Replace the per-enctype .decrypt callbacks (gss_krb5_aes_decrypt
and krb5_etm_decrypt) with a single gss_krb5_aead_decrypt()
wrapper that delegates to crypto_krb5_decrypt().
The new wrapper builds a scatterlist covering the secured
region (confounder through checksum), passes it to the AEAD
decrypt operation, and derives the confounder and checksum
lengths from the data offset and length that
crypto_krb5_decrypt() reports. The caller's token header
verification and buffer adjustment logic is unchanged.
Assisted-by: Claude:claude-opus-4-6
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Acked-by: Anna Schumaker <anna.schumaker@hammerspace.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Replace the per-enctype .encrypt callbacks (gss_krb5_aes_encrypt and
krb5_etm_encrypt) with a single gss_krb5_aead_encrypt() wrapper that
delegates to crypto_krb5_encrypt().
The xdr_buf setup -- GSS header insertion, confounder space
allocation, and token header copy -- remains unchanged. The
difference is that the CBC-CTS encryption and HMAC computation are
now a single AEAD operation through the crypto/krb5 library. Both
the MtE construction (RFC 3962) and the EtM construction (RFC 8009)
are handled transparently by the AEAD transform.
The plaintext page data must be copied from the page cache pages to
the scratch output pages before building the scatterlist, since the
AEAD operates in-place rather than using separate input and output
scatterlists.
Assisted-by: Claude:claude-opus-4-6
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Acked-by: Anna Schumaker <anna.schumaker@hammerspace.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Allocate crypto_aead handles for encryption (one per direction)
and crypto_shash handles for checksumming (one per direction)
using the crypto/krb5 library's key preparation functions.
These four handles derive their subkeys from the session key
and the RFC 4121 usage numbers and are ready for use in
encrypt, decrypt, get_mic, and verify_mic operations.
The existing crypto_sync_skcipher and crypto_ahash handles
remain in place for now; subsequent patches switch the
per-message operations to the new handles and then remove
the old ones.
Assisted-by: Claude:claude-opus-4-6
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Acked-by: Anna Schumaker <anna.schumaker@hammerspace.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The crypto/krb5 library returns standard negative errno values,
but the GSS mechanism layer reports results as GSS_S_* major
status codes. A translation is needed at each call site that
will be switched to the new library.
Rather than open-coding the mapping in every wrapper, provide a
single helper function.
Assisted-by: Claude:claude-opus-4-6
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Acked-by: Anna Schumaker <anna.schumaker@hammerspace.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The crypto/krb5 library accepts data in scatterlist form, but
the GSS-API layer presents RPC payloads as struct xdr_buf.
Bridge that gap with a pair of helper functions:
xdr_buf_to_sg() - populate a caller-supplied scatterlist
array from a byte range
xdr_buf_to_sg_alloc() - populate a caller-supplied inline
scatterlist, chaining to a heap-
allocated overflow for large payloads
The inline array (typically stack-allocated at eight entries)
covers the common case of small RPCs with no heap allocation
on the encrypt/decrypt path. Only buffers spanning many pages
incur a kmalloc for the chained extension.
The segment-walking logic follows the same head, page array,
tail traversal as xdr_process_buf(), but populates a
scatterlist directly rather than invoking a per-segment
callback. sg_next() traversal makes the walker safe for
chained scatterlists. Once subsequent patches reroute all
per-message crypto operations through crypto/krb5,
xdr_process_buf() loses its last callers and is removed.
Assisted-by: Claude:claude-opus-4-6
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Acked-by: Anna Schumaker <anna.schumaker@hammerspace.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Each krb5_ctx currently points to a gss_krb5_enctype, the
rpcsec_gss_krb5 module's own enctype descriptor. To begin
using the common crypto/krb5 library, store a pointer to the
corresponding struct krb5_enctype (from <crypto/krb5.h>) as
well.
The lookup is performed in gss_import_v2_context() immediately
after the existing gss_krb5_lookup_enctype() call. If
crypto_krb5_find_enctype() cannot find a matching enctype the
context import fails, ensuring the module never operates with
a partially-initialized krb5_ctx.
Assisted-by: Claude:claude-opus-4-6
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Acked-by: Anna Schumaker <anna.schumaker@hammerspace.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The rpcsec_gss_krb5 module currently contains its own Kerberos 5
crypto implementation (key derivation, encryption, checksumming)
that duplicates functionality available in the common crypto/krb5
library. As a first step toward migrating to that library, add a
Kconfig select so that building rpcsec_gss_krb5 pulls in the
common Kerberos 5 crypto support.
The per-enctype Kconfig options (AES_SHA1, CAMELLIA, AES_SHA2)
remain: they continue to gate which encryption types are offered
by the GSS mechanism. The individual crypto algorithm selects
they carry become redundant once the migration is complete, since
CRYPTO_KRB5 already selects all needed ciphers and hashes.
Assisted-by: Claude:claude-opus-4-6
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Acked-by: Anna Schumaker <anna.schumaker@hammerspace.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Commit 8a81f16de6 ("NFSD: Add a "default" block size") introduced
NFSSVC_DEFBLKSIZE at 1MB, well below the 4MB NFSSVC_MAXBLKSIZE
ceiling, with the stated intent that a later change would raise the
default.
Raising the default reduces per-RPC overhead on fast networks by
amortizing header processing and scheduling costs across larger
payloads. The halving loop in nfsd_get_default_max_blksize()
constrains the returned value to 1/4096 of available RAM, so the
new 4MB default takes effect only on systems with at least 16GB of
RAM. Smaller machines continue to receive the same computed value
as before. Administrators can still override the computed value
through /proc/fs/nfsd/max_block_size.
On systems where the new default takes effect,
svc_sock_setbufsize() sizes each service socket's send and receive
buffers as nreqs * max_mesg * 2. Quadrupling max_mesg therefore
quadruples the per-socket buffer reservation at a fixed thread
count, which operators tuning large thread pools should account
for.
Note well: Your NFS client implementation must support large read
and write size settings to benefit from this change.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Roland Mainz <roland.mainz@nrubsig.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
When NFSD_CMD_UNLOCK_EXPORT revokes NFSv4 state for an export path,
GC-managed nfsd_file entries for files under that path may remain
in the file cache. These cached handles hold the underlying
filesystem busy, preventing a subsequent unmount.
Add nfsd_file_close_export(), which walks the nfsd_file hash table
and closes GC-eligible entries whose underlying file resides on the
same filesystem and is a descendant of the export path. Because
nfsd_file entries do not carry an export reference, the ancestry
check uses is_subdir() on the file's dentry. False positives --
closing a cached handle that did not originate from the target
export -- are harmless; the handle is simply reopened on the next
access.
The handler calls nfsd_file_close_export() before revoking NFSv4
state, mirroring the order used by NFSD_CMD_UNLOCK_FILESYSTEM
(which cancels copies and releases NLM locks before revoking
state). Both calls run under nfsd_mutex.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Tested-by: Dai Ngo <dai.ngo@oracle.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
When a filesystem is exported to NFS clients, NFSv4 state
(opens, locks, delegations, layouts) holds references that
prevent the underlying filesystem from being unmounted.
NFSD_CMD_UNLOCK_FILESYSTEM addresses this at superblock
granularity, but administrators unexporting a single path on a
shared filesystem (e.g., one of several exports on the same device)
need finer control.
Add NFSD_CMD_UNLOCK_EXPORT, which revokes NFSv4 state acquired
through exports of a specific path. Matching is by path identity
(dentry + vfsmount) via the sc_export field on each nfs4_stid,
so multiple svc_export objects for the same path -- one per
auth_domain -- are handled correctly without requiring the caller
to name a specific client.
The command takes a single "path" attribute. Userspace (exportfs
-u) sends this after removing the last client for a given path,
enabling the underlying filesystem to be unmounted. When multiple
clients share an export path, individual unexports do not trigger
state revocation; only the final one does.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Tested-by: Dai Ngo <dai.ngo@oracle.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Add an sc_export field to struct nfs4_stid so that each stateid
records the export under which it was acquired. The export
reference is taken via exp_get() at stateid creation and released
via exp_put() in nfs4_put_stid().
Open stateids record the export from current_fh->fh_export.
Lock stateids and delegations inherit the export from their
parent open stateid. Layout stateids inherit from their
parent stateid. Directory delegations record the export from
cstate->current_fh.
A subsequent commit uses sc_export to scope state revocation to a
specific export, avoiding the need to walk inode dentry aliases at
revocation time.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Tested-by: Dai Ngo <dai.ngo@oracle.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Replace idr_for_each_entry_ul() with a while loop over
idr_get_next_ul() for consistency with find_one_export_stid(),
added in a subsequent commit.
No change in behavior.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Tested-by: Dai Ngo <dai.ngo@oracle.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Add NFSD_CMD_UNLOCK_FILESYSTEM as a dedicated netlink command for
revoking NFS state under a filesystem path, providing a netlink
equivalent of /proc/fs/nfsd/unlock_fs.
The command requires a "path" string attribute containing the
filesystem path whose state should be released. The handler
resolves the path to its superblock, then cancels async copies,
releases NLM locks, and revokes NFSv4 state on that superblock.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Tested-by: Dai Ngo <dai.ngo@oracle.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The existing write_unlock_ip procfs interface releases NLM file
locks held by a specific client IP address, but procfs provides
no structured way to extend that operation to other scopes such as
revoking NFSv4 state.
Add NFSD_CMD_UNLOCK_IP as a dedicated netlink command for
releasing NLM locks by client address. The command accepts a
binary sockaddr_in or sockaddr_in6 in its address attribute.
The handler validates the address family and length, then calls
nlmsvc_unlock_all_by_ip() to release matching NLM locks. Because
lockd is a single global instance, that call operates across
all network namespaces regardless of which namespace the caller
inhabits.
A separate netlink command for filesystem-scoped unlock is added in
a subsequent commit.
The nfsd_ctl_unlock_ip tracepoint is updated from string-based
address logging to __sockaddr, which stores the binary sockaddr
and formats it with %pISpc. This affects both the new netlink path
and the existing procfs write_unlock_ip path, giving consistent
structured output in both cases.
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Tested-by: Dai Ngo <dai.ngo@oracle.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
The per-stateid revocation logic in nfsd4_revoke_states() handles
four stateid types in a deeply nested switch. Extract two helpers:
revoke_ol_stid() performs admin-revocation of an open or lock
stateid with st_mutex already held: marks the stateid as
SC_STATUS_ADMIN_REVOKED, closes POSIX locks for lock stateids,
and releases file access.
revoke_one_stid() dispatches by sc_type, acquires st_mutex with
the appropriate lockdep class for open and lock stateids, and
handles delegation unhash and layout close inline.
No functional change. Preparation for adding export-scoped state
revocation which reuses revoke_one_stid().
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Tested-by: Dai Ngo <dai.ngo@oracle.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
nfsd4_drop_revoked_stid() has no SC_TYPE_LAYOUT case, so when a
client sends FREE_STATEID for an admin-revoked layout stid, the
default branch releases cl_lock and returns without unhashing or
releasing the stid. The stid remains in the IDR and on the
per-client list until the client is destroyed.
Remove the layout stid from the per-client list and call
nfs4_put_stid() to drop the creation reference. When the
refcount reaches zero, nfsd4_free_layout_stateid() handles the
remaining cleanup: cancelling the fence worker, removing from
the per-file list, and freeing the slab object.
Fixes: 1e33e1414b ("nfsd: allow layout state to be admin-revoked.")
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Tested-by: Dai Ngo <dai.ngo@oracle.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
iommu_dma_iova_link_swiotlb() processes a mapping that is unaligned in three
parts, the head, middle and trailer. If the middle is empty because there
are no aligned pages it will call down to iommu_map() with a 0 size
which the iommupt implementation will fail as illegal.
It then tries to do an error unwind and starts from the wrong spot
corrupting the mapping so the eventual destruction triggers a WARN_ON.
Check for 0 length and avoid mapping and use offset not 0 as the starting
point to unlink.
This is frequently triggered by using some kinds of thunderbolt NVMe
drives that trigger forced SWIOTLB for unaligned memory. NVMe seems to
pass in oddly aligned buffers for the passthrough commands from smartctl
that hit this condition.
Cc: stable@vger.kernel.org
Fixes: 433a76207d ("dma-mapping: Implement link/unlink ranges API")
Reported-by: Mark Lord <mlord@pobox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Samiullah Khawaja <skhawaja@google.com>
Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/0-v1-8536728bc89f+469-swiotlb_warn_jgg@nvidia.com
Vlad Poenaru says:
====================
bpf, lpm_trie: Allow sleepable BPF programs to use LPM tries
trie_lookup_elem() annotates its rcu_dereference_check() walks with only
rcu_read_lock_bh_held(), so a sleepable BPF program that touches an LPM
trie (e.g. a sleepable LSM hook calling bpf_map_lookup_elem()) trips a
"suspicious RCU usage" lockdep splat on debug kernels: it holds only
rcu_read_lock_trace(), which that annotation does not accept.
Patch 1 relaxes the rcu_dereference annotations in the trie walks so they
no longer trip lockdep from the Tasks Trace context, including the
trie_update_elem()/trie_delete_elem() writer walks (protected by
trie->lock). Patch 2 adds BPF_MAP_TYPE_LPM_TRIE to the verifier's
sleepable map whitelist so sleepable programs can reference an LPM trie
directly, not just as the inner map of a map-of-maps. LPM trie nodes are
reclaimed via bpf_mem_cache_free_rcu(), which chains a regular RCU grace
period into a Tasks Trace grace period before freeing -- the same
discipline BPF_MAP_TYPE_HASH relies on for sleepable access.
Changes since v1:
- Split into a 2-patch series.
- Patch 1 now also converts the trie_update_elem()/trie_delete_elem()
walks from rcu_dereference() to rcu_dereference_protected(*p, 1),
addressing review feedback that v1 only fixed the lookup path and left
the same splat on the writer paths.
- New patch 2 adds the verifier whitelist entry so the fix is actually
reachable for directly-referenced LPM tries.
- Retitled v1 ("Allow lookups from sleepable BPF programs").
v1: https://lore.kernel.org/all/20260529174233.2954240-1-vlad.wing@gmail.com/
====================
Link: https://patch.msgid.link/20260609135558.193287-1-vlad.wing@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
The previous change relaxed the rcu_dereference annotations in
lpm_trie.c so the trie walks no longer trip lockdep when reached from a
sleepable BPF program holding only rcu_read_lock_trace(). By itself
that only helps tries reached as the inner map of a map-of-maps, or
from the classic-RCU syscall path: a sleepable program that references
an LPM trie directly is still rejected at load time by
check_map_prog_compatibility(), whose sleepable whitelist omits
BPF_MAP_TYPE_LPM_TRIE:
Sleepable programs can only use array, hash, ringbuf and local storage maps
LPM trie nodes are allocated from a bpf_mem_alloc (trie->ma) and freed
with bpf_mem_cache_free_rcu(), which chains a regular RCU grace period
into a Tasks Trace grace period before the node -- and the value
embedded in it that trie_lookup_elem() returns to the program -- is
released. That is the same reclaim discipline BPF_MAP_TYPE_HASH relies
on for sleepable access, so a value handed to a sleepable reader cannot
be freed while the program is still running under rcu_read_lock_trace().
The writer paths take trie->lock across the walk and never relied on the
RCU read-side lock to keep nodes alive.
Add BPF_MAP_TYPE_LPM_TRIE to the sleepable map whitelist so these
programs can use LPM tries directly.
Signed-off-by: Vlad Poenaru <vlad.wing@gmail.com>
Reviewed-by: Emil Tsalapatis <emil@etsalapatis.com>
Link: https://lore.kernel.org/r/20260609135558.193287-3-vlad.wing@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
trie_lookup_elem() annotates its rcu_dereference_check() walks with
only rcu_read_lock_bh_held(). Because rcu_dereference_check(p, c)
resolves to "c || rcu_read_lock_held()", this passes for XDP/NAPI and
classic RCU readers but fails for sleepable BPF programs, which enter
via __bpf_prog_enter_sleepable() and hold only rcu_read_lock_trace().
trie_update_elem() and trie_delete_elem() have the same problem in a
different form: they walk the trie with plain rcu_dereference(), which
asserts rcu_read_lock_held() unconditionally. Both are reachable from
sleepable BPF programs via the bpf_map_update_elem / bpf_map_delete_elem
helpers, and from the syscall path under classic rcu_read_lock(). In
the writer paths the trie is actually protected by trie->lock (an
rqspinlock taken across the walk); we never relied on the RCU read-side
lock to keep nodes alive there.
A sleepable LSM hook that ends up touching an LPM trie therefore
triggers lockdep on debug kernels:
=============================
WARNING: suspicious RCU usage
7.1.0-... Tainted: G E
-----------------------------
kernel/bpf/lpm_trie.c:249 suspicious rcu_dereference_check() usage!
1 lock held by net_tests/540:
#0: (rcu_tasks_trace_srcu_struct){....}-{0:0},
at: __bpf_prog_enter_sleepable+0x26/0x280
Call Trace:
dump_stack_lvl
lockdep_rcu_suspicious
trie_lookup_elem
bpf_prog_..._enforce_security_socket_connect
bpf_trampoline_...
security_socket_connect
__sys_connect
do_syscall_64
This is lockdep-only -- no UAF, since Tasks Trace RCU does serialize
against the trie's reclaim path -- but it spams the console once per
distinct callsite on every debug kernel running a sleepable BPF LSM
that touches an LPM trie, which is increasingly common.
For the lookup path, switch the rcu_dereference_check() annotation
from rcu_read_lock_bh_held() to bpf_rcu_lock_held(), which accepts all
three contexts (classic, BH, Tasks Trace). Other map types already
follow this convention.
For trie_update_elem() and trie_delete_elem(), annotate the walks as
rcu_dereference_protected(*p, 1) -- matching trie_free() in the same
file -- since trie->lock is held across the walk. rqspinlock has no
lockdep_map, so the predicate degenerates to '1' rather than
lockdep_is_held(&trie->lock); the protection is real but not
machine-verifiable. trie_get_next_key() also uses bare
rcu_dereference() but is reachable only from the BPF syscall, which
holds classic rcu_read_lock() before dispatching, so it is left
untouched.
Fixes: 694cea395f ("bpf: Allow RCU-protected lookups to happen from bh context")
Cc: stable@vger.kernel.org
Signed-off-by: Vlad Poenaru <vlad.wing@gmail.com>
Reviewed-by: Emil Tsalapatis <emil@etsalapatis.com>
Link: https://lore.kernel.org/r/20260609135558.193287-2-vlad.wing@gmail.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Implement the query port speed callback to report the port effective
bandwidth directly in 100 Mb/s granularity.
Link: https://patch.msgid.link/r/20260608083927.4116-1-tomsela@amazon.com
Reviewed-by: Michael Margolin <mrgolin@amazon.com>
Reviewed-by: Yonatan Nachum <ynachum@amazon.com>
Signed-off-by: Tom Sela <tomsela@amazon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Add support for reporting 800 Gbps as 8X NDR and 1600 Gbps as 8X XDR
link speeds.
Link: https://patch.msgid.link/r/20260608083736.48454-1-tomsela@amazon.com
Reviewed-by: Michael Margolin <mrgolin@amazon.com>
Reviewed-by: Yonatan Nachum <ynachum@amazon.com>
Signed-off-by: Tom Sela <tomsela@amazon.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
A comment in fs/nfs/dir.c incorrectly refers to CONFIG_NFSV4 instead of
CONFIG_NFS_V4. Correct it.
Discovered while searching for CONFIG_* symbols referenced in code but
not defined in any Kconfig file.
Signed-off-by: Ethan Nelson-Moore <enelsonmoore@gmail.com>
Signed-off-by: Anna Schumaker <anna.schumaker@hammerspace.com>
Don't just overwrite the original pointer passed to krealloc()
with its return value without checking latter:
MEM = krealloc(MEM, SZ, GFP);
If krealloc() returns NULL, that erases the pointer
to the still allocated memory, hence leaks this memory.
Instead, use a temporary variable, check it's not NULL
and only then assign it to the original pointer:
TMP = krealloc(MEM, SZ, GFP);
if (!TMP) return;
MEM = TMP;
While on it, use krealloc_array().
Fixes: 6d45c81d22 ("drm/vc4: Add support for branching in shader validation.")
Signed-off-by: Alexander A. Klimov <grandmaster@al2klimov.de>
Signed-off-by: Maíra Canal <mcanal@igalia.com>
Link: https://patch.msgid.link/20260606123817.37222-1-grandmaster@al2klimov.de
Use a flexible array member to remove one allocation. Simplifies code
slightly.
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
The driver carries custom system suspend/resume handling that manually
tracks a suspended state and conditionally calls
geni_i2c_runtime_suspend()
from the noirq suspend path, then adjusts runtime PM state by hand. This
duplicates PM core behavior and adds unnecessary complexity.
Drop the manual state tracking and switch to pm_runtime_force_suspend()
and pm_runtime_force_resume() for system sleep. These helpers already
perform the required checks, call the runtime PM callbacks when needed,
and keep runtime PM state transitions consistent.
Reviewed-by: Mukesh Kumar Savaliya <mukesh.savaliya@oss.qualcomm.com>
Signed-off-by: Praveen Talari <praveen.talari@oss.qualcomm.com>
Signed-off-by: Andi Shyti <andi.shyti@kernel.org>
Link: https://lore.kernel.org/r/20260520-use_pm_runtime_apis-v1-1-6a5238fc6cb6@oss.qualcomm.com
The libarena parallel spmc selftest is nondeterministic by design.
As a result it depends up to a point on the relative timing between the
producer and consumer threads. This introduces the possibility for two
kinds of spurious failures that this patch addresses.
1) Spurious timeouts. The test proceeds in phases, and threads use a
common counter as a barrier to avoid proceeding to the next phase
until all threads are ready to do so. If a thread takes too long to
reach the barrier, the already waiting threads may time out.
Increase the current timeout. The timeout's value is a balance
between the maximum amount of time spent on the test and the
possibility of spurious failures. Right now the timeout is too short.
Err on the side of caution and significantly increase it to avoid
spurious failures.
2) Spurious resize failures. Some selftests require the spmc queue to
resize itself. This in turn requires for the producer side to be
materially faster than the consumer side so that the queue gets full
enough for a resize. However, in the benchmark the spmc queue's producer
is outnumbered 3:1. To offset it we add busy waits for consume
queues. However, we still see occasional failures due to the queue
never resizing.
Minimize the possibility for this in two ways: First, remove one of
the consumers. The 2 consumers still exercise the "race between
consumers" scenario. Second, increase the busy wait duration to
decrease the rate by which the consumers act on the queue.
While at it, also replace a stray invalid error value "153" with EINVAL.
Fixes: 42998f8192 ("selftests/bpf: libarena: parallel test harness and spmc parallel selftest")
Reported-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Emil Tsalapatis <emil@etsalapatis.com>
Link: https://lore.kernel.org/r/20260609063630.10245-1-emil@etsalapatis.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
This issue was pointed out by Sashiko.
vbbck is fed internally from vio18. For the MT6359, the default supply
name was incorrectly set as "VIO18", instead of the supply's default
"VIO18". In practice this still works, but it causes the regulator
description copy and replace to always happen. For the MT6359P the
name is correct.
Fix the supply name for MT6359 so that both instances are the same and
correct. Also copy the comment about the internal supply from the MT6359
list to the MT6359P list.
Fixes: 10be8fc1d5 ("regulator: mt6359: Add regulator supply names")
Signed-off-by: Chen-Yu Tsai <wenst@chromium.org>
Reviewed-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>
Link: https://patch.msgid.link/20260609083630.1600070-1-wenst@chromium.org
Signed-off-by: Mark Brown <broonie@kernel.org>
Sashiko noticed mlx4 was using whatever random access flags were provided
when IB_MR_REREG_ACCESS is not used. Since IB_MR_REREG_TRANS needs
access_flags it used the random ones which means it doesn't work sensibly
if userspace provides only IB_MR_REREG_TRANS.
Keep track of the current access_flag of the MR and use it if the user
does not specify one.
Also fixup a little confusion around mmr.access, it is the HW access flags
so the convert_access() was missing. But nothing reads this by the time
rereg_mr can happen.
Fixes: 9376932d0c ("IB/mlx4_ib: Add support for user MR re-registration")
Link: https://patch.msgid.link/r/0-v1-29ca7a402625+ddd6-mlx4_rereg_flags_jgg@nvidia.com
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Only accept an ACPI machine table entry when machine_check is absent
or returns true, matching other AMD SoundWire machine select paths.
Signed-off-by: Vijendar Mukunda <Vijendar.Mukunda@amd.com>
Link: https://patch.msgid.link/20260609144146.3311301-1-Vijendar.Mukunda@amd.com
Signed-off-by: Mark Brown <broonie@kernel.org>
A couple of symbol declarations shadow earlier variables in the region
attach paths. Shadowing makes it harder to tell which object is being
referenced and can obscure future bugs.
Reuse the existing 'cxld' variable in cxl_port_attach_region() and
rename the endpoint decoder iterator in cxl_region_attach() to avoid
shadowing the function parameter.
No functional change.
Found with sparse.
Signed-off-by: Alison Schofield <alison.schofield@intel.com>
Reviewed-by: Li Ming <ming.li@zohomail.com>
Reviewed-by: Dave Jiang <dave.jiang@intel.com>
Link: https://patch.msgid.link/20260605040504.865728-1-alison.schofield@intel.com
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Only accept an ACPI machine table entry when machine_check is absent
or returns true, matching other AMD SoundWire machine select paths.
Signed-off-by: Vijendar Mukunda <Vijendar.Mukunda@amd.com>
Link: https://patch.msgid.link/20260609143230.3310356-1-Vijendar.Mukunda@amd.com
Signed-off-by: Mark Brown <broonie@kernel.org>
sma1307_check_fault_worker() stores dynamically allocated uevent strings in
envp[0]. Several fault conditions are checked in sequence, so a later fault
can overwrite envp[0] before the final kfree() and leak the previous
allocation.
The same flow can leave an OT1 volume entry in envp[1] while envp[0]
has been overwritten by a later non-OT1 fault, causing an inconsistent
uevent payload.
Use static STATUS strings and a stack buffer for the optional VOLUME entry.
This removes the allocations from the worker and keeps VOLUME tied only
to the OT1 events that produce it.
Fixes: 576c57e6b4 ("ASoC: sma1307: Add driver for Iron Device SMA1307")
Signed-off-by: Cássio Gabriel <cassiogabrielcontato@gmail.com>
Link: https://patch.msgid.link/20260609-asoc-sma1307-uevent-leak-v1-1-cd7f5b062ab7@gmail.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Peter Ujfalusi <peter.ujfalusi@linux.intel.com> says:
This series hardens SOF kcontrol data paths for both IPC3 and IPC4 by
fixing size-handling bugs in put/get/update flows and tightening bounds
checks around firmware/user-provided payload lengths.
The changes include:
Fix TOCTOU-style size misuse in IPC3/IPC4 bytes put paths by validating and
using the incoming payload size.
Add notification/update payload size validation before parsing control data.
Use overflow-checked arithmetic when computing expected IPC3 control sizes.
Ensure update/copy bounds are validated against actual allocation limits.
Fix IPC3 bytes_ext bounds checks to account for struct header offset, closing
a heap overflow/over-read issue from unprivileged userspace TLV access.
Overall, the series makes control payload processing robust against malformed or
inconsistent sizes and prevents out-of-bounds accesses.
Link: https://patch.msgid.link/20260609083458.31193-1-peter.ujfalusi@linux.intel.com
The ipc_control_data buffer is allocated as kzalloc(max_size), where
max_size covers the entire struct sof_ipc_ctrl_data including its
flexible array payload. However, the bounds checks in bytes_ext_put
and _bytes_ext_get compared user data lengths against max_size
directly, ignoring that cdata->data sits at an offset of
sizeof(struct sof_ipc_ctrl_data) bytes into the allocation.
This allowed writing up to sizeof(struct sof_ipc_ctrl_data) bytes past
the end of the heap buffer from unprivileged userspace via the ALSA TLV
kcontrol interface, and similarly allowed over-reading adjacent heap
data on the get path.
Fix all bounds checks to subtract sizeof(*cdata) from max_size so they
reflect the actual space available at the cdata->data offset. Also fix
the error-path restore in bytes_ext_put which wrote to cdata->data
instead of cdata, causing the same overflow.
Fixes: 67ec2a0916 ("ASoC: SOF: Add bytes_ext control IPC ops for IPC3")
Cc: stable@vger.kernel.org
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>
Reviewed-by: Liam Girdwood <liam.r.girdwood@intel.com>
Reviewed-by: Bard Liao <yung-chuan.liao@linux.intel.com>
Link: https://patch.msgid.link/20260609083458.31193-7-peter.ujfalusi@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
In sof_ipc3_bytes_put(), the size used for the memcpy is derived from
the old data->size already in the buffer, not the incoming new data's
size field. If the new data has a different size, the copy length is
wrong: it may truncate valid data or copy stale bytes.
Similarly, sof_ipc3_bytes_get() checks data->size against max_size
without accounting for the sizeof(struct sof_ipc_ctrl_data) offset
of the flex array within the allocation.
Fix bytes_put to validate and use the incoming data's sof_abi_hdr.size
from ucontrol before copying. Fix bytes_get to subtract sizeof(*cdata)
from the bounds check to match the actual available space.
Fixes: 544ac8858f ("ASoC: SOF: Add bytes_get/put control IPC ops for IPC3")
Cc: stable@vger.kernel.org
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>
Reviewed-by: Liam Girdwood <liam.r.girdwood@intel.com>
Reviewed-by: Bard Liao <yung-chuan.liao@linux.intel.com>
Link: https://patch.msgid.link/20260609083458.31193-6-peter.ujfalusi@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
In snd_sof_update_control(), firmware-provided cdata->num_elems is
checked against local_cdata->data->size but never against the actual
allocation size. If local_cdata->data->size was previously set to an
inconsistent value, the memcpy could write past the allocated buffer.
Add a bounds check to ensure num_elems fits within the available space
in the ipc_control_data allocation before copying.
Fixes: 10f461d79c ("ASoC: SOF: Add IPC3 topology control ops")
Cc: stable@vger.kernel.org
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>
Reviewed-by: Liam Girdwood <liam.r.girdwood@intel.com>
Reviewed-by: Bard Liao <yung-chuan.liao@linux.intel.com>
Link: https://patch.msgid.link/20260609083458.31193-5-peter.ujfalusi@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
In sof_ipc3_control_update(), the expected_size calculation uses
firmware-provided cdata->num_elems in arithmetic that could overflow
on 32-bit platforms, wrapping to a small value. This would allow the
cdata->rhdr.hdr.size comparison to pass with mismatched sizes,
potentially leading to out-of-bounds access in snd_sof_update_control.
Use check_mul_overflow() and check_add_overflow() to detect and reject
overflowed size calculations.
Fixes: 10f461d79c ("ASoC: SOF: Add IPC3 topology control ops")
Cc: stable@vger.kernel.org
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>
Reviewed-by: Liam Girdwood <liam.r.girdwood@intel.com>
Reviewed-by: Bard Liao <yung-chuan.liao@linux.intel.com>
Link: https://patch.msgid.link/20260609083458.31193-4-peter.ujfalusi@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Validate MODULE_NOTIFICATION payload length before reading
bytes/channel data in control update handling.
Fixes: 2a28b5240f ("ASoC: SOF: ipc4-control: Add support for generic bytes control")
Cc: stable@vger.kernel.org
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>
Reviewed-by: Liam Girdwood <liam.r.girdwood@intel.com>
Reviewed-by: Bard Liao <yung-chuan.liao@linux.intel.com>
Link: https://patch.msgid.link/20260609083458.31193-3-peter.ujfalusi@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
In sof_ipc4_bytes_put(), the copy size is derived from the old
data->size in the buffer rather than the incoming new data's size
field from ucontrol. If the new data has a different size, the copy
uses the wrong length: it may truncate valid data or copy stale bytes.
Fix by validating and using the incoming data's sof_abi_hdr.size from
ucontrol before copying.
Fixes: a062c8899f ("ASoC: SOF: ipc4-control: Add support for bytes control get and put")
Cc: stable@vger.kernel.org
Signed-off-by: Peter Ujfalusi <peter.ujfalusi@linux.intel.com>
Reviewed-by: Liam Girdwood <liam.r.girdwood@intel.com>
Reviewed-by: Bard Liao <yung-chuan.liao@linux.intel.com>
Link: https://patch.msgid.link/20260609083458.31193-2-peter.ujfalusi@linux.intel.com
Signed-off-by: Mark Brown <broonie@kernel.org>
Revisit and reorganize the locking and lock coverage of the
ap->lock spinlock as used in the two sysfs functions
se_bind_store() and se_associate_store().
A kernel run reported a possible deadlock situation, caused by
holding the spinlock (ap->lock) while triggering a uevent.
The fix rearranges the code protected by the spinlock by excluding
the uevent invocation, which does not require protection.
Additionally, the start of the protected region is moved earlier
to cover more lines, ensuring a consistent view of the AP queue
state between reading and updating its struct fields.
=====================================================
WARNING: SOFTIRQ-safe -> SOFTIRQ-unsafe lock order detected
7.1.0-20260601.rc6.git12.516b5dbd4d4a.300.fc44.s390x+debug #1 Not tainted
-----------------------------------------------------
setupseguest.sh/11034 [HC0[0]:SC0[2]:HE1:SE0] is trying to acquire:
000001c991f498e8 (fs_reclaim){+.+.}-{0:0}, at: __kmalloc_cache_noprof+0x5a/0x6d0
and this task is already holding:
000000c4a1a12378 (&aq->lock){+.-.}-{2:2}, at: se_bind_store+0x96/0x3a0
which would create a new lock dependency:
(&aq->lock){+.-.}-{2:2} -> (fs_reclaim){+.+.}-{0:0}
but this new dependency connects a SOFTIRQ-irq-safe lock:
(&aq->lock){+.-.}-{2:2}
... which became SOFTIRQ-irq-safe at:
__lock_acquire+0x5ae/0x15a0
lock_acquire+0x14c/0x400
_raw_spin_lock_bh+0x58/0xb0
ap_tasklet_fn+0x72/0xd0
tasklet_action_common+0x174/0x1b0
handle_softirqs+0x180/0x5c0
irq_exit_rcu+0x196/0x200
do_ext_irq+0x12a/0x4d0
ext_int_handler+0xc6/0xf0
folio_zero_user+0x1c6/0x240
folio_zero_user+0x182/0x240
vma_alloc_anon_folio_pmd+0xa0/0x1d0
__do_huge_pmd_anonymous_page+0x3a/0x200
__handle_mm_fault+0x56c/0x590
handle_mm_fault+0xa2/0x370
do_exception+0x292/0x590
__do_pgm_check+0x136/0x3e0
pgm_check_handler+0x114/0x160
to a SOFTIRQ-irq-unsafe lock:
(fs_reclaim){+.+.}-{0:0}
... which became SOFTIRQ-irq-unsafe at:
...
__lock_acquire+0x5ae/0x15a0
lock_acquire+0x14c/0x400
__fs_reclaim_acquire+0x44/0x50
fs_reclaim_acquire+0xbe/0x100
fs_reclaim_correct_nesting+0x20/0x70
dotest+0x5e/0x148
locking_selftest+0x2854/0x2a88
start_kernel+0x3b2/0x4f0
startup_continue+0x2e/0x40
other info that might help us debug this:
Possible interrupt unsafe locking scenario:
CPU0 CPU1
---- ----
lock(fs_reclaim);
local_irq_disable();
lock(&aq->lock);
lock(fs_reclaim);
<Interrupt>
lock(&aq->lock);
*** DEADLOCK ***
4 locks held by setupseguest.sh/11034:
#0: 000000c485d01440 (sb_writers#4){.+.+}-{0:0}, at: vfs_write+0x2fc/0x380
#1: 000000c4d2283288 (&of->mutex#2){+.+.}-{3:3}, at: kernfs_fop_write_iter+0x12a0x270
#2: 000000c4a1830e48 (kn->active#172){.+.+}-{0:0}, at: kernfs_fop_write_iter+0x1e/0x270
#3: 000000c4a1a12378 (&aq->lock){+.-.}-{2:2}, at: se_bind_store+0x96/0x3a0
the dependencies between SOFTIRQ-irq-safe lock and the holding lock:
-> (&aq->lock){+.-.}-{2:2} {
HARDIRQ-ON-W at:
__lock_acquire+0x5ae/0x15a0
lock_acquire+0x14c/0x400
_raw_spin_lock_bh+0x58/0xb0
ap_queue_init_state+0x2e/0x50
ap_scan_domains+0x5d6/0x620
ap_scan_adapter+0x4c0/0x810
ap_scan_bus+0x70/0x350
ap_scan_bus_wq_callback+0x56/0x80
process_one_work+0x2ba/0x820
worker_thread+0x21a/0x400
kthread+0x164/0x190
__ret_from_fork+0x4c/0x340
ret_from_fork+0xa/0x30
IN-SOFTIRQ-W at:
__lock_acquire+0x5ae/0x15a0
lock_acquire+0x14c/0x400
_raw_spin_lock_bh+0x58/0xb0
ap_tasklet_fn+0x72/0xd0
tasklet_action_common+0x174/0x1b0
handle_softirqs+0x180/0x5c0
irq_exit_rcu+0x196/0x200
do_ext_irq+0x12a/0x4d0
ext_int_handler+0xc6/0xf0
folio_zero_user+0x1c6/0x240
folio_zero_user+0x182/0x240
vma_alloc_anon_folio_pmd+0xa0/0x1d0
__do_huge_pmd_anonymous_page+0x3a/0x200
__handle_mm_fault+0x56c/0x590
handle_mm_fault+0xa2/0x370
do_exception+0x292/0x590
__do_pgm_check+0x136/0x3e0
pgm_check_handler+0x114/0x160
INITIAL USE at:
__lock_acquire+0x5ae/0x15a0
lock_acquire+0x14c/0x400
_raw_spin_lock_bh+0x58/0xb0
ap_queue_init_state+0x2e/0x50
ap_scan_domains+0x5d6/0x620
ap_scan_adapter+0x4c0/0x810
ap_scan_bus+0x70/0x350
ap_scan_bus_wq_callback+0x56/0x80
process_one_work+0x2ba/0x820
worker_thread+0x21a/0x400
kthread+0x164/0x190
__ret_from_fork+0x4c/0x340
ret_from_fork+0xa/0x30
}
... key at: [<000001c9936e8aa0>] __key.7+0x0/0x10
the dependencies between the lock to be acquired
and SOFTIRQ-irq-unsafe lock:
-> (fs_reclaim){+.+.}-{0:0} {
HARDIRQ-ON-W at:
__lock_acquire+0x5ae/0x15a0
lock_acquire+0x14c/0x400
__fs_reclaim_acquire+0x44/0x50
fs_reclaim_acquire+0xbe/0x100
fs_reclaim_correct_nesting+0x20/0x70
dotest+0x5e/0x148
locking_selftest+0x2854/0x2a88
start_kernel+0x3b2/0x4f0
startup_continue+0x2e/0x40
SOFTIRQ-ON-W at:
__lock_acquire+0x5ae/0x15a0
lock_acquire+0x14c/0x400
__fs_reclaim_acquire+0x44/0x50
fs_reclaim_acquire+0xbe/0x100
fs_reclaim_correct_nesting+0x20/0x70
dotest+0x5e/0x148
locking_selftest+0x2854/0x2a88
start_kernel+0x3b2/0x4f0
startup_continue+0x2e/0x40
INITIAL USE at:
__lock_acquire+0x5ae/0x15a0
lock_acquire+0x14c/0x400
__fs_reclaim_acquire+0x44/0x50
fs_reclaim_acquire+0xbe/0x100
fs_reclaim_correct_nesting+0x20/0x70
dotest+0x5e/0x148
locking_selftest+0x2854/0x2a88
start_kernel+0x3b2/0x4f0
startup_continue+0x2e/0x40
}
... key at: [<000001c991f498e8>] __fs_reclaim_map+0x0/0x30
... acquired at:
check_prev_add+0x178/0xf40
__lock_acquire+0x12aa/0x15a0
lock_acquire+0x14c/0x400
__fs_reclaim_acquire+0x44/0x50
fs_reclaim_acquire+0xbe/0x100
__kmalloc_cache_noprof+0x5a/0x6d0
kobject_uevent_env+0xd4/0x420
ap_send_se_bind_uevent+0x48/0x70
se_bind_store+0x146/0x3a0
kernfs_fop_write_iter+0x18c/0x270
vfs_write+0x23c/0x380
ksys_write+0x88/0x120
__do_syscall+0x170/0x750
system_call+0x72/0x90
stack backtrace:
CPU: 6 UID: 0 PID: 11034 Comm: setupseguest.sh Not tainted 7.1.0-20260601.rc6.git2.516b5dbd4d4a.300.fc44.s390x+debug #1 PREEMPT
Hardware name: IBM 9175 ME1 701 (KVM/Linux)
Call Trace:
[<000001c98ffa0a7e>] dump_stack_lvl+0xae/0x108
[<000001c9900a6d7a>] print_bad_irq_dependency+0x47a/0x480
[<000001c9900a7184>] check_irq_usage+0x404/0x4c0
[<000001c9900a73b8>] check_prev_add+0x178/0xf40
[<000001c9900aaf1a>] __lock_acquire+0x12aa/0x15a0
[<000001c9900ab35c>] lock_acquire+0x14c/0x400
[<000001c9903be454>] __fs_reclaim_acquire+0x44/0x50
[<000001c9903be51e>] fs_reclaim_acquire+0xbe/0x100
[<000001c9903cf4ca>] __kmalloc_cache_noprof+0x5a/0x6d0
[<000001c9910ca9d4>] kobject_uevent_env+0xd4/0x420
[<000001c990d84098>] ap_send_se_bind_uevent+0x48/0x70
[<000001c990d87416>] se_bind_store+0x146/0x3a0
[<000001c99057da7c>] kernfs_fop_write_iter+0x18c/0x270
[<000001c99047712c>] vfs_write+0x23c/0x380
[<000001c990477438>] ksys_write+0x88/0x120
[<000001c9910f64e0>] __do_syscall+0x170/0x750
[<000001c99110a412>] system_call+0x72/0x90
INFO: lockdep is turned off.
Fixes: 4179c39842 ("s390/ap: Implement SE bind and associate uevents")
Reported-by: Ingo Franzki <ifranzki@linux.ibm.com>
Suggested-by: Finn Callies <fcallies@linux.ibm.com>
Reviewed-by: Finn Callies <fcallies@linux.ibm.com>
Signed-off-by: Harald Freudenberger <freude@linux.ibm.com>
Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
Firmware will set dsp_ack to 1 when firmware sends response for the IPC
command issued by host. Similarly dsp_msg flag will be updated to 1.
During ACP D0 entry, the value read from the sof_dsp_ack_write scratch
flag can be uninitialized. A non-zero garbage value is treated as a
pending DSP IPC ack before SOF_FW_BOOT_COMPLETE, causing a spurious
"IPC reply before FW_BOOT_COMPLETE" log.
Fix the condition checks for ipc flags.
Fixes: 738a2b5e2c ("ASoC: SOF: amd: Add IPC support for ACP IP block")
Link: https://github.com/thesofproject/linux/pull/5642
Signed-off-by: Vijendar Mukunda <Vijendar.Mukunda@amd.com>
Tested-by: Umang Jain <uajain@igalia.com>
Link: https://patch.msgid.link/20260609160938.3717513-1-Vijendar.Mukunda@amd.com
Signed-off-by: Mark Brown <broonie@kernel.org>
In the NVMe multipath code, if nvme_add_ns_head_cdev() fails during
nvme_mpath_set_live(), the error is ignored. However, during teardown,
nvme_remove_head() unconditionally calls nvme_cdev_del(). This teardown
asymmetry leads to a kernel panic if the character device was never
successfully initialized.
BUG: kernel NULL pointer dereference, address: 00000000000000d0
device_del+0x39/0x3c0
cdev_device_del+0x15/0x50
nvme_cdev_del+0xe/0x20 [nvme_core]
nvme_mpath_shutdown_disk+0x38/0x60 [nvme_core]
nvme_ns_remove+0x177/0x1f0 [nvme_core]
nvme_remove_namespaces+0xdc/0x130 [nvme_core]
nvme_do_delete_ctrl+0x71/0xd0 [nvme_core]
Additionally, a memory leak exists in the nvme_cdev_add() failure path.
Previously, dev_set_name() was called before ida_alloc(). If ida_alloc()
subsequently failed, device_initialize() was never called, meaning
put_device() could not be used to clean up the kobject, leaking the
memory allocated by dev_set_name().
* Introduces the NVME_NSHEAD_CDEV_LIVE and NVME_NS_CDEV_LIVE bits to track
the successful creation of the character devices. Teardown routines now
check these bits before attempting deletion.
* Refactor nvme_cdev_add() to accept the formatted device name as a
parameter, moving dev_set_name() after the IDA allocation and
immediately before device_initialize(). This ensures any internally
allocated strings are safely cleaned up by put_device() upon failure.
Signed-off-by: Maurizio Lombardi <mlombard@redhat.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Jakub Kicinski says:
====================
net: ethtool: let ops locked drivers run without rtnl_lock
With the ethtool_get_link_ksettings() situation hopefully ironed out
the previous series (commit 6a5d837f0c) let's return to the main
part of the series.
We have been slowly moving towards removing the rtnl_lock dependency
in driver ops since the concept of "ops-locked" drivers have been
introduced last year. Since last year will take the netdev instance
lock before invoking any ndo or ethtool op of "ops-locked" drivers.
We dipped our toes into rtnl_lock-less ops with the queue binding API.
Queue stats, NAPI, and other netdev-netlink objects are also queried
without holding rtnl_lock already. It's time to take the next logical
step and lift the requirement from ethtool ops.
The direct motivation for this patchset is that ethtool ops often
involve communicating with device FW, and may take a long time
to complete. Aggressive polling of device state on machines
with 10+ NICs have been shown to significantly increase rtnl_lock
pressure.
There's a handful of areas which still need rtnl_lock (see below).
I decided to convert everything to rtnl_lock-less by default, and
add a set of flags which let the drivers request rtnl_lock to still
be taken. I don't love this, but I'm worried that opt-in would be
even more confusing.
Known issues / exclusions:
- qdiscs - qdisc configuration currently assumes rtnl_lock, this
is mostly impacting set_channels callback. qdisc config is probably
the easiest one of the exclusions to tackle, it's fairly self-contained.
- features - even tho feature changes are (correctly) plumbed to
the driver thru ndos they are part of ethtool uAPI. ethtool itself
calls netdev_features_change() if it has spotted device feature change
before vs after to the callback. Some drivers also call
netdev_features_change() directly in response to various changes,
e.g. setting priv flags.
Since features have to propagate to upper and lower devices anything
that touches features is quite hard to move from under rtnl_lock.
- phylink - phylink and SFP depend on rtnl_lock today, I suspect
that this is purely for historic reasons. I started poking at
it and don't really see a need for a global lock. But accessing
the netdev instance lock from the SFP entry points will require
some attention from the phylink folks.
- phydev - similar to phylink, looks quite doable. But no ops-locked
driver currently has a phydev (fbnic only uses phylink) so phydev
related paths retain a ASSERT_RTNL() for now.
Tested on mlx5, bnxt and fbnic.
====================
Link: https://patch.msgid.link/20260605002912.3456868-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Catch up various bits of documentation after the locking changes.
Reviewed-by: Nicolai Buchwitz <nb@tipi-net.de>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Acked-by: Stanislav Fomichev <sdf@fomichev.me>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://patch.msgid.link/20260605002912.3456868-13-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
