->release_work) <--- nvmet_wq
process_one_work()
nvmet_tcp_release_queue_work()
nvmet_cq_put(&queue->nvme_cq)
nvmet_cq_destroy()
nvmet_ctrl_put(cq->ctrl)
nvmet_ctrl_free()
flush_work(&ctrl->async_event_work) <--- nvmet_wq
Previously Scheduled by :-
nvmet_add_async_event
queue_work(nvmet_wq, &ctrl->async_event_work);
This trips lockdep with a possible recursive locking warning.
[ 5223.015876] run blktests nvme/003 at 2026-04-07 20:53:55
[ 5223.061801] loop0: detected capacity change from 0 to 2097152
[ 5223.072206] nvmet: adding nsid 1 to subsystem blktests-subsystem-1
[ 5223.088368] nvmet_tcp: enabling port 0 (127.0.0.1:4420)
[ 5223.126086] nvmet: Created discovery controller 1 for subsystem nqn.2014-08.org.nvmexpress.discovery for NQN nqn.2014-08.org.nvmexpress:uuid:0f01fb42-9f7f-4856-b0b3-51e60b8de349.
[ 5223.128453] nvme nvme1: new ctrl: NQN "nqn.2014-08.org.nvmexpress.discovery", addr 127.0.0.1:4420, hostnqn: nqn.2014-08.org.nvmexpress:uuid:0f01fb42-9f7f-4856-b0b3-51e60b8de349
[ 5233.199447] nvme nvme1: Removing ctrl: NQN "nqn.2014-08.org.nvmexpress.discovery"
[ 5233.227718] ============================================
[ 5233.231283] WARNING: possible recursive locking detected
[ 5233.234696] 7.0.0-rc3nvme+ #20 Tainted: G O N
[ 5233.238434] --------------------------------------------
[ 5233.241852] kworker/u192:6/2413 is trying to acquire lock:
[ 5233.245429] ffff888111632548 ((wq_completion)nvmet-wq){+.+.}-{0:0}, at: touch_wq_lockdep_map+0x26/0x90
[ 5233.251438]
but task is already holding lock:
[ 5233.255254] ffff888111632548 ((wq_completion)nvmet-wq){+.+.}-{0:0}, at: process_one_work+0x5cc/0x6e0
[ 5233.261125]
other info that might help us debug this:
[ 5233.265333] Possible unsafe locking scenario:
[ 5233.269217] CPU0
[ 5233.270795] ----
[ 5233.272436] lock((wq_completion)nvmet-wq);
[ 5233.275241] lock((wq_completion)nvmet-wq);
[ 5233.278020]
*** DEADLOCK ***
[ 5233.281793] May be due to missing lock nesting notation
[ 5233.286195] 3 locks held by kworker/u192:6/2413:
[ 5233.289192] #0: ffff888111632548 ((wq_completion)nvmet-wq){+.+.}-{0:0}, at: process_one_work+0x5cc/0x6e0
[ 5233.294569] #1: ffffc9000e2a7e40 ((work_completion)(&queue->release_work)){+.+.}-{0:0}, at: process_one_work+0x1c5/0x6e0
[ 5233.300128] #2: ffffffff82d7dc40 (rcu_read_lock){....}-{1:3}, at: __flush_work+0x62/0x530
[ 5233.304290]
stack backtrace:
[ 5233.306520] CPU: 4 UID: 0 PID: 2413 Comm: kworker/u192:6 Tainted: G O N 7.0.0-rc3nvme+ #20 PREEMPT(full)
[ 5233.306524] Tainted: [O]=OOT_MODULE, [N]=TEST
[ 5233.306525] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.17.0-0-gb52ca86e094d-prebuilt.qemu.org 04/01/2014
[ 5233.306527] Workqueue: nvmet-wq nvmet_tcp_release_queue_work [nvmet_tcp]
[ 5233.306532] Call Trace:
[ 5233.306534] <TASK>
[ 5233.306536] dump_stack_lvl+0x73/0xb0
[ 5233.306552] print_deadlock_bug+0x225/0x2f0
[ 5233.306556] __lock_acquire+0x13f0/0x2290
[ 5233.306563] lock_acquire+0xd0/0x300
[ 5233.306565] ? touch_wq_lockdep_map+0x26/0x90
[ 5233.306571] ? __flush_work+0x20b/0x530
[ 5233.306573] ? touch_wq_lockdep_map+0x26/0x90
[ 5233.306577] touch_wq_lockdep_map+0x3b/0x90
[ 5233.306580] ? touch_wq_lockdep_map+0x26/0x90
[ 52
---truncated--- |
| In the Linux kernel, the following vulnerability has been resolved:
dm-integrity: fix a bug if the bio is out of limits
If dm_integrity_check_limits fails, the code would exit with
DM_MAPIO_KILL. However, the range would be already locked at this point,
and it wouldn't be unlocked, resulting in a deadlock. Let's move the
limit check up, so that when it exits, no resources are leaked. |
| In the Linux kernel, the following vulnerability has been resolved:
drm/xe/guc: Hold device ref until queue teardown completes
GuC exec queue destruction can run asynchronously. If the final device
put happens from a destroy worker, drmm cleanup can end up draining
the same workqueue and deadlock.
Hold a drm_device reference for the queue lifetime and drop it after
queue teardown completes. This keeps drmm cleanup from running while
async destroy work is still pending.
Move GuC destroy work to a module-lifetime Xe workqueue and flush it
on PCI remove so hot-unbind/rebind still waits for pending destroy work.
With queue-held device refs, guc_submit_sw_fini() cannot run with live
GuC IDs. Replace the fini wait with an assertion and remove the unused
fini_wq.
v2:
- Rebase
v3:
- Switch to queue-lifetime drm_dev_get()/drm_dev_put() model. (Matt)
- Queue async teardown on system_dfl_wq instead of xe->destroy_wq. (Matt)
- Drop separate deferred drm_dev_put worker.
- Remove stale drain_workqueue(xe->destroy_wq) from guc_submit_sw_fini().
v4:
- Replace the guc_submit_sw_fini() wait with an assertion and remove
the now-unused fini_wq. (sashiko)
v5:
- Move destroy work to a module-lifetime Xe workqueue instead of
system_dfl_wq. (Matt)
- Flush the module-lifetime destroy workqueue during PCI remove to
preserve the old device-remove wait semantics.
v6:
- Keep SVM pagemap destroy work on the per-device destroy_wq to avoid
letting it outlive the xe_device/drm_device. (Sashiko)
- Use WQ_MEM_RECLAIM for xe->destroy_wq because SVM pagemap destroy work
can be queued from the reclaim path.
v7:
- Drop the per-device xe->destroy_wq and use the module-level destroy WQ
for SVM pagemap destroy as well. (Matt)
- Rename xe_exec_queue_destroy_wq_*() helpers to xe_destroy_wq_*()
helpers because the WQ is no longer exec-queue specific. (Matt)
v8:
- Rebase.
v9:
- Keep SVM pagemap destroy work on the per-device WQ_MEM_RECLAIM
destroy_wq because it can be queued from reclaim and embeds
the dev_pagemap used by devres teardown. (Sashiko)
- Keep the module-level destroy WQ GuC-only and drop WQ_MEM_RECLAIM
from it.
- Update the module-WQ kdoc to document the GuC/SVM split.
v10:
- Keep xe->destroy_wq per-cpu while adding WQ_MEM_RECLAIM to fix the
workqueue allocation warning.
v11:
- Drop the SVM pagemap destroy comment as it was revision-specific.
(Thomas)
v12:
- Rebase.
(cherry picked from commit da1124abac689cc2b1d8995e5f0a816f8a122edb) |
| Thelia before 3.1.5 and 3.2.x before 3.2.4 contains a business logic vulnerability that allows authenticated customers to bypass coupon usage limits because usage is counted only when orders become paid. Attackers can place multiple pay-later orders while a usage remains, or cancel paid orders to inflate max_usage, gaining unlimited discounts. |
| In the Linux kernel, the following vulnerability has been resolved:
misc: nsm: only unlock nsm_dev on post-lock error paths
nsm_dev_ioctl() jumps to the common out label even when the initial
copy_from_user() fails before nsm->lock has been taken. The error path
then blindly unlocks a mutex that was never acquired.
This issue was found by our static analysis tool and then manually
reviewed against the current tree.
The grounded PoC kept the miscdevice ioctl entry and the pre-lock
copy_from_user(&raw, argp, _IOC_SIZE(cmd)) failure path by issuing
NSM_IOCTL_RAW with an invalid user pointer. That failure reaches the
shared out label before mutex_lock(&nsm->lock). Lockdep reported:
WARNING: bad unlock balance detected!
exploit/193 is trying to release lock (&global_nsm.lock) at:
nsm_dev_ioctl+0x5f/0xcf [vuln_msv]
but there are no more locks to release!
no locks held by exploit/193.
Return immediately on the pre-lock copy_from_user() failure and keep the
common unlock label for the post-lock paths only. |
| In the Linux kernel, the following vulnerability has been resolved:
audit: fix recursive locking deadlock in audit_dupe_exe()
A deadlock occurs in the audit subsystem when duplicating
executable-related rules.
When a file is moved (e.g., via do_renameat2()), the VFS layer locks
the parent directory (I_MUTEX_PARENT), which synchronously triggers an
fsnotify_move event. If an existing executable audit rule matches the
file being moved, the audit subsystem catches this event and calls
audit_dupe_exe() to duplicate the watch and update the rule. Then,
audit_alloc_mark() would call kern_path_parent() to resolve the path,
leading to a blind attempt to acquire the exact same I_MUTEX_PARENT lock
already held by the task, resulting in the following recursive locking
deadlock:
============================================
WARNING: possible recursive locking detected
6.12.0-55.27.1.el10_0.x86_64+debug #1 Not tainted
--------------------------------------------
mv/5099 is trying to acquire lock:
ffff888132845358 (&inode->i_sb->s_type->i_mutex_dir_key/1){+.+.}-{3:3},
at: __kern_path_locked+0x10a/0x2f0
but task is already holding lock:
ffff888132846b58 (&inode->i_sb->s_type->i_mutex_dir_key/1){+.+.}-{3:3},
at: lock_two_directories+0x13f/0x2b0
other info that might help us debug this:
Possible unsafe locking scenario:
CPU0
----
lock(&inode->i_sb->s_type->i_mutex_dir_key/1);
lock(&inode->i_sb->s_type->i_mutex_dir_key/1);
*** DEADLOCK ***
May be due to missing lock nesting notation
6 locks held by mv/5099:
#0: ffff888112a9c440 (sb_writers#13)
at: do_renameat2+0x34c/0xbc0
#1: ffff888112a9c790 (&type->s_vfs_rename_key#3)
at: do_renameat2+0x415/0xbc0
#2: ffff888132846b58 (&inode->i_sb->s_type->i_mutex_dir_key/1)
at: lock_two_directories+0x13f/0x2b0
#3: ffff888132845358 (&inode->i_sb->s_type->i_mutex_dir_key/5)
at: lock_two_directories+0x175/0x2b0
#4: ffffffffb3a1fb10 (&fsnotify_mark_srcu)
at: fsnotify+0x454/0x28a0
#5: ffffffffaf886230 (audit_filter_mutex)
at: audit_update_watch+0x36/0x11e0
stack backtrace:
Call Trace:
<TASK>
dump_stack_lvl+0x6f/0xb0
print_deadlock_bug.cold+0xbd/0xca
validate_chain+0x83a/0xf00
__lock_acquire+0xcac/0x1d20
lock_acquire.part.0+0x11b/0x360
down_write_nested+0x9f/0x230
__kern_path_locked+0x10a/0x2f0
kern_path_locked+0x26/0x40
audit_alloc_mark+0xfb/0x4f0
audit_dupe_exe+0x6c/0xe0
audit_dupe_rule+0x6c2/0xc00
audit_update_watch+0x4cc/0x11e0
audit_watch_handle_event+0x12c/0x1b0
send_to_group+0x5d0/0x8b0
fsnotify+0x615/0x28a0
fsnotify_move+0x1d8/0x630
vfs_rename+0xdcd/0x1df0
do_renameat2+0x9d4/0xbc0
__x64_sys_renameat+0x192/0x260
do_syscall_64+0x92/0x180
entry_SYSCALL_64_after_hwframe+0x76/0x7e
RIP: 0033:0x7f0491fe8c4e
Code: 0f 1f 40 00 48 8b 15 c1 e1 16 00 f7 d8 64 89 02 b8 ff ff ff ff
c3 66 0f 1f 44 00 00 f3 0f 1e fa 49 89 ca b8 08 01 00 00 0f 05 <48>
3d 00 f0 ff ff 77 0a c3 66 0f 1f 84 00 00 00 00 00 48 8b 15 89
RSP: 002b:00007ffc7210bf38 EFLAGS: 00000246 ORIG_RAX: 0000000000000108
RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f0491fe8c4e
RDX: 0000000000000003 RSI: 00007ffc7210e6c8 RDI: 00000000ffffff9c
RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000001
R10: 00005575eb2dae2a R11: 0000000000000246 R12: 00005575eb2dae2a
R13: 00007ffc7210e6c8 R14: 0000000000000003 R15: 00000000ffffff9c
</TASK>
The aforementioned deadlock can be consistently reproduced by running
the script below:
audit-dupe-exe-deadlock.sh
--------------------------
#!/bin/bash
auditctl -D
mkdir -p /tmp/foo
touch /tmp/file
auditctl -a always,exit -F exe=/tmp/file -F path=/tmp/file -S all -k dr
mv /tmp/file /tmp/foo/file
rm -Rf /tmp/foo
This patch fixes the issue by introducing struct audit_watch_ctx to pass
the fsnotify event context down to audit_alloc_mark(). By utilizing the
already-resolved directory inode provided by the event, we bypass the
kern_path_parent() path resol
---truncated--- |
| In the Linux kernel, the following vulnerability has been resolved:
sched/rt: Have RT_PUSH_IPI be default off for non PREEMPT_RT
RT migration is done aggressively. When a CPU schedules out a high
priority RT task for a lower priority task, it will look to see if there's
any RT tasks that are waiting to run on another CPU that is of higher
priority than the task this CPU is about to run. If it finds one, it will
pull that task over to the CPU and allow it to run there instead.
Normally, this pulling is done by looking at the RT overloaded mask (rto)
which contains all the CPUs in the scheduler domain with RT tasks that are
waiting to run due to a higher priority RT task currently running on their
CPU. The CPU that is about to schedule a lower priority task will grab the
rq lock of the overloaded CPU and move the RT task from that CPU's runqueue
to the local one and schedule the higher priority RT task.
This caused issues when a lot of CPUs would schedule a lower priority task
at the same time. They would all try to grab the same runqueue lock of
the CPU with the overloaded RT tasks. Only the first CPU that got in will
get that task. All the others would wait until they got the runqueue lock
and see there's nothing to pull and do nothing. On systems with lots of
CPUs, this caused a large latency (up to 500us) which is beyond what
PREEMPT_RT is to allow.
The solution to that was to create an RT_PUSH_IPI logic. When any CPU
wanted to pull a task, instead of grabbing the runqueue lock of the
overloaded CPU, it would start by sending an IPI to the overloaded CPU,
and that IPI handler would have the CPU with the waiting RT task do a push
instead. Then that handler would send an IPI to the next CPU with
overloaded RT tasks, and so on. Note, after the first CPU starts this
process, if another CPU wanted to do a pull, it would see that the process
has already begun and would only increment a counter to have the IPIs
continue again.
The RT_PUSH_IPI solved the latency problem with PREEMPT_RT but could cause
a new issue with non PREEMPT_RT. Namely, softirqs run in a threaded
context on PREEMPT_RT but they can run in an interrupt context in non-RT.
If an IPI lands on a CPU that has just woken up multiple RT tasks and the
current CPU is running a non RT or a low priority RT task, instead of
doing a push, it would simply do a schedule on that CPU. But if a softirq
was also executing on this CPU, the schedule would need to wait until the
softirq finished. Until then, the CPU would still be considered overloaded
as there are RT tasks still waiting to run on it.
A live lock occurred on a workload that was doing heavy networking traffic
on a large machine where the softirqs would run 500us out of 750us. And it
would also be waking up RT tasks, causing the RT pull logic to be
constantly executed.
When a softirq triggered on a CPU with RT tasks queued but not running
yet, and the other CPUs would see this CPU as being overloaded, they would
send an IPI over to it. The CPU would notice that the waiting RT tasks are
of higher priority than the currently running task and simply schedule
that CPU instead. But because the softirq was executing, before it could
schedule, it would receive another IPI to do the same. The amount of IPIs
would slow down the currently running softirq so much that before it could
return back to task context, it would execute another softirq never
allowing the CPU to schedule. This live locked that CPU.
As RT_PUSH_IPI was created to help PREEMPT_RT, make it default off if
PREEMPT_RT is not enabled. |
