Recent commit changed the semantics from NOT_VALID to VALID.
I didn't realize that the flags are not stored atomically
with the entry in XArray. There's still a race of reader
observing a VALID mark for a slot, getting interrupted,
writer replacing the entry with a different one, reader
continuing, fetching the entry which is now a different
pointer than the pointer for which VALID was meant.
The biggest consequence of this is that we may see a UAF
since net_shaper_rollback() assumed that entries without
VALID can be freed without observing RCU.
Looks like the XArray marks are buying us nothing at this
point. Let's convert the code to an explicit valid field.
The smp_load_acquire() / smp_store_release() barriers are
marginally cleaner.
Reported-by: Sashiko <sashiko-bot@kernel.org>
Fixes: 93954b40f6 ("net-shapers: implement NL set and delete operations")
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260515221325.1685455-3-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
As previously discussed we don't care about making the shaper
state fully RCU-compliant because the hierarchy itself can't
be dumped in one go over Netlink. Let's annotate the reads
and writes to make that clear.
The field-by-field assignments will also be useful for the
next commit which adds explicit "valid" field (which we don't
want to override with the current full struct assignment).
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260515221325.1685455-2-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
mlx5e_xfrm_add_state() handles acquire-flow temporary SAs by allocating
software state and skipping hardware offload setup.
That path jumps to the common success label before taking the eswitch mode
block. After tunnel-mode validation was moved earlier, the common success
label unconditionally calls mlx5_eswitch_unblock_mode(). For acquire SAs,
this decrements esw->offloads.num_block_mode without a matching increment.
Return directly after installing the acquire SA offload handle, so only the
paths that successfully called mlx5_eswitch_block_mode() call the matching
unblock.
Fixes: 22239eb258 ("net/mlx5e: Prevent tunnel reformat when tunnel mode not allowed")
Signed-off-by: Prathamesh Deshpande <prathameshdeshpande7@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Reviewed-by: Tariq Toukan <tariqt@nvidia.com>
Link: https://patch.msgid.link/20260510225903.13184-1-prathameshdeshpande7@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Gal Pressman says:
====================
udp: gso: Fix __udp_gso_segment() after GSO_PARTIAL UDP length change
This series fixes two issues introduced by commit b10b446ce7 ("udp:
gso: Use single MSS length in UDP header for GSO_PARTIAL"), which
switched __udp_gso_segment() to write the single MSS length into the UDP
header for GSO_PARTIAL skbs.
Patch 1 (from Alice) fixes the UDP checksum adjustment in
__udp_gso_segment().
The patch adjusts the checksum by the correct delta, and since msslen
and newlen become equivalent before the loop, drops one of the two
variables to simplify the code.
Patch 2 handles the case where the last segment of a GSO_PARTIAL skb is
itself a GSO skb. This happens when the original packet size is an exact
multiple of MSS, so the post-loop segment is not a remainder skb but a
full GSO chunk and must also carry the single MSS length in its UDP
header.
====================
Link: https://patch.msgid.link/20260518062250.3019914-1-gal@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Following the cited commit, __udp_gso_segment() writes single MSS length
in the UDP header.
The cited patch doesn't account for the fact that the last segment could
be a GSO skb by itself. This could happen when the size of the packet is
a multiple of MSS, hence the first segment is also the last one (there
is no need for a remainder skb).
When the post-loop segment is a GSO skb, assign the single MSS length in
the UDP header.
Fixes: b10b446ce7 ("udp: gso: Use single MSS length in UDP header for GSO_PARTIAL")
Reported-by: Matthew Schwartz <matthew.schwartz@linux.dev>
Closes: https://lore.kernel.org/all/6c3fb15e-711d-4b8d-b152-e03d9b05293f@linux.dev/
Tested-by: Matthew Schwartz <matthew.schwartz@linux.dev>
Reviewed-by: Dragos Tatulea <dtatulea@nvidia.com>
Signed-off-by: Gal Pressman <gal@nvidia.com>
Link: https://patch.msgid.link/20260518062250.3019914-3-gal@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The cited commit started using msslen for uh->len, but still uses newlen
to adjust uh->check. Although the checksum is ignored in most cases due
to the hardware offload, __udp_gso_segment attempts to maintain the
correct one. Fix uh->check and adjust it by the right value.
Additionally, after the fix, newlen becomes assigned and unused before
the loop. The code can be simplified a bit if mss adjustment is dropped,
so that newlen becomes equal to msslen before the loop, and msslen can
be also dropped, saving a few lines of code.
This brings us back to one variable, drops an unneeded arithmetic for
mss, and fixes the UDP checksum.
Fixes: b10b446ce7 ("udp: gso: Use single MSS length in UDP header for GSO_PARTIAL")
Signed-off-by: Alice Mikityanska <alice@isovalent.com>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Signed-off-by: Gal Pressman <gal@nvidia.com>
Link: https://patch.msgid.link/20260518062250.3019914-2-gal@nvidia.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Booting with "nopcid" clears X86_FEATURE_PCID and keeps CR4.PCIDE from being
set to one. On AMD CPUs that support INVLPGB, broadcast TLB flushing remains
enabled.
There are two checks that decide whether the global ASID code runs,
mm_global_asid() and consider_global_asid(), that key off of the
X86_FEATURE_INVLPGB feature. Once an mm becomes active on more than three
CPUs, consider_global_asid() assigns it a global ASID, after which
flush_tlb_mm_range() takes the broadcast_tlb_flush() path using a non-zero
PCID. Issuing an INVLPGB with a non-zero PCID while CR4.PCIDE is not set
results in a #GP:
Oops: general protection fault, kernel NULL pointer dereference 0x1: 0000 [#1] SMP NOPTI
CPU: 158 UID: 0 PID: 3119 Comm: snap Not tainted 7.1.0-rc3 #1 PREEMPT(full)
Hardware name: ...
RIP: 0010:broadcast_tlb_flush
Code: ... 89 da 48 83 c8 07 <0f> 01 fe eb 08 cc cc cc ...
Call Trace:
<TASK>
flush_tlb_mm_range
ptep_clear_flush
wp_page_copy
? _raw_spin_unlock
__handle_mm_fault
handle_mm_fault
do_user_addr_fault
exc_page_fault
asm_exc_page_fault
All processors that support broadcast TLB invalidation also have PCID support,
so it is only the "nopcid" scenario that is of concern. In this situation just
disable the broadcast TLB support using the CPUID dependency support by making
X86_FEATURE_INVLPGB dependent on X86_FEATURE_PCID.
[ bp: Massage commit message. ]
Fixes: 4afeb0ed17 ("x86/mm: Enable broadcast TLB invalidation for multi-threaded processes")
Suggested-by: Dave Hansen <dave.hansen@intel.com>
Assisted-by: Claude:claude-opus-4.7
Signed-off-by: Tom Lendacky <thomas.lendacky@amd.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Acked-by: Rik van Riel <riel@surriel.com>
Cc: <stable@kernel.org>
Link: https://patch.msgid.link/b915acfd63e8b2a094fdeb8dc608738072518764.1779296450.git.thomas.lendacky@amd.com
On modern Xeon platforms, such as Granite Rapids, Sierra Forest, and
Clearwater Forest, there are two flavors of requestable C6 states: C6
and C6P. C6 allows only core C6 (CC6), while C6P allows both CC6 and
package C6 (PC6). PC6 saves more power but also has a higher exit
latency, so many users disable it in BIOS.
When PC6 is disabled, C6P becomes identical to C6 — the CPU treats C6P
requests as C6 requests. Exposing both C6 and C6P to user space in this
situation is confusing: two states with the same name look different but
behave the same. It also adds unnecessary overhead to the cpuidle
subsystem, which is a fast path: the governor evaluates every registered
state on idle entry.
Drop C-states that are redundant when PC6 is disabled by marking them
with CPUIDLE_FLAG_UNUSABLE, which causes cpuidle to exclude them when
registering idle states.
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Link: https://patch.msgid.link/20260425072532.358365-5-dedekind1@gmail.com
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Introduce the skx_is_pc6_disabled() for checking if PC6 is disabled and
switch the following functions to use it:
- skx_idle_state_table_update()
- spr_idle_state_table_update()
At the same time, clean them up improving the commentary and moving it to
the function kernel-doc.
Purely a clean up, no functional changes intended.
Reviewed-by: Andy Shevchenko <andriy.shevchenko@intel.com>
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Link: https://patch.msgid.link/20260425072532.358365-4-dedekind1@gmail.com
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Add two constants for the package C-state limit fields in
MSR_PKG_CST_CONFIG_CONTROL.
The SKX_ prefix stands for "Skylake Xeon" and makes it explicit that
the mask is CPU model-specific. The same values have applied to all
Xeon platforms starting from SKX.
Reviewed-by: Andy Shevchenko <andriy.shevchenko@intel.com>
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
Link: https://patch.msgid.link/20260425072532.358365-2-dedekind1@gmail.com
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
bt_accept_dequeue() unlinks a not-yet-accepted child from the parent
accept queue and release_sock()s it before returning, so the returned
sk has no caller reference and is unlocked.
l2cap_sock_cleanup_listen() walks these children on listening-socket
close. A concurrent HCI disconnect drives hci_rx_work ->
l2cap_conn_del() which runs l2cap_chan_del() + l2cap_sock_kill() and
frees the child sk and its l2cap_chan; cleanup_listen() then uses both:
BUG: KASAN: slab-use-after-free in l2cap_sock_kill
l2cap_sock_kill / l2cap_sock_cleanup_listen / __x64_sys_close
Freed by: l2cap_conn_del -> l2cap_sock_close_cb -> l2cap_sock_kill
This is distinct from the two fixes already in this area: commit
e83f5e24da ("Bluetooth: serialize accept_q access") serialises the
accept_q list/poll and takes temporary refs inside bt_accept_dequeue(),
and CVE-2025-39860 serialises the userspace close()/accept() race by
calling cleanup_listen() under lock_sock() in l2cap_sock_release().
Neither covers l2cap_conn_del() running from hci_rx_work, so this UAF
still reproduces on current bluetooth/master.
Take the reference at the source: bt_accept_dequeue() does sock_hold()
while sk is still locked, before release_sock(); callers sock_put().
cleanup_listen() pins the chan with l2cap_chan_hold_unless_zero() under
a brief child sk lock (serialising vs l2cap_sock_teardown_cb()), drops
it before l2cap_chan_lock(), and skips a duplicate l2cap_sock_kill() on
SOCK_DEAD. conn->lock is not taken here: cleanup_listen() runs under
the parent sk lock and that would invert
conn->lock -> chan->lock -> sk_lock (lockdep).
KASAN/SMP: an unprivileged listen/close vs HCI-disconnect race produced
12 use-after-free reports per run before this change; 0, and no lockdep
report, over 1600+ raced iterations after it on bluetooth/master.
Fixes: 15f02b9105 ("Bluetooth: L2CAP: Add initial code for Enhanced Credit Based Mode")
Cc: stable@vger.kernel.org
Reported-by: Siwei Zhang <oss@fourdim.xyz>
Reviewed-by: Siwei Zhang <oss@fourdim.xyz>
Signed-off-by: Safa Karakuş <safa.karakus@secunnix.com>
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
Vulnerabilities leading to Use-After-Free (UAF) and Null Pointer
Dereference (NPD) conditions were observed in the lifecycle management
of hci_uart.
The primary issue arises because the workqueues (init_ready and
write_work) are only flushed/cancelled if the HCI_UART_PROTO_READY
flag is set during TTY close. If a hangup occurs before setup completes,
hci_uart_tty_close() skips the teardown of these workqueues and
proceeds to free the `hu` struct. When the scheduled work executes
later, it blindly dereferences the freed `hu` struct.
Furthermore, several data races and UAFs were identified in the teardown
sequence:
1. Calling hci_uart_flush() from hci_uart_close() without effectively
disabling write_work causes a race condition where both can concurrently
double-free hu->tx_skb. This happens because protocol timers can
concurrently invoke hci_uart_tx_wakeup() and requeue write_work.
2. Calling hci_free_dev(hdev) before hu->proto->close(hu) causes a UAF
when vendor specific protocol close callbacks dereference hu->hdev.
3. In the initialization error paths, failing to take the proto_lock
write lock before clearing PROTO_READY leads to races with active
readers. Additionally, hci_uart_tty_receive() accesses hu->hdev
outside the read lock, leading to UAFs if the initialization error
path frees hdev concurrently.
Fix these synchronization and lifecycle issues by:
1. Re-ordering hci_uart_tty_close() to clear HCI_UART_PROTO_READY first,
followed immediately by a cancel_work_sync(&hu->write_work). Clearing
the flag locks out concurrent protocol timers from successfully invoking
hci_uart_tx_wakeup(), effectively rendering the cancellation permanent
and preventing the tx_skb double-free.
2. Note: Clearing PROTO_READY early causes hci_uart_close() to skip
hu->proto->flush(). This is perfectly safe in the tty_close path
because hu->proto->close() executes shortly after, which intrinsically
purges all protocol SKB queues and tears down the state.
3. Relocating hu->proto->close(hu) strictly prior to hci_free_dev(hdev)
across all close and error paths to prevent vendor-level UAFs.
4. Moving the hdev->stat.byte_rx increment in hci_uart_tty_receive()
inside the proto_lock read-side critical section to safely synchronize
with device unregistration.
5. Adding cancel_work_sync(&hu->write_work) to hci_uart_close() to safely
flush the workqueue before hci_uart_flush() is invoked via the HCI core.
6. Utilizing cancel_work_sync() instead of disable_work_sync() across
all paths to prevent permanently breaking user-space retry capabilities.
Fixes: 3b799254cf ("Bluetooth: hci_uart: Cancel init work before unregistering")
Cc: stable@vger.kernel.org
Signed-off-by: Mingyu Wang <25181214217@stu.xidian.edu.cn>
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
MGMT_OP_ADD_EXT_ADV_DATA is registered as a variable-length command,
with MGMT_ADD_EXT_ADV_DATA_SIZE as the fixed header size. The handler
then uses cp->adv_data_len and cp->scan_rsp_len to validate and copy
cp->data, but it never checks that those bytes are part of the mgmt
command payload.
A short command can therefore make add_ext_adv_data() pass an
out-of-bounds pointer into tlv_data_is_valid(). If the bytes beyond
the command buffer are addressable, they can also be copied into the
advertising instance as scan response data, where the caller can read
them back via MGMT_OP_GET_ADV_INSTANCE. The trigger requires
CAP_NET_ADMIN in the initial user namespace; KASAN reports an 8-byte
slab-out-of-bounds read.
Reject commands whose length does not match the fixed header plus both
advertising data lengths before parsing cp->data.
Fixes: 1241057283 ("Bluetooth: Break add adv into two mgmt commands")
Cc: stable@vger.kernel.org
Assisted-by: Claude:claude-opus-4-7
Signed-off-by: Michael Bommarito <michael.bommarito@gmail.com>
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
The setup_packet of control urb is not freed if usb_submit_urb fails or
the submitted urb is killed. Add free in these two paths.
Fixes: a1c49c434e ("Bluetooth: btusb: Add protocol support for MediaTek MT7668U USB devices")
Signed-off-by: Jiajia Liu <liujiajia@kylinos.cn>
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
ISO data PDUs carry a packet-boundary flag indicating START, CONT, END
or SINGLE. The ISO_CONT branch of iso_recv() guards against a missing
ISO_START by checking conn->rx_len before touching conn->rx_skb, but
ISO_END does not.
If a peer sends an ISO_END as the first packet on a fresh ISO
connection, conn->rx_skb is still NULL and conn->rx_len is zero, so
skb_put(conn->rx_skb, ...) dereferences NULL and oopses. For BIS,
where receivers sync to a broadcaster without pairing, any broadcaster
on the air can trigger this.
Mirror the ISO_CONT check at the top of ISO_END so a stray end fragment
is logged and dropped instead of crashing the host.
Fixes: ccf74f2390 ("Bluetooth: Add BTPROTO_ISO socket type")
Cc: stable@vger.kernel.org
Assisted-by: Claude:claude-opus-4-7
Signed-off-by: David Carlier <devnexen@gmail.com>
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
btintel_pcie_get_mac_access() and btintel_pcie_release_mac_access()
were programming STOP_MAC_ACCESS_DIS and XTAL_CLK_REQ in addition to
the MAC_ACCESS_REQ handshake. These bits are not part of the host
MAC-access handshake on the supported parts; the driver was
programming them incorrectly. Drop the writes so the register update
contains only the bits the controller actually consumes.
Fixes: b9465e6670 ("Bluetooth: btintel_pcie: Read hardware exception data")
Signed-off-by: Kiran K <kiran.k@intel.com>
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
This fixes not setting the bit for HCI_EVT_LE_ALL_REMOTE_FEATURES_COMPLETE
when extended features bit is set otherwise the controller may not
generate HCI_EVT_LE_ALL_REMOTE_FEATURES_COMPLETE causing
hci_le_read_all_remote_features_sync to timeout waiting for it.
Also remove dead code.
Fixes: a106e50be7 ("Bluetooth: HCI: Add support for LL Extended Feature Set")
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
bnep_add_connection() needs to keep holding the bnep_session_sem while
reading dev->name (just like bnep_get_connlist() does); otherwise the
bnep_session() thread can concurrently free the net_device, which can for
example be triggered by a concurrent bnep_del_connection().
(This UAF is fairly uninteresting from a security perspective;
calling bnep_add_connection() requires passing a capable(CAP_NET_ADMIN)
check. It also requires completely tearing down a netdev during a fairly
tight race window.)
Cc: stable@vger.kernel.org
Fixes: 1da177e4c3 ("Linux-2.6.12-rc2")
Signed-off-by: Jann Horn <jannh@google.com>
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
Certain drivers release resources (pinned pages, etc.) into system
memory during the prepare freeze PM op, making them swappable.
Currently, hibernate_preallocate_memory() is called before prepare
freeze, so those drivers have no opportunity to release resources
first. If a driver is holding a large amount of unswappable system
RAM, this can cause hibernate_preallocate_memory() to fail.
Move the call to hibernate_preallocate_memory() after prepare freeze.
According to the documentation for the prepare callback, devices should
be left in a usable state, so storage drivers should still be able to
service I/O requests. This allows drivers to release unswappable
resources prior to preallocation, so they can be swapped out through
hibernate_preallocate_memory()'s reclaim path.
Also remove shrink_shmem_memory() since hibernate_preallocate_memory()
will have reclaimed enough memory for the hibernation image.
Signed-off-by: Matthew Leach <matthew.leach@collabora.com>
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
[ rjw: Subject and changelog tweaks ]
Link: https://patch.msgid.link/20260403-hibernation-fixes-v3-1-31bc9fa3ba2d@collabora.com
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
hist_field_name() returns "" everywhere except the fully-qualified
VAR_REF/EXPR case, where snprintf() truncation returns NULL early
and bypasses the bottom NULL->"" guard. Callers don't expect NULL:
strcat(expr, hist_field_name(field, 0)) at trace_events_hist.c:1758
and the strcmp() in the sort-key match loop at :4804 both deref it.
system and event_name are bounded by MAX_EVENT_NAME_LEN, but the
field name on a VAR_REF is kstrdup'd from a histogram variable
name parsed out of the trigger string and has no length cap, so
a long enough var name in a fully qualified reference can reach
the truncation path.
Keep the length check but leave field_name as "" on overflow.
Link: https://patch.msgid.link/20260508195747.25492-1-devnexen@gmail.com
Fixes: 5ec1d1e97d ("tracing: Rebuild full_name on each hist_field_name() call")
Signed-off-by: David Carlier <devnexen@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Given how rbd_lock_add_request() and rbd_img_exclusive_lock() are
written, lock_dwork may be (re)queued more than it's actually needed:
for example in case a new I/O request comes in while we are in the
middle of rbd_acquire_lock() on behalf of another I/O request. This is
expected and with rbd_release_lock() preemptively canceling lock_dwork
is benign under normal operation.
A more problematic example is maybe_kick_acquire():
if (have_requests || delayed_work_pending(&rbd_dev->lock_dwork)) {
dout("%s rbd_dev %p kicking lock_dwork\n", __func__, rbd_dev);
mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, 0);
}
It's not unrealistic for lock_dwork to get canceled right after
delayed_work_pending() returns true and for mod_delayed_work() to
requeue it right there anyway. This is a classic TOCTOU race.
When it comes to unmapping the image, there is an implicit assumption
of no self-initiated exclusive lock activity past the point of return
from rbd_dev_image_unlock() which unlocks the lock if it happens to be
held. This unlock is assumed to be final and lock_dwork (as well as
all other exclusive lock tasks, really) isn't expected to get queued
again. However, lock_dwork is canceled only in cancel_tasks_sync()
(i.e. later in the unmap sequence) and on top of that the cancellation
can get in effect nullified by maybe_kick_acquire(). This may result
in rbd_acquire_lock() executing after rbd_dev_device_release() and
rbd_dev_image_release() run and free and/or reset a bunch of things.
One of the possible failure modes then is a violated
rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
in rbd_dev_header_info() which is called via rbd_dev_refresh() from
rbd_post_acquire_action().
Redo exclusive lock task draining to provide saner semantics and try
to meet the assumptions around rbd_dev_image_unlock().
Cc: stable@vger.kernel.org
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Reviewed-by: Viacheslav Dubeyko <Slava.Dubeyko@ibm.com>
A number of fixes to the dynamic epp feature which was new
to kernel 7.1, including making it opt in only.
-----BEGIN PGP SIGNATURE-----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=TNUP
-----END PGP SIGNATURE-----
Merge tag 'amd-pstate-v7.1-2026-05-14' of ssh://gitolite.kernel.org/pub/scm/linux/kernel/git/superm1/linux into pm-cpufreq-fixes
Merge amd-pstate fixes for 7.1 (05/14/2026) from Mario Limonciello:
"A number of fixes to the dynamic epp feature which was new
to kernel 7.1, including making it opt in only."
* tag 'amd-pstate-v7.1-2026-05-14' of ssh://gitolite.kernel.org/pub/scm/linux/kernel/git/superm1/linux:
cpufreq/amd-pstate: Drop Kconfig option for dynamic EPP
cpufreq/amd-pstate-ut: Drop policy reference before driver switch
cpufreq/amd-pstate: Use "epp_default_dc" as default when dynamic_epp is disabled
cpufreq/amd-pstate: Reorder notifier unregistration and floor perf reset
cpufreq/amd-pstate: Allow writes to dynamic_epp when state isn't modified
cpufreq/amd-pstate: Return -ENOMEM on failure to allocate profile_name
cpufreq/amd-pstate: Grab "amd_pstate_driver_lock" when toggling dynamic_epp
Commit 36df6e3dbd ("cgroup: make css_rstat_updated nmi safe") used
this_cpu_cmpxchg() for the lockless insertion, and therefore required
both ARCH_HAVE_NMI_SAFE_CMPXCHG and ARCH_HAS_NMI_SAFE_THIS_CPU_OPS in
the NMI guard: on archs without the latter, this_cpu_cmpxchg() falls
back to "local_irq_save() + plain cmpxchg", and local_irq_save()
cannot mask NMIs.
Commit 3309b63a22 ("cgroup: rstat: use LOCK CMPXCHG in
css_rstat_updated") later replaced this_cpu_cmpxchg() with plain
try_cmpxchg() to fix cross-CPU lockless-list corruption, but left the
NMI guard untouched. After that switch, css_rstat_updated() no longer
performs any this_cpu_*() RMW operations and only relies on the arch
having NMI-safe cmpxchg, so ARCH_HAS_NMI_SAFE_THIS_CPU_OPS is no
longer required in the guard.
Relax the guard accordingly so that archs which have HAVE_NMI and
ARCH_HAVE_NMI_SAFE_CMPXCHG but not ARCH_HAS_NMI_SAFE_THIS_CPU_OPS
(e.g. sparc, powerpc on PPC64/BOOK3S) can benefit from the existing
CONFIG_MEMCG_NMI_SAFETY_REQUIRES_ATOMIC path. Without this, the css
is never queued in NMI on those archs, and the atomics staged by
account_{slab,kmem}_nmi_safe() are not drained by flush_nmi_stats().
Fixes: 3309b63a22 ("cgroup: rstat: use LOCK CMPXCHG in css_rstat_updated")
Signed-off-by: Cunlong Li <shenxiaogll@gmail.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
nvme_update_ns_info_block() trusts id->lbaf[lbaf].ds from the
controller and assigns it directly to ns->head->lba_shift without
bounds checking. nvme_lba_to_sect() then does:
return lba << (head->lba_shift - SECTOR_SHIFT);
When called with lba = le64_to_cpu(id->nsze) to compute the device
capacity, an attacker-controlled controller can choose ds < 9 or a
combination of (ds, nsze) that makes the left shift overflow
sector_t. The former is a C undefined behaviour that UBSAN reports
as a BUG; the latter silently yields a bogus capacity that the
block layer then trusts for bounds checking.
Validate ds against SECTOR_SHIFT and use check_shl_overflow() to
compute capacity so that any (ds, nsze) combination that would
overflow sector_t is rejected. The namespace is skipped with
-ENODEV instead of crashing the kernel. This is reachable by a
malicious NVMe device, a buggy firmware, or an attacker-controlled
NVMe-oF target.
The check is performed before queue_limits_start_update() and
blk_mq_freeze_queue(), so the error path is a plain `goto out` with
no cleanup needed.
Stack trace (UBSAN, ds < 9 variant):
RIP: nvme_lba_to_sect drivers/nvme/host/nvme.h:699 [inline]
RIP: nvme_update_ns_info_block.cold+0x5/0x7
Call Trace:
nvme_update_ns_info+0x175/0xd90 drivers/nvme/host/core.c:2467
nvme_validate_ns drivers/nvme/host/core.c:4299 [inline]
nvme_scan_ns drivers/nvme/host/core.c:4350
nvme_scan_ns_async+0xa5/0xe0 drivers/nvme/host/core.c:4383
async_run_entry_fn
process_one_work
worker_thread
kthread
Found by Syzkaller.
Acked-by: Sungwoo Kim <iam@sung-woo.kim>
Acked-by: Dave Tian <daveti@purdue.edu>
Acked-by: Weidong Zhu <weizhu@fiu.edu>
Signed-off-by: Chao Shi <coshi036@gmail.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Since commit
13f4e660a1 ("thermal/core: Split __thermal_cooling_device_register()
into two functions")
thermal_cooling_device_setup_sysfs() is called before the
->get_max_state() callback in thermal_cooling_device_add().
However, cooling_device_stats_setup() allocates space based on
cdev->max_state, which is not initialized at that point.
With CONFIG_THERMAL_STATISTICS=y, an out of bounds access happens
inside thermal_cooling_device_stats_update(), followed by a kernel
crash:
Unable to handle kernel paging request at virtual address ffff800081329e60
Call trace:
queued_spin_lock_slowpath+0x1cc/0x320 (P)
thermal_cooling_device_stats_update+0x28/0xa4
__thermal_cdev_update+0x74/0x88
thermal_cdev_update+0x44/0x58
step_wise_manage+0x1b8/0x300
__thermal_zone_device_update+0x270/0x414
thermal_zone_device_check+0x28/0x40
process_one_work+0x150/0x290
worker_thread+0x18c/0x300
kthread+0x114/0x120
ret_from_fork+0x10/0x20
To fix this, restore the original ordering of ->get_max_state() and
thermal_cooling_device_setup_sysfs(). Note that with this reordering,
the dev_set_name() and ->get_max_state() error paths now reach
thermal_cdev_release() without setup_sysfs() having run. This is safe
because cdev->stats is NULL in that case and destroy_sysfs() is a no-op.
Fixes: 13f4e660a1 ("thermal/core: Split __thermal_cooling_device_register() into two functions")
Signed-off-by: Ovidiu Panait <ovidiu.panait.rb@renesas.com>
Reviewed-by: Daniel Lezcano <daniel.lezcano@oss.qualcomm.com>
Link: https://patch.msgid.link/20260520165835.90974-1-ovidiu.panait.rb@renesas.com
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Currently, nvme_alloc_admin_tag_set() silently drops and releases
the existing admin_q if it called on a controller that already
had one (e.g., during a controller reset).
However, transport drivers should not be reallocating the admin tag
set and queue during a reset. Dropping the old queue and allocating
a new one destroys user-configured timeouts and may race against
nvme_admin_timeout_store()
Since all transport drivers are now expected to preserve the admin queue
across resets, calling nvme_alloc_admin_tag_set() when ctrl->admin_q
is already populated is a bug.
Remove the silent cleanup and replace it with a WARN_ON_ONCE() to
explicitly catch any transport drivers that violate this lifecycle rule
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Daniel Wagner <dwagner@suse.de>
Signed-off-by: Maurizio Lombardi <mlombard@redhat.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Currently, resetting a loopback controller unconditionally invokes
nvme_alloc_admin_tag_set() inside nvme_loop_configure_admin_queue().
Doing so drops the old queue and allocates a new one. Consequently,
this reverts the admin queue's timeout (q->rq_timeout) back to the
module default (NVME_ADMIN_TIMEOUT), completely wiping out any custom
timeout values the user may have configured via sysfs and potentially
racing against the sysfs nvme_admin_timeout_store() function
that may dereference the admin_q pointer during the RESETTING state.
Decouple the admin tag set lifecycle from the admin queue
configuration and destruction paths, which are executed during resets;
Specifically:
* Move nvme_alloc_admin_tag_set() into nvme_loop_create_ctrl() so it
is only allocated once during the initial controller creation.
* Defer the destruction of the admin tag set to
nvme_loop_delete_ctrl_host() and the terminal error-handling
paths of nvme_loop_reset_ctrl_work() and
nvme_loop_create_ctrl().
Reviewed-by: Daniel Wagner <dwagner@suse.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Signed-off-by: Maurizio Lombardi <mlombard@redhat.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Currently, the final reference for the fabrics admin queue (fabrics_q)
is dropped inside nvme_remove_admin_tag_set(). However, the primary admin
queue (admin_q) defers dropping its final reference until
nvme_free_ctrl().
Move the blk_put_queue() call for fabrics_q from
nvme_remove_admin_tag_set() to nvme_free_ctrl(). This aligns the
lifecycle management of both admin queues, ensuring they are freed
symmetrically when the controller is finally torn down.
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Daniel Wagner <dwagner@suse.de>
Signed-off-by: Maurizio Lombardi <mlombard@redhat.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Currently, there is no method to adjust the timeout values on a
per controller basis with nvme I/O queues.
Add an io_timeout attribute to nvme so that different nvme controllers
which may have different timeout requirements can have custom I/O
timeouts set.
The I/O timeout is also applied to the connect queue (connect_q).
In NVMe over Fabrics, the connect queue is utilized specifically to
issue Connect commands that establish the I/O queues.
Reviewed-by: Mohamed Khalfella <mkhalfella@purestorage.com>
Reviewed-by: Daniel Wagner <dwagner@suse.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Maurizio Lombardi <mlombard@redhat.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Currently, there is no method to adjust the timeout values on a
per-controller basis with nvme admin queues.
Add an admin_timeout attribute to nvme so that different nvme controllers
which may have different timeout requirements can have custom admin
timeouts set.
The admin timeout is also applied to the fabrics queue (fabrics_q).
The fabrics queue is utilized for fabric-specific administrative and
control operations, such as Connect and Property Get/Set commands.
Reviewed-by: Daniel Wagner <dwagner@suse.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Mohamed Khalfella <mkhalfella@purestorage.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Maurizio Lombardi <mlombard@redhat.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
All callers of nvme_wait_freeze_timeout() currently pass the exact same
NVME_IO_TIMEOUT default as their timeout argument.
Remove it and use a local variable.
Reviewed-by: Daniel Wagner <dwagner@suse.de>
Reviewed-by: Mohamed Khalfella <mkhalfella@purestorage.com>
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Maurizio Lombardi <mlombard@redhat.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
When initializing an nvme request which is about to be send to the block
layer, we do not need to initialize its timeout. If it's left
uninitialized at 0 the block layer will use the request queue's timeout
in blk_add_timer (via nvme_start_request which is called from
nvme_*_queue_rq). These timeouts are setup to either NVME_IO_TIMEOUT or
NVME_ADMIN_TIMEOUT when the request queues were created.
Because the io_timeout of the IO queues can be modified via sysfs, the
following situation can occur:
1) NVME_IO_TIMEOUT = 30 (default module parameter)
2) nvme1n1 is probed. IO queues default timeout is 30 s
3) manually change the IO timeout to 90 s
echo 90000 > /sys/class/nvme/nvme1/nvme1n1/queue/io_timeout
4) Any call of __submit_sync_cmd on nvme1n1 to an IO queue will issue
commands with the 30 s timeout instead of the wanted 90 s which might
be more suitable for this device.
Commit 470e900c80 ("nvme: refactor nvme_alloc_request") silently
changed the behavior for ioctl's already because it unconditionally
overrides the request's timeout that was set in nvme_init_request. If it
was unset by the user of the ioctl if will be overridden with 0 meaning
the block layer will pick the request queue's IO timeout.
Following up on that, this patch further improves the consistency of IO
timeout usage. However, there are still uses of NVME_IO_TIMEOUT which
could be inconsistent with what is set in the device's request_queue by
the user.
Reviewed-by: Mohamed Khalfella <mkhalfella@purestorage.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Daniel Wagner <dwagner@suse.de>
Reviewed-by: Hannes Reinecke <hare@kernel.org>
Signed-off-by: Maximilian Heyne <mheyne@amazon.de>
Signed-off-by: Maurizio Lombardi <mlombard@redhat.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
The mdacon driver supports using ISA MDA or Hercules-compatible display
adapters as a secondary text console. This was commonly used in the
1990s and earlier for debugging software which took over the primary
display. It is highly unlikely anyone is doing so nowadays because
serial consoles and much better methods of debugging exist.
The driver is not enabled by any defconfig, nor any of the
dozens of distro configs collected at [1]. It has been relegated to VTs
13-16 since commit 0b9cf3aa6b ("mdacon messing up default vc's - set
default to vc13-16 again") in Linux 2.6.27 (and before Linux 2.5.53 -
see the link in the message of the above commit). The change in 2.6.27
was done because it was incorrectly detecting non-MDA adapters as MDA
and taking over all VTs, rendering them unusable.
Furthermore, vgacon supports using MDA/Hercules-compatible adapters as
the primary text console, so any systems with only one of these
adapters were already using vgacon and will not experience any loss in
functionality from the removal of this driver.
Given all of these factors, the mdacon driver is likely entirely
unused. Remove it.
[1] f0bee86a13
Signed-off-by: Ethan Nelson-Moore <enelsonmoore@gmail.com>
Signed-off-by: Helge Deller <deller@gmx.de>
The hgafb driver supports graphics adapters compatible with the Hercules
adapter from 1984. These were ISA cards or onboard devices that supported
monochrome 720x348 graphics. This driver was created in 1999 by Ferenc Bakonyi.
In the entire Git history (since Linux 2.6.12-rc2), there has only been one
commit in 2010 which indicated that the driver was in use, commit 529ed806d4
("video: Fix the HGA framebuffer driver"). The commit message states:
Only tested with fbcon, since most fbdev-based software appears
to only support 12bpp and up. It does not appear that this driver has
worked for at least the entire 2.6.x series, perhaps since 2002.
Given the age and limited capabilities of the hardware and the lack of
users, remove this driver and move the former maintainer to CREDITS.
Signed-off-by: Ethan Nelson-Moore <enelsonmoore@gmail.com>
Acked-by: Thomas Zimmermann <tzimmermann@suse.de>
Signed-off-by: Helge Deller <deller@gmx.de>
Running clang-tidy on a program that uses getopt() from nolibc
this warning appears:
getopt.h:80:6: warning: Out of bound access to memory after the end of the string literal [clang-analyzer-security.ArrayBound]
80 | if (optstring[i] == ':') {
This looks like a very unlikely case that an argument
inside of argv is being changed between getopt() calls.
Adding a check for d becoming 0 in the guard after the loop
stops getopt() getting far enough to access beyond the end
of the array and seems to correct the issue.
Fixes: bae3cd708e ("tools/nolibc: add getopt()")
Assisted-by: Claude:claude-4.6-sonnet # reproducer
Signed-off-by: Daniel Palmer <daniel@thingy.jp>
Link: https://patch.msgid.link/20260520111931.1027758-1-daniel@thingy.jp
[Thomas: clean up commit message a bit]
Signed-off-by: Thomas Weißschuh <linux@weissschuh.net>
The ldo_vcn33_[12]_wifi and ldo_vcn33_[12]_bt are just two regulator
outputs instead of four. The wifi and bt parts refer to separate enable
bits that are OR-ed together to affect the actual regulator output. The
separate bits allow the wifi and bt stacks to enable their power without
coordination between them. These have been deprecated in favor of proper
nodes matching the output.
Add proper ldo_vcn33_[12] regulators to replace the existing ones. The
enable status is synced to just one of the two enable bits, and the
other is forced off. This makes the handling in other bits simpler.
The existing *_(bt|wifi) regulators are converted to no-op regulators
that are fed from their new respective ldo_vcn33_[12] regulator. This
allows existing device trees to continue to work.
Signed-off-by: Chen-Yu Tsai <wenst@chromium.org>
Link: https://patch.msgid.link/20260514091520.2718987-7-wenst@chromium.org
Signed-off-by: Mark Brown <broonie@kernel.org>
The MT6359 regulator DT binding defines the supply names for the PMIC.
Add support for them by adding .supply_name field settings for each
regulator. The buck regulators each have their own supply. The name
of the supply is related to the name of the buck regulator. The LDOs
have shared supplies.
Add the supply name to the declaration of each regulator. At the moment
they are declared explicitly, but the buck regulator macro can be made
to derive both the match string and supply name from the base name once
the *_sshub regulators are figured out and removed. For context, the
*_sshub regulators are not separate regulators, but separate settings
for the same name regulators without the "_sshub" suffix.
Signed-off-by: Chen-Yu Tsai <wenst@chromium.org>
Link: https://patch.msgid.link/20260514091520.2718987-6-wenst@chromium.org
Signed-off-by: Mark Brown <broonie@kernel.org>
The regulator descriptions and extended descriptions don't change at
runtime. The only reason they are not const is that the regulator
driver data is non-const.
Const-ify the descriptions and all references to them. For the driver
data, explicitly cast it to non-const void *.
Signed-off-by: Chen-Yu Tsai <wenst@chromium.org>
Link: https://patch.msgid.link/20260514091520.2718987-5-wenst@chromium.org
Signed-off-by: Mark Brown <broonie@kernel.org>
vcn33_[12]_bt and vcn33_[12]_wifi refer to the same output. There are
two enable bits in the registers so that BT and WiFi drivers can toggle
them separately without any coordination. If either bit is set, then the
regulator output is enabled.
Deprecate the existing regulators, and add proper regulators matching
the outputs: vcn33_1 and vcn33_2.
Signed-off-by: Chen-Yu Tsai <wenst@chromium.org>
Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@oss.qualcomm.com>
Link: https://patch.msgid.link/20260514091520.2718987-4-wenst@chromium.org
Signed-off-by: Mark Brown <broonie@kernel.org>
The name of the regulator should match what the board design specifies
for the power rail. There should be no limitations on what the name can
be, and they definitely don't always follow the PMIC's own names.
Drop the restrictions on regulator-name.
Fixes: 8771456635 ("dt-bindings: regulator: Add document for MT6359 regulator")
Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@oss.qualcomm.com>
Signed-off-by: Chen-Yu Tsai <wenst@chromium.org>
Link: https://patch.msgid.link/20260514091520.2718987-3-wenst@chromium.org
Signed-off-by: Mark Brown <broonie@kernel.org>
A previous commit deferred this to the task_work part of it, so it could
be protected by ->uring_lock. But that's actually not necessary here,
and in fact the head clearing is not enough to make that safe. For those
two reasons, just re-instate the local splicing.
Fixes: 49ae66eb8c ("io_uring: defer linked-timeout chain splice out of hrtimer context")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Fix a regression introduced by commit 61bbcfb505 ("srcu: Push
srcu_node allocation to GP when non-preemptible"): SRCU may queue works
on CPUs that are "possible" but never have been online. In such a case,
the work callbacks may not be executed until the corresponding CPU gets
online, and as the callbacks accumulates, workqueue lockups will fire.
Fix this by avoiding queuing works on CPUs that have never been online.
-----BEGIN PGP SIGNATURE-----
iQFhBAABCABLFiEEj5IosQTPz8XU1wRHSXnow7UH+rgFAmoMy/EbFIAAAAAABAAO
bWFudTIsMi41KzEuMTIsMCwzERxib3F1bkBrZXJuZWwub3JnAAoJEEl56MO1B/q4
lT8H/RlNu00LC24b0JxPYRBZJz3TSM2WlnGJQ+5LXSQgk2ecqzoTzwDE7oC3naPC
QDkwGpSif8Y5OKaEnlVavtDcHdNa824mKrgRo/nXkk5fqrrMMubHOHe5Y0fwy5z5
upPoEvEs0XtbW4Mm6lI4uRw+qvIH16+Ud9SMrfZMwLRGaO8axBXi3rijUtfAMRGv
xBFqJX15Z/ixWkA6aHGuM1fI4WdApUen4/W3oUC+Ka4Lpgtt29GmIOV3n/topNQq
R8bZM9QC+7f6Vk1s49ywD9WZYa1b4Pig74XDOXn3328kulGBxVtOnOz4sXMgBmBF
ZLvJ4xXy8+u1eM0DFcSExJCCvHc=
=PyvW
-----END PGP SIGNATURE-----
Merge tag 'rcu-fixes.v7.1-20260519a' of git://git.kernel.org/pub/scm/linux/kernel/git/rcu/linux
Pull RCU fixes from Boqun Feng:
"Fix a regression introduced by commit 61bbcfb505 ("srcu: Push
srcu_node allocation to GP when non-preemptible").
SRCU may queue works on CPUs that are "possible" but never have been
online. In such a case, the work callbacks may not be executed until
the corresponding CPU gets online, and as the callbacks accumulates,
workqueue lockups will fire.
Fix this by avoiding queuing works on CPUs that have never been
online"
* tag 'rcu-fixes.v7.1-20260519a' of git://git.kernel.org/pub/scm/linux/kernel/git/rcu/linux:
srcu: Don't queue workqueue handlers to never-online CPUs
sysfs_kf_seq_show() defends against buggy show() callbacks that return
larger than PAGE_SIZE by clamping the value and printing a warning.
sysfs_kf_read(), the prealloc variant, has no such defense.
The only current in-tree user of __ATTR_PREALLOC is drivers/md/md.c,
whose show() callbacks are well-behaved, so this is hardening against
future drivers doing foolish things and out-of-tree code doing even more
foolish things.
Cc: NeilBrown <neil@brown.name>
Cc: Tejun Heo <tj@kernel.org>
Fixes: 2b75869bba ("sysfs/kernfs: allow attributes to request write buffer be pre-allocated.")
Assisted-by: gregkh_clanker_t1000
Reviewed-by: Rafael J. Wysocki (Intel) <rafael@kernel.org>
Reviewed-by: Danilo Krummrich <dakr@kernel.org>
Link: https://patch.msgid.link/2026052000-drove-unicycle-d61b@gregkh
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When sysfs_update_group() is called for a named group and create_files()
fails (e.g. -ENOMEM), internal_create_group() calls kernfs_remove(kn) on
the group directory. In the update path, kn was obtained via
kernfs_find_and_get() and refers to a directory that already existed
before this call. Removing it silently destroys a sysfs group that the
caller did not create.
Only remove the directory if we created it ourselves. On update failure
the directory remains as it is left empty by remove_files() inside
create_files(), but can be repopulated by a retry.
Cc: Rajat Jain <rajatja@google.com>
Fixes: c855cf2759 ("sysfs: Fix internal_create_group() for named group updates")
Cc: stable <stable@kernel.org>
Assisted-by: gkh_clanker_t1000
Reviewed-by: Rafael J. Wysocki (Intel) <rafael@kernel.org>
Reviewed-by: Danilo Krummrich <dakr@kernel.org>
Link: https://patch.msgid.link/2026052003-uniquely-hastily-c093@gregkh
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
virtio_gpu_cursor_plane_update() and virtio_gpu_resource_flush() lock
the framebuffer BO's dma_resv via virtio_gpu_array_lock_resv() and
ignore its return value. The function can fail with -EINTR from
dma_resv_lock_interruptible() (signal during lock wait) or with
-ENOMEM from dma_resv_reserve_fences() (fence slot allocation),
leaving the resv lock not held. The queue path then walks the object
array and calls dma_resv_add_fence(), which requires the lock held;
with lockdep enabled this trips dma_resv_assert_held():
WARNING: drivers/dma-buf/dma-resv.c:296 at dma_resv_add_fence+0x71e/0x840
Call Trace:
virtio_gpu_array_add_fence
virtio_gpu_queue_ctrl_sgs
virtio_gpu_queue_fenced_ctrl_buffer
virtio_gpu_cursor_plane_update
drm_atomic_helper_commit_planes
drm_atomic_helper_commit_tail
commit_tail
drm_atomic_helper_commit
drm_atomic_commit
drm_atomic_helper_update_plane
__setplane_atomic
drm_mode_cursor_universal
drm_mode_cursor_common
drm_mode_cursor_ioctl
drm_ioctl
__x64_sys_ioctl
Beyond the WARN, mutating the dma_resv fence list without the lock
races with concurrent readers/writers and can corrupt the list.
Both call sites run inside the .atomic_update plane callback, which
DRM atomic helpers do not allow to fail (by the time it runs, the
commit has been signed off to userspace and there is no clean
rollback path). Moving the lock acquisition to .prepare_fb was
rejected because the broader lock scope deadlocks against other BO
locking paths in the same atomic commit.
Introduce virtio_gpu_lock_one_resv_uninterruptible() that uses
dma_resv_lock() instead of dma_resv_lock_interruptible(). This
eliminates the -EINTR failure mode -- the realistic syzbot trigger
-- without extending the lock hold across the commit. The helper
locks a single BO and rejects nents > 1 with -EINVAL; both fix
sites lock exactly one BO.
Use it from virtio_gpu_cursor_plane_update() and
virtio_gpu_resource_flush(); check the return value to handle the
remaining -ENOMEM case from dma_resv_reserve_fences() by freeing
the objs and skipping the plane update for that frame. The
framebuffer BOs touched here are not shared with other contexts
and lock contention is expected to be brief, so the loss of
signal-interruptibility is acceptable.
Other callers of virtio_gpu_array_lock_resv() (the ioctl paths)
continue to use the interruptible variant.
The bug was reported by syzbot, triggered via fault injection
(fail_nth) on the DRM_IOCTL_MODE_CURSOR path, which forces the
-ENOMEM branch in dma_resv_reserve_fences().
Reported-by: syzbot+72bd3dd3a5d5f39a0271@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=72bd3dd3a5d5f39a0271
Fixes: 5cfd31c5b3 ("drm/virtio: fix virtio_gpu_cursor_plane_update().")
Cc: stable@vger.kernel.org
Signed-off-by: Deepanshu Kartikey <kartikey406@gmail.com>
Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
Link: https://patch.msgid.link/20260519082247.34470-1-kartikey406@gmail.com
While being less compact, using named initializers allows to more easily
see which members of the structs are assigned which value without having
to lookup the declaration of the struct. And it's also more robust
against changes to the struct definition.
The mentioned robustness is relevant for a planned change to struct
i2c_device_id that replaces .driver_data by an anonymous union.
This patch doesn't modify the compiled array, only its representation
in source form benefits. The former was confirmed with x86 and arm64
builds.
Signed-off-by: Uwe Kleine-König (The Capable Hub) <u.kleine-koenig@baylibre.com>
Link: https://patch.msgid.link/20260519093806.1567914-2-u.kleine-koenig@baylibre.com
Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@oss.qualcomm.com>
