linux-2.6-microblaze.git
8 months agosvcrdma: De-duplicate completion ID initialization helpers
Chuck Lever [Mon, 11 Dec 2023 15:24:08 +0000 (10:24 -0500)]
svcrdma: De-duplicate completion ID initialization helpers

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Move the svc_rdma_cc_init() call
Chuck Lever [Mon, 4 Dec 2023 14:58:33 +0000 (09:58 -0500)]
svcrdma: Move the svc_rdma_cc_init() call

Now that the chunk_ctxt for Reads is no longer dynamically allocated
it can be initialized once for the life of the object that contains
it (struct svc_rdma_recv_ctxt).

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Remove struct svc_rdma_read_info
Chuck Lever [Mon, 4 Dec 2023 14:58:26 +0000 (09:58 -0500)]
svcrdma: Remove struct svc_rdma_read_info

The remaining fields of struct svc_rdma_read_info are no longer
referenced.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Update the synopsis of svc_rdma_read_special()
Chuck Lever [Mon, 4 Dec 2023 14:58:20 +0000 (09:58 -0500)]
svcrdma: Update the synopsis of svc_rdma_read_special()

Since the RDMA Read I/O state is now contained in the recv_ctxt,
svc_rdma_read_special() can use that recv_ctxt to derive the
read_info rather than the other way around. This removes another
usage of the ri_readctxt field, enabling its removal in a
subsequent patch.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Update the synopsis of svc_rdma_read_call_chunk()
Chuck Lever [Mon, 4 Dec 2023 14:58:13 +0000 (09:58 -0500)]
svcrdma: Update the synopsis of svc_rdma_read_call_chunk()

Since the RDMA Read I/O state is now contained in the recv_ctxt,
svc_rdma_read_call_chunk() can use that recv_ctxt to derive the
read_info rather than the other way around. This removes another
usage of the ri_readctxt field, enabling its removal in a
subsequent patch.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Update synopsis of svc_rdma_read_multiple_chunks()
Chuck Lever [Mon, 4 Dec 2023 14:58:07 +0000 (09:58 -0500)]
svcrdma: Update synopsis of svc_rdma_read_multiple_chunks()

Since the RDMA Read I/O state is now contained in the recv_ctxt,
svc_rdma_read_multiple_chunks() can use that recv_ctxt to derive the
read_info rather than the other way around. This removes another
usage of the ri_readctxt field, enabling its removal in a
subsequent patch.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Update synopsis of svc_rdma_copy_inline_range()
Chuck Lever [Mon, 4 Dec 2023 14:58:01 +0000 (09:58 -0500)]
svcrdma: Update synopsis of svc_rdma_copy_inline_range()

Since the RDMA Read I/O state is now contained in the recv_ctxt,
svc_rdma_copy_inline_range() can use that recv_ctxt to derive the
read_info rather than the other way around. This removes another
usage of the ri_readctxt field, enabling its removal in a
subsequent patch.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Update the synopsis of svc_rdma_read_data_item()
Chuck Lever [Mon, 4 Dec 2023 14:57:54 +0000 (09:57 -0500)]
svcrdma: Update the synopsis of svc_rdma_read_data_item()

Since the RDMA Read I/O state is now contained in the recv_ctxt,
svc_rdma_build_read_data_item() can use that recv_ctxt to derive
that information rather than the other way around. This removes
another usage of the ri_readctxt field, enabling its removal in a
subsequent patch.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Update synopsis of svc_rdma_read_chunk_range()
Chuck Lever [Mon, 4 Dec 2023 14:57:48 +0000 (09:57 -0500)]
svcrdma: Update synopsis of svc_rdma_read_chunk_range()

Since the RDMA Read I/O state is now contained in the recv_ctxt,
svc_rdma_build_read_chunk_range() can use that recv_ctxt to derive
that information rather than the other way around. This removes
another usage of the ri_readctxt field, enabling its removal in a
subsequent patch.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Update synopsis of svc_rdma_build_read_chunk()
Chuck Lever [Mon, 4 Dec 2023 14:57:41 +0000 (09:57 -0500)]
svcrdma: Update synopsis of svc_rdma_build_read_chunk()

Since the RDMA Read I/O state is now contained in the recv_ctxt,
svc_rdma_build_read_chunk() can use that recv_ctxt to derive that
information rather than the other way around. This removes another
usage of the ri_readctxt field, enabling its removal in a
subsequent patch.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Update synopsis of svc_rdma_build_read_segment()
Chuck Lever [Mon, 4 Dec 2023 14:57:35 +0000 (09:57 -0500)]
svcrdma: Update synopsis of svc_rdma_build_read_segment()

Since the RDMA Read I/O state is now contained in the recv_ctxt,
svc_rdma_build_read_segment() can use the recv_ctxt to derive that
information rather than the other way around. This removes one usage
of the ri_readctxt field, enabling its removal in a subsequent
patch.

At the same time, the use of ri_rqst can similarly be replaced with
a passed-in function parameter.

Start with build_read_segment() because it is a common utility
function at the bottom of the Read chunk path.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Move read_info::ri_pageoff into struct svc_rdma_recv_ctxt
Chuck Lever [Mon, 4 Dec 2023 14:57:28 +0000 (09:57 -0500)]
svcrdma: Move read_info::ri_pageoff into struct svc_rdma_recv_ctxt

Further clean up: move the starting byte offset field into
svc_rdma_recv_ctxt.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Move svc_rdma_read_info::ri_pageno to struct svc_rdma_recv_ctxt
Chuck Lever [Mon, 4 Dec 2023 14:57:22 +0000 (09:57 -0500)]
svcrdma: Move svc_rdma_read_info::ri_pageno to struct svc_rdma_recv_ctxt

Further clean up: move the page index field into svc_rdma_recv_ctxt.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Start moving fields out of struct svc_rdma_read_info
Chuck Lever [Mon, 4 Dec 2023 14:57:16 +0000 (09:57 -0500)]
svcrdma: Start moving fields out of struct svc_rdma_read_info

Since the request's svc_rdma_recv_ctxt will stay around for the
duration of the RDMA Read operation, the contents of struct
svc_rdma_read_info can reside in the request's svc_rdma_recv_ctxt
rather than being allocated separately. This will eventually save a
call to kmalloc() in a hot path.

Start this clean-up by moving the Read chunk's svc_rdma_chunk_ctxt.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Move struct svc_rdma_chunk_ctxt to svc_rdma.h
Chuck Lever [Mon, 4 Dec 2023 14:57:09 +0000 (09:57 -0500)]
svcrdma: Move struct svc_rdma_chunk_ctxt to svc_rdma.h

Prepare for nestling these into the send and recv ctxts so they
no longer have to be allocated dynamically.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Remove the svc_rdma_chunk_ctxt::cc_rdma field
Chuck Lever [Mon, 4 Dec 2023 14:57:03 +0000 (09:57 -0500)]
svcrdma: Remove the svc_rdma_chunk_ctxt::cc_rdma field

In every instance, the pointer address in that field is now
available by other means.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Pass a pointer to the transport to svc_rdma_cc_release()
Chuck Lever [Mon, 4 Dec 2023 14:56:57 +0000 (09:56 -0500)]
svcrdma: Pass a pointer to the transport to svc_rdma_cc_release()

Enable the eventual removal of the svc_rdma_chunk_ctxt::cc_rdma
field.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Explicitly pass the transport to svc_rdma_post_chunk_ctxt()
Chuck Lever [Mon, 4 Dec 2023 14:56:50 +0000 (09:56 -0500)]
svcrdma: Explicitly pass the transport to svc_rdma_post_chunk_ctxt()

Enable the eventual removal of the svc_rdma_chunk_ctxt::cc_rdma
field.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Explicitly pass the transport into Read chunk I/O paths
Chuck Lever [Mon, 4 Dec 2023 14:56:44 +0000 (09:56 -0500)]
svcrdma: Explicitly pass the transport into Read chunk I/O paths

Enable the eventual removal of the svc_rdma_chunk_ctxt::cc_rdma
field.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Explicitly pass the transport into Write chunk I/O paths
Chuck Lever [Mon, 4 Dec 2023 14:56:37 +0000 (09:56 -0500)]
svcrdma: Explicitly pass the transport into Write chunk I/O paths

Enable the eventual removal of the svc_rdma_chunk_ctxt::cc_rdma
field.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Acquire the svcxprt_rdma pointer from the CQ context
Chuck Lever [Mon, 4 Dec 2023 14:56:31 +0000 (09:56 -0500)]
svcrdma: Acquire the svcxprt_rdma pointer from the CQ context

Enable the removal of the svc_rdma_chunk_ctxt::cc_rdma field in a
subsequent patch.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Reduce size of struct svc_rdma_rw_ctxt
Chuck Lever [Mon, 4 Dec 2023 14:56:24 +0000 (09:56 -0500)]
svcrdma: Reduce size of struct svc_rdma_rw_ctxt

SG_CHUNK_SIZE is 128, making struct svc_rdma_rw_ctxt + the first
SGL array more than 4200 bytes in length, pushing the memory
allocation well into order 1.

Even so, the RDMA rw core doesn't seem to use more than max_send_sge
entries in that array (typically 32 or less), so that is all wasted
space.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Update some svcrdma DMA-related tracepoints
Chuck Lever [Mon, 27 Nov 2023 16:33:50 +0000 (11:33 -0500)]
svcrdma: Update some svcrdma DMA-related tracepoints

A send/recv_ctxt already records transport-related information
in the cq.id, thus there is no need to record the IP addresses of
the transport endpoints.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: DMA error tracepoints should report completion IDs
Chuck Lever [Mon, 27 Nov 2023 16:33:43 +0000 (11:33 -0500)]
svcrdma: DMA error tracepoints should report completion IDs

Update the DMA error flow tracepoints to report the completion ID of
the failing context. This ties the wait/failure to a particular
operation or request, which is more useful than knowing only the
failing transport.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: SQ error tracepoints should report completion IDs
Chuck Lever [Mon, 27 Nov 2023 16:33:37 +0000 (11:33 -0500)]
svcrdma: SQ error tracepoints should report completion IDs

Update the Send Queue's error flow tracepoints to report the
completion ID of the waiting or failing context. This ties the
wait/failure to a particular operation or request, which is a little
more useful than knowing only the transport that is about to close.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agorpcrdma: Introduce a simple cid tracepoint class
Chuck Lever [Mon, 27 Nov 2023 16:33:30 +0000 (11:33 -0500)]
rpcrdma: Introduce a simple cid tracepoint class

De-duplicate some code, making it easier to add new tracepoints that
report only a completion ID.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Add lockdep class keys for transport locks
Chuck Lever [Mon, 27 Nov 2023 16:33:24 +0000 (11:33 -0500)]
svcrdma: Add lockdep class keys for transport locks

Two svcrdma-related transport locks can become quite contended.
Collate their use and make them easy to find in /proc/lock_stat for
better observability.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Clean up locking
Chuck Lever [Tue, 21 Nov 2023 16:40:46 +0000 (11:40 -0500)]
svcrdma: Clean up locking

There's no need to protect llist_entry() with a spin lock.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Add an async version of svc_rdma_write_info_free()
Chuck Lever [Tue, 21 Nov 2023 16:40:39 +0000 (11:40 -0500)]
svcrdma: Add an async version of svc_rdma_write_info_free()

DMA unmapping can take quite some time, so it should not be handled
in a single-threaded completion handler. Defer releasing write_info
structs to the recently-added workqueue.

With this patch, DMA unmapping can be handled in parallel, and it
does not cause head-of-queue blocking of Write completions.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Add an async version of svc_rdma_send_ctxt_put()
Chuck Lever [Tue, 21 Nov 2023 16:40:33 +0000 (11:40 -0500)]
svcrdma: Add an async version of svc_rdma_send_ctxt_put()

DMA unmapping can take quite some time, so it should not be handled
in a single-threaded completion handler. Defer releasing send_ctxts
to the recently-added workqueue.

With this patch, DMA unmapping can be handled in parallel, and it
does not cause head-of-queue blocking of Send completions.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Add a utility workqueue to svcrdma
Chuck Lever [Tue, 21 Nov 2023 16:40:26 +0000 (11:40 -0500)]
svcrdma: Add a utility workqueue to svcrdma

To handle work in the background, set up an UNBOUND workqueue for
svcrdma. Subsequent patches will make use of it.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Pre-allocate svc_rdma_recv_ctxt objects
Chuck Lever [Tue, 21 Nov 2023 16:40:20 +0000 (11:40 -0500)]
svcrdma: Pre-allocate svc_rdma_recv_ctxt objects

The original reason for allocating svc_rdma_recv_ctxt objects during
Receive completion was to ensure the objects were allocated on the
NUMA node closest to the underlying IB device.

Since commit c5d68d25bd6b ("svcrdma: Clean up allocation of
svc_rdma_recv_ctxt"), however, the device's favored node is
explicitly passed to the memory allocator.

To enable switching Receive completion to soft IRQ context, move
memory allocation out of completion handling, since it can be
costly, and it can sleep.

A limited number of objects is now allocated at "accept" time.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agosvcrdma: Eliminate allocation of recv_ctxt objects in backchannel
Chuck Lever [Tue, 21 Nov 2023 16:40:13 +0000 (11:40 -0500)]
svcrdma: Eliminate allocation of recv_ctxt objects in backchannel

The svc_rdma_recv_ctxt free list uses a lockless list to avoid the
need for a spin lock in the fast path. llist_del_first(), which is
used by svc_rdma_recv_ctxt_get(), requires serialization, however,
when there are multiple list producers that are unserialized.

I mistakenly thought there was only one caller of
svc_rdma_recv_ctxt_get() (svc_rdma_refresh_recvs()), thus explicit
serialization would not be necessary. But there is another caller:
svc_rdma_bc_sendto(), and these two are not serialized against each
other. I haven't seen ill effects that I could directly ascribe to
a lack of serialization. It's just an observation based on code
audit.

When DMA-mapping before sending a Reply, the passed-in struct
svc_rdma_recv_ctxt is used only for its write and reply PCLs. These
are currently always empty in the backchannel case. So, instead of
passing a full svc_rdma_recv_ctxt object to
svc_rdma_map_reply_msg(), let's pass in just the Write and Reply
PCLs.

This change makes it unnecessary for the backchannel to acquire a
dummy svc_rdma_recv_ctxt object when sending an RPC Call. The need
for svc_rdma_recv_ctxt free list serialization is now completely
avoided.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agoNFSv4, NFSD: move enum nfs_cb_opnum4 to include/linux/nfs4.h
ChenXiaoSong [Sat, 2 Dec 2023 21:07:25 +0000 (21:07 +0000)]
NFSv4, NFSD: move enum nfs_cb_opnum4 to include/linux/nfs4.h

Callback operations enum is defined in client and server, move it to
common header file.

Signed-off-by: ChenXiaoSong <chenxiaosong@kylinos.cn>
Acked-by: Anna Schumaker <Anna.Schumaker@netapp.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agonfsd: remove unnecessary NULL check
Dan Carpenter [Mon, 4 Dec 2023 12:30:06 +0000 (15:30 +0300)]
nfsd: remove unnecessary NULL check

We check "state" for NULL on the previous line so it can't be NULL here.
No need to check again.

Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/r/202312031425.LffZTarR-lkp@intel.com/
Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agoSUNRPC: Remove RQ_SPLICE_OK
Chuck Lever [Fri, 17 Nov 2023 22:14:46 +0000 (17:14 -0500)]
SUNRPC: Remove RQ_SPLICE_OK

This flag is no longer used.

Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agoNFSD: Modify NFSv4 to use nfsd_read_splice_ok()
Chuck Lever [Fri, 17 Nov 2023 22:14:40 +0000 (17:14 -0500)]
NFSD: Modify NFSv4 to use nfsd_read_splice_ok()

Avoid the use of an atomic bitop, and prepare for adding a run-time
switch for using splice reads.

Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agoNFSD: Replace RQ_SPLICE_OK in nfsd_read()
Chuck Lever [Fri, 17 Nov 2023 22:14:33 +0000 (17:14 -0500)]
NFSD: Replace RQ_SPLICE_OK in nfsd_read()

RQ_SPLICE_OK is a bit of a layering violation. Also, a subsequent
patch is going to provide a mechanism for always disabling splice
reads.

Splicing is an issue only for NFS READs, so refactor nfsd_read() to
check the auth type directly instead of relying on an rq_flag
setting.

The new helper will be added into the NFSv4 read path in a
subsequent patch.

Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agoSUNRPC: Add a server-side API for retrieving an RPC's pseudoflavor
Chuck Lever [Fri, 17 Nov 2023 22:14:27 +0000 (17:14 -0500)]
SUNRPC: Add a server-side API for retrieving an RPC's pseudoflavor

NFSD will use this new API to determine whether nfsd_splice_read is
safe to use. This avoids the need to add a dependency to NFSD for
CONFIG_SUNRPC_GSS.

Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agoNFSD: Document lack of f_pos_lock in nfsd_readdir()
Chuck Lever [Mon, 20 Nov 2023 01:17:11 +0000 (20:17 -0500)]
NFSD: Document lack of f_pos_lock in nfsd_readdir()

Al Viro notes that normal system calls hold f_pos_lock when calling
->iterate_shared and ->llseek; however nfsd_readdir() does not take
that mutex when calling these methods.

It should be safe however because the struct file acquired by
nfsd_readdir() is not visible to other threads.

Reviewed-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agoNFSD: Remove nfsd_drc_gc() tracepoint
Chuck Lever [Mon, 13 Nov 2023 13:45:07 +0000 (08:45 -0500)]
NFSD: Remove nfsd_drc_gc() tracepoint

This trace point was for debugging the DRC's garbage collection. In
the field it's just noise.

Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agoNFSD: Make the file_delayed_close workqueue UNBOUND
Chuck Lever [Sun, 22 Oct 2023 22:50:10 +0000 (18:50 -0400)]
NFSD: Make the file_delayed_close workqueue UNBOUND

workqueue: nfsd_file_delayed_close [nfsd] hogged CPU for >13333us 8
times, consider switching to WQ_UNBOUND

There's no harm in closing a cached file descriptor on another core.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agoNFSD: use read_seqbegin() rather than read_seqbegin_or_lock()
Oleg Nesterov [Thu, 26 Oct 2023 14:50:18 +0000 (16:50 +0200)]
NFSD: use read_seqbegin() rather than read_seqbegin_or_lock()

The usage of read_seqbegin_or_lock() in nfsd_copy_write_verifier()
is wrong. "seq" is always even and thus "or_lock" has no effect,
this code can never take ->writeverf_lock for writing.

I guess this is fine, nfsd_copy_write_verifier() just copies 8 bytes
and nfsd_reset_write_verifier() is supposed to be very rare operation
so we do not need the adaptive locking in this case.

Yet the code looks wrong and sub-optimal, it can use read_seqbegin()
without changing the behaviour.

[ cel: Note also that it eliminates this Sparse warning:

fs/nfsd/nfssvc.c:360:6: warning: context imbalance in 'nfsd_copy_write_verifier' -
different lock contexts for basic block

]

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: NeilBrown <neilb@suse.de>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agonfsd: new Kconfig option for legacy client tracking
Jeff Layton [Fri, 13 Oct 2023 13:03:53 +0000 (09:03 -0400)]
nfsd: new Kconfig option for legacy client tracking

We've had a number of attempts at different NFSv4 client tracking
methods over the years, but now nfsdcld has emerged as the clear winner
since the others (recoverydir and the usermodehelper upcall) are
problematic.

As a case in point, the recoverydir backend uses MD5 hashes to encode
long form clientid strings, which means that nfsd repeatedly gets dinged
on FIPS audits, since MD5 isn't considered secure. Its use of MD5 is not
cryptographically significant, so there is no danger there, but allowing
us to compile that out allows us to sidestep the issue entirely.

As a prelude to eventually removing support for these client tracking
methods, add a new Kconfig option that enables them. Mark it deprecated
and make it default to N.

Acked-by: NeilBrown <neilb@suse.de>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agoLinux 6.7
Linus Torvalds [Sun, 7 Jan 2024 20:18:38 +0000 (12:18 -0800)]
Linux 6.7

8 months agoMerge tag 'i2c-for-6.7-final' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa...
Linus Torvalds [Sat, 6 Jan 2024 19:35:37 +0000 (11:35 -0800)]
Merge tag 'i2c-for-6.7-final' of git://git./linux/kernel/git/wsa/linux

Pull i2c fixes from Wolfram Sang:
 "Improve the detection when to run atomic transfer handlers for kernels
  with preemption disabled. This removes some false positive splats a
  number of users were seeing if their driver didn't have support for
  atomic transfers.

  Also, fix a typo in the docs while we are here"

* tag 'i2c-for-6.7-final' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
  i2c: core: Fix atomic xfer check for non-preempt config
  Documentation/i2c: fix spelling error in i2c-address-translators

8 months agoi2c: core: Fix atomic xfer check for non-preempt config
Benjamin Bara [Thu, 4 Jan 2024 08:17:08 +0000 (09:17 +0100)]
i2c: core: Fix atomic xfer check for non-preempt config

Since commit aa49c90894d0 ("i2c: core: Run atomic i2c xfer when
!preemptible"), the whole reboot/power off sequence on non-preempt kernels
is using atomic i2c xfer, as !preemptible() always results to 1.

During device_shutdown(), the i2c might be used a lot and not all busses
have implemented an atomic xfer handler. This results in a lot of
avoidable noise, like:

[   12.687169] No atomic I2C transfer handler for 'i2c-0'
[   12.692313] WARNING: CPU: 6 PID: 275 at drivers/i2c/i2c-core.h:40 i2c_smbus_xfer+0x100/0x118
...

Fix this by allowing non-atomic xfer when the interrupts are enabled, as
it was before.

Link: https://lore.kernel.org/r/20231222230106.73f030a5@yea
Link: https://lore.kernel.org/r/20240102150350.3180741-1-mwalle@kernel.org
Link: https://lore.kernel.org/linux-i2c/13271b9b-4132-46ef-abf8-2c311967bb46@mailbox.org/
Fixes: aa49c90894d0 ("i2c: core: Run atomic i2c xfer when !preemptible")
Cc: stable@vger.kernel.org # v5.2+
Signed-off-by: Benjamin Bara <benjamin.bara@skidata.com>
Tested-by: Michael Walle <mwalle@kernel.org>
Tested-by: Tor Vic <torvic9@mailbox.org>
[wsa: removed a comment which needs more work, code is ok]
Signed-off-by: Wolfram Sang <wsa@kernel.org>
8 months agoMerge tag 'mm-hotfixes-stable-2024-01-05-11-35' of git://git.kernel.org/pub/scm/linux...
Linus Torvalds [Fri, 5 Jan 2024 21:46:18 +0000 (13:46 -0800)]
Merge tag 'mm-hotfixes-stable-2024-01-05-11-35' of git://git./linux/kernel/git/akpm/mm

Pull misc mm fixes from Andrew Morton:
 "12 hotfixes.

  Two are cc:stable and the remainder either address post-6.7 issues or
  aren't considered necessary for earlier kernel versions"

* tag 'mm-hotfixes-stable-2024-01-05-11-35' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm:
  mm: shrinker: use kvzalloc_node() from expand_one_shrinker_info()
  mailmap: add entries for Mathieu Othacehe
  MAINTAINERS: change vmware.com addresses to broadcom.com
  arch/mm/fault: fix major fault accounting when retrying under per-VMA lock
  mm/mglru: skip special VMAs in lru_gen_look_around()
  MAINTAINERS: hand over hwpoison maintainership to Miaohe Lin
  MAINTAINERS: remove hugetlb maintainer Mike Kravetz
  mm: fix unmap_mapping_range high bits shift bug
  mm: memcg: fix split queue list crash when large folio migration
  mm: fix arithmetic for max_prop_frac when setting max_ratio
  mm: fix arithmetic for bdi min_ratio
  mm: align larger anonymous mappings on THP boundaries

8 months agoMerge tag 'nfsd-6.7-3' of git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux
Linus Torvalds [Fri, 5 Jan 2024 21:12:29 +0000 (13:12 -0800)]
Merge tag 'nfsd-6.7-3' of git://git./linux/kernel/git/cel/linux

Pull nfsd fix from Chuck Lever:

 - Fix another regression in the NFSD administrative API

* tag 'nfsd-6.7-3' of git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux:
  nfsd: drop the nfsd_put helper

8 months agoMerge tag 'firewire-fixes-6.7-final' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Fri, 5 Jan 2024 20:26:26 +0000 (12:26 -0800)]
Merge tag 'firewire-fixes-6.7-final' of git://git./linux/kernel/git/ieee1394/linux1394

Pull firewire fix from Takashi Sakamoto:
 "A single patch to suppress unexpected system reboot in AMD Ryzen
  machines with PCIe card consisting of Asmedia ASM1083/1085 and
  VT6306/6307/6308.

  When the 1394 OHCI driver for the card accesses a specific register
  in PCI memory space, the system reboot often occurs.

  The issue affects all versions of Linux kernel as long as the 1394
  OHCI driver is included. The mechanism of unexpected system reboot is
  not clear, so the driver is changed to avoid the access itself when
  detecting the combination of hardware"

* tag 'firewire-fixes-6.7-final' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee1394/linux1394:
  firewire: ohci: suppress unexpected system reboot in AMD Ryzen machines and ASM108x/VT630x PCIe cards

8 months agoMerge tag 'mmc-v6.7-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc
Linus Torvalds [Fri, 5 Jan 2024 20:12:33 +0000 (12:12 -0800)]
Merge tag 'mmc-v6.7-rc4' of git://git./linux/kernel/git/ulfh/mmc

Pull MMC fixes from Ulf Hansson:
 "MMC core:
   - Fix releasing the host by canceling the delayed work
   - Fix pause retune on all RPMB partitions

  MMC host:
   - meson-mx-sdhc: Fix HW hang during card initialization
   - sdhci-sprd: Fix eMMC init failure after HW reset"

* tag 'mmc-v6.7-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/mmc:
  mmc: sdhci-sprd: Fix eMMC init failure after hw reset
  mmc: core: Cancel delayed work before releasing host
  mmc: rpmb: fixes pause retune on all RPMB partitions.
  mmc: meson-mx-sdhc: Fix initialization frozen issue

8 months agoMerge tag 'drm-fixes-2024-01-05' of git://anongit.freedesktop.org/drm/drm
Linus Torvalds [Fri, 5 Jan 2024 20:02:20 +0000 (12:02 -0800)]
Merge tag 'drm-fixes-2024-01-05' of git://anongit.freedesktop.org/drm/drm

Pull more drm fixes from Dave Airlie:
 "The amdgpu ones are fairly normal, the one that is a bit large is a
  fix for a newly introduced IP in 6.7 so unlikely to cause regressions.

  The nouveau ones are mostly memory leaks and debugging cleanups from
  the GSP (new nvidia firmware) enablement. There are some GSP changes
  to the message passing code and a subsequent fix for eDP panel turn
  on, that means my laptop can turn on the panel in GSP mode. These are
  fairly low chance of disrupting things since GSP is new in 6.7. The
  final not all in GSP fix is a deadlock seen with i915/nouveau when GSP
  is used where the the fence and irq paths have locking inversions,
  I've pushed some irq enablement out to a workqueue, and this has seen
  some fairly decent testing.

  amdgpu:
   - DP MST fix
   - SMU 13.0.6 fixes
   - fix displays on macbooks using vega12
   - fix VSC and colorimetry on DP/eDP

  nouveau:
   - fix deadlock between fence signalling and irq paths
   - fix GSP memory leaks
   - fix GSP leftover debug
   - hide some GSP callback messages
   - fix GSP display disable path
   - fix GSP ACPI interaction
   - handle errors in ctrl messages
   - use errors info to fix DP link training"

* tag 'drm-fixes-2024-01-05' of git://anongit.freedesktop.org/drm/drm:
  drm/nouveau/dp: Honor GSP link training retry timeouts
  nouveau: push event block/allowing out of the fence context
  nouveau/gsp: always free the alloc messages on r535
  nouveau/gsp: don't free ctrl messages on errors
  nouveau/gsp: convert gsp errors to generic errors
  drm/nouveau/gsp: Fix ACPI MXDM/MXDS method invocations
  nouveau/gsp: free userd allocation.
  nouveau/gsp: free acpi object after use
  nouveau: fix disp disabling with GSP
  nouveau/gsp: drop some acpi related debug
  nouveau/gsp: add three notifier callbacks that we see in normal operation (v2)
  drm/amd/pm: Use gpu_metrics_v1_5 for SMUv13.0.6
  drm/amd/pm: Add gpu_metrics_v1_5
  drm/amd/pm: Add mem_busy_percent for GCv9.4.3 apu
  drm/amd/display: Fix sending VSC (+ colorimetry) packets for DP/eDP displays without PSR
  drm/amdgpu: skip gpu_info fw loading on navi12
  drm/amd/display: add nv12 bounding box
  drm/amd/pm: Update metric table for jpeg/vcn data
  drm/amd/pm: Use separate metric table for APU
  drm/amd/display: pbn_div need be updated for hotplug event

8 months agomm: shrinker: use kvzalloc_node() from expand_one_shrinker_info()
Tetsuo Handa [Wed, 3 Jan 2024 01:52:22 +0000 (10:52 +0900)]
mm: shrinker: use kvzalloc_node() from expand_one_shrinker_info()

syzbot is reporting uninit-value at shrinker_alloc(), for commit
307bececcd12 ("mm: shrinker: add a secondary array for
shrinker_info::{map, nr_deferred}") which assumed that the ->unit was
allocated with __GFP_ZERO forgot to replace kvmalloc_node() in
expand_one_shrinker_info() with kvzalloc_node().

Link: https://lkml.kernel.org/r/9226cc0a-10e0-4489-80c5-58c3b5b4359c@I-love.SAKURA.ne.jp
Reported-by: syzbot <syzbot+1e0ed05798af62917464@syzkaller.appspotmail.com>
Closes: https://syzkaller.appspot.com/bug?extid=1e0ed05798af62917464
Fixes: 307bececcd12 ("mm: shrinker: add a secondary array for shrinker_info::{map, nr_deferred}")
Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Acked-by: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
8 months agoMerge tag 'soc-fixes-6.7-3a' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc
Linus Torvalds [Fri, 5 Jan 2024 17:39:24 +0000 (09:39 -0800)]
Merge tag 'soc-fixes-6.7-3a' of git://git./linux/kernel/git/soc/soc

Pull ARM SoC fixes from Arnd Bergmann:
 "These are two correctness fixes for handing DT input in the
  Allwinner (sunxi) SMP startup code"

* tag 'soc-fixes-6.7-3a' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc:
  ARM: sun9i: smp: fix return code check of of_property_match_string
  ARM: sun9i: smp: Fix array-index-out-of-bounds read in sunxi_mc_smp_init

8 months agoMerge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
Linus Torvalds [Fri, 5 Jan 2024 17:16:15 +0000 (09:16 -0800)]
Merge tag 'for-linus' of git://git./virt/kvm/kvm

Pull kvm fix from Paolo Bonzini:

 - Fix boolean logic in intel_guest_get_msrs

* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
  KVM: x86/pmu: fix masking logic for MSR_CORE_PERF_GLOBAL_CTRL

8 months agoMerge tag 'probes-fixes-v6.7-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 5 Jan 2024 17:07:59 +0000 (09:07 -0800)]
Merge tag 'probes-fixes-v6.7-rc8' of git://git./linux/kernel/git/trace/linux-trace

Pull kprobes/x86 fix from Masami Hiramatsu:

 - Fix to emulate indirect call which size is not 5 byte.

   Current code expects the indirect call instructions are 5 bytes, but
   that is incorrect. Usually indirect call based on register is shorter
   than that, thus the emulation causes a kernel crash by accessing
   wrong instruction boundary. This uses the instruction size to
   calculate the return address correctly.

* tag 'probes-fixes-v6.7-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace:
  x86/kprobes: fix incorrect return address calculation in kprobe_emulate_call_indirect

8 months agoMerge tag '6.7-rc8-smb3-mchan-fixes' of git://git.samba.org/sfrench/cifs-2.6
Linus Torvalds [Fri, 5 Jan 2024 16:52:25 +0000 (08:52 -0800)]
Merge tag '6.7-rc8-smb3-mchan-fixes' of git://git.samba.org/sfrench/cifs-2.6

Pull smb client fixes from Steve French:
 "Three important multichannel smb3 client fixes found in recent
  testing:

   - fix oops due to incorrect refcounting of interfaces after
     disabling multichannel

   - fix possible unrecoverable session state after disabling
     multichannel with active sessions

   - fix two places that were missing use of chan_lock"

* tag '6.7-rc8-smb3-mchan-fixes' of git://git.samba.org/sfrench/cifs-2.6:
  cifs: do not depend on release_iface for maintaining iface_list
  cifs: cifs_chan_is_iface_active should be called with chan_lock held
  cifs: after disabling multichannel, mark tcon for reconnect

8 months agofirewire: ohci: suppress unexpected system reboot in AMD Ryzen machines and ASM108x...
Takashi Sakamoto [Tue, 2 Jan 2024 11:01:50 +0000 (20:01 +0900)]
firewire: ohci: suppress unexpected system reboot in AMD Ryzen machines and ASM108x/VT630x PCIe cards

VIA VT6306/6307/6308 provides PCI interface compliant to 1394 OHCI. When
the hardware is combined with Asmedia ASM1083/1085 PCIe-to-PCI bus bridge,
it appears that accesses to its 'Isochronous Cycle Timer' register (offset
0xf0 on PCI memory space) often causes unexpected system reboot in any
type of AMD Ryzen machine (both 0x17 and 0x19 families). It does not
appears in the other type of machine (AMD pre-Ryzen machine, Intel
machine, at least), or in the other OHCI 1394 hardware (e.g. Texas
Instruments).

The issue explicitly appears at a commit dcadfd7f7c74 ("firewire: core:
use union for callback of transaction completion") added to v6.5 kernel.
It changed 1394 OHCI driver to access to the register every time to
dispatch local asynchronous transaction. However, the issue exists in
older version of kernel as long as it runs in AMD Ryzen machine, since
the access to the register is required to maintain bus time. It is not
hard to imagine that users experience the unexpected system reboot when
generating bus reset by plugging any devices in, or reading the register
by time-aware application programs; e.g. audio sample processing.

This commit suppresses the unexpected system reboot in the combination of
hardware. It avoids the access itself. As a result, the software stack can
not provide the hardware time anymore to unit drivers, userspace
applications, and nodes in the same IEEE 1394 bus. It brings apparent
disadvantage since time-aware application programs require it, while
time-unaware applications are available again; e.g. sbp2.

Cc: stable@vger.kernel.org
Reported-by: Jiri Slaby <jirislaby@kernel.org>
Closes: https://bugzilla.suse.com/show_bug.cgi?id=1215436
Reported-by: Mario Limonciello <mario.limonciello@amd.com>
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=217994
Reported-by: Tobias Gruetzmacher <tobias-lists@23.gs>
Closes: https://sourceforge.net/p/linux1394/mailman/message/58711901/
Closes: https://bugzilla.redhat.com/show_bug.cgi?id=2240973
Closes: https://bugs.launchpad.net/linux/+bug/2043905
Link: https://lore.kernel.org/r/20240102110150.244475-1-o-takashi@sakamocchi.jp
Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
8 months agonfsd: drop the nfsd_put helper
Jeff Layton [Wed, 3 Jan 2024 13:36:52 +0000 (08:36 -0500)]
nfsd: drop the nfsd_put helper

It's not safe to call nfsd_put once nfsd_last_thread has been called, as
that function will zero out the nn->nfsd_serv pointer.

Drop the nfsd_put helper altogether and open-code the svc_put in its
callers instead. That allows us to not be reliant on the value of that
pointer when handling an error.

Fixes: 2a501f55cd64 ("nfsd: call nfsd_last_thread() before final nfsd_put()")
Reported-by: Zhi Li <yieli@redhat.com>
Cc: NeilBrown <neilb@suse.de>
Signed-off-by: Jeffrey Layton <jlayton@redhat.com>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
8 months agodrm/nouveau/dp: Honor GSP link training retry timeouts
Lyude Paul [Fri, 22 Dec 2023 04:32:00 +0000 (14:32 +1000)]
drm/nouveau/dp: Honor GSP link training retry timeouts

Turns out that one of the ways that Nvidia's driver handles the pre-LT
timeout for eDP panels is by providing a retry timeout in their link
training callbacks that we're expected to wait for. Up until now we didn't
pay any attention to this parameter.

So, start honoring the timeout if link training fails - and retry up to 3
times. The "3 times" bit comes from OpenRM's link training code.

[airlied: this fixes the panel on one of my laptops]

Signed-off-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231222043308.3090089-12-airlied@gmail.com
8 months agonouveau: push event block/allowing out of the fence context
Dave Airlie [Fri, 22 Dec 2023 04:31:59 +0000 (14:31 +1000)]
nouveau: push event block/allowing out of the fence context

There is a deadlock between the irq and fctx locks,
the irq handling takes irq then fctx lock
the fence signalling takes fctx then irq lock

This splits the fence signalling path so the code that hits
the irq lock is done in a separate work queue.

This seems to fix crashes/hangs when using nouveau gsp with
i915 primary GPU.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231222043308.3090089-11-airlied@gmail.com
8 months agonouveau/gsp: always free the alloc messages on r535
Dave Airlie [Fri, 22 Dec 2023 04:31:58 +0000 (14:31 +1000)]
nouveau/gsp: always free the alloc messages on r535

Fixes a memory leak seen with kmemleak.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231222043308.3090089-10-airlied@gmail.com
8 months agonouveau/gsp: don't free ctrl messages on errors
Dave Airlie [Fri, 22 Dec 2023 04:31:57 +0000 (14:31 +1000)]
nouveau/gsp: don't free ctrl messages on errors

It looks like for some messages the upper layers need to get access to the
results of the message so we can interpret it.

Rework the ctrl push interface to not free things and cleanup properly
whereever it errors out.

Requested-by: Lyude
Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231222043308.3090089-9-airlied@gmail.com
8 months agonouveau/gsp: convert gsp errors to generic errors
Dave Airlie [Fri, 22 Dec 2023 04:31:56 +0000 (14:31 +1000)]
nouveau/gsp: convert gsp errors to generic errors

This should let the upper layers retry as needed on EAGAIN.

There may be other values we will care about in the future, but
this covers our present needs.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231222043308.3090089-8-airlied@gmail.com
8 months agodrm/nouveau/gsp: Fix ACPI MXDM/MXDS method invocations
Lyude Paul [Fri, 22 Dec 2023 04:31:55 +0000 (14:31 +1000)]
drm/nouveau/gsp: Fix ACPI MXDM/MXDS method invocations

Currently we get an error from ACPI because both of these arguments expect
a single argument, and we don't provide one. I'm not totally clear on what
that argument does, but we're able to find the missing value from
_acpiCacheMethodData() in src/kernel/platform/acpi_common.c in nvidia's
driver. So, let's add that - which doesn't get eDP displays to power on
quite yet, but gets rid of the argument warning at least.

Signed-off-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231222043308.3090089-7-airlied@gmail.com
8 months agonouveau/gsp: free userd allocation.
Dave Airlie [Fri, 22 Dec 2023 04:31:54 +0000 (14:31 +1000)]
nouveau/gsp: free userd allocation.

This was being leaked.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231222043308.3090089-6-airlied@gmail.com
8 months agonouveau/gsp: free acpi object after use
Dave Airlie [Fri, 22 Dec 2023 04:31:53 +0000 (14:31 +1000)]
nouveau/gsp: free acpi object after use

This fixes a memory leak for the acpi dod object.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231222043308.3090089-5-airlied@gmail.com
8 months agonouveau: fix disp disabling with GSP
Dave Airlie [Fri, 22 Dec 2023 04:31:52 +0000 (14:31 +1000)]
nouveau: fix disp disabling with GSP

This func ptr here is normally static allocation, but gsp r535
uses a dynamic pointer, so we need to handle that better.

This fixes a crash with GSP when you use config=disp=0 to avoid
disp problems.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231222043308.3090089-4-airlied@gmail.com
8 months agonouveau/gsp: drop some acpi related debug
Dave Airlie [Fri, 22 Dec 2023 04:31:51 +0000 (14:31 +1000)]
nouveau/gsp: drop some acpi related debug

These were leftover debug, if we need to bring them back do so
for debugging later.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231222043308.3090089-3-airlied@gmail.com
8 months agonouveau/gsp: add three notifier callbacks that we see in normal operation (v2)
Dave Airlie [Fri, 22 Dec 2023 04:31:50 +0000 (14:31 +1000)]
nouveau/gsp: add three notifier callbacks that we see in normal operation (v2)

Add NULL callbacks for some things GSP calls that we don't handle, but know about
so we avoid the logging.

v2: Timur suggested allowing null fn.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20231222043308.3090089-2-airlied@gmail.com
8 months agoMerge tag 'amd-drm-fixes-6.7-2024-01-04' of https://gitlab.freedesktop.org/agd5f...
Dave Airlie [Fri, 5 Jan 2024 01:28:34 +0000 (11:28 +1000)]
Merge tag 'amd-drm-fixes-6.7-2024-01-04' of https://gitlab.freedesktop.org/agd5f/linux into drm-fixes

amdgpu:
- DP MST fix
- SMU 13.0.6 fixes
- Fix displays on macbooks using vega12
- Fix VSC and colorimetry on DP/eDP

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexander.deucher@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240104152139.4931-1-alexander.deucher@amd.com
8 months agoMerge tag 'net-6.7-rc9' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net
Linus Torvalds [Fri, 5 Jan 2024 00:34:50 +0000 (16:34 -0800)]
Merge tag 'net-6.7-rc9' of git://git./linux/kernel/git/netdev/net

Pull networking fixes from Jakub Kicinski:
 "Including fixes from wireless and netfilter.

  We haven't accumulated much over the break. If it wasn't for the
  uninterrupted stream of fixes for Intel drivers this PR would be very
  slim. There was a handful of user reports, however, either they stood
  out because of the lower traffic or users have had more time to test
  over the break. The ones which are v6.7-relevant should be wrapped up.

  Current release - regressions:

   - Revert "net: ipv6/addrconf: clamp preferred_lft to the minimum
     required", it caused issues on networks where routers send prefixes
     with preferred_lft=0

   - wifi:
      - iwlwifi: pcie: don't synchronize IRQs from IRQ, prevent deadlock
      - mac80211: fix re-adding debugfs entries during reconfiguration

  Current release - new code bugs:

   - tcp: print AO/MD5 messages only if there are any keys

  Previous releases - regressions:

   - virtio_net: fix missing dma unmap for resize, prevent OOM

  Previous releases - always broken:

   - mptcp: prevent tcp diag from closing listener subflows

   - nf_tables:
      - set transport header offset for egress hook, fix IPv4 mangling
      - skip set commit for deleted/destroyed sets, avoid double deactivation

   - nat: make sure action is set for all ct states, fix openvswitch
     matching on ICMP packets in related state

   - eth: mlxbf_gige: fix receive hang under heavy traffic

   - eth: r8169: fix PCI error on system resume for RTL8168FP

   - net: add missing getsockopt(SO_TIMESTAMPING_NEW) and cmsg handling"

* tag 'net-6.7-rc9' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (52 commits)
  net/tcp: Only produce AO/MD5 logs if there are any keys
  net: Implement missing SO_TIMESTAMPING_NEW cmsg support
  bnxt_en: Remove mis-applied code from bnxt_cfg_ntp_filters()
  net: ravb: Wait for operating mode to be applied
  asix: Add check for usbnet_get_endpoints
  octeontx2-af: Re-enable MAC TX in otx2_stop processing
  octeontx2-af: Always configure NIX TX link credits based on max frame size
  net/smc: fix invalid link access in dumping SMC-R connections
  net/qla3xxx: fix potential memleak in ql_alloc_buffer_queues
  virtio_net: fix missing dma unmap for resize
  igc: Fix hicredit calculation
  ice: fix Get link status data length
  i40e: Restore VF MSI-X state during PCI reset
  i40e: fix use-after-free in i40e_aqc_add_filters()
  net: Save and restore msg_namelen in sock_sendmsg
  netfilter: nft_immediate: drop chain reference counter on error
  netfilter: nf_nat: fix action not being set for all ct states
  net: bcmgenet: Fix FCS generation for fragmented skbuffs
  mptcp: prevent tcp diag from closing listener subflows
  MAINTAINERS: add Geliang as reviewer for MPTCP
  ...

8 months agox86/csum: clean up `csum_partial' further
Linus Torvalds [Tue, 27 Jun 2023 20:55:32 +0000 (13:55 -0700)]
x86/csum: clean up `csum_partial' further

Commit 688eb8191b47 ("x86/csum: Improve performance of `csum_partial`")
ended up improving the code generation for the IP csum calculations, and
in particular special-casing the 40-byte case that is a hot case for
IPv6 headers.

It then had _another_ special case for the 64-byte unrolled loop, which
did two chains of 32-byte blocks, which allows modern CPU's to improve
performance by doing the chains in parallel thanks to renaming the carry
flag.

This just unifies the special cases and combines them into just one
single helper the 40-byte csum case, and replaces the 64-byte case by a
80-byte case that just does that single helper twice.  It avoids having
all these different versions of inline assembly, and actually improved
performance further in my tests.

There was never anything magical about the 64-byte unrolled case, even
though it happens to be a common size (and typically is the cacheline
size).

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 months agox86/csum: Remove unnecessary odd handling
Noah Goldstein [Sun, 24 Sep 2023 14:35:49 +0000 (09:35 -0500)]
x86/csum: Remove unnecessary odd handling

The special case for odd aligned buffers is unnecessary and mostly
just adds overhead. Aligned buffers is the expectations, and even for
unaligned buffer, the only case that was helped is if the buffer was
1-byte from word aligned which is ~1/7 of the cases. Overall it seems
highly unlikely to be worth to extra branch.

It was left in the previous perf improvement patch because I was
erroneously comparing the exact output of `csum_partial(...)`, but
really we only need `csum_fold(csum_partial(...))` to match so its
safe to remove.

All csum kunit tests pass.

Signed-off-by: Noah Goldstein <goldstein.w.n@gmail.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: David Laight <david.laight@aculab.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
8 months agoMerge tag 'platform-drivers-x86-v6.7-7' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Thu, 4 Jan 2024 18:58:51 +0000 (10:58 -0800)]
Merge tag 'platform-drivers-x86-v6.7-7' of git://git./linux/kernel/git/pdx86/platform-drivers-x86

Pull x86 platform driver fix from Ilpo Järvinen:
 "Unfortunately the P2SB deadlock fix broke some older HW and we need
  some time to figure out the best way to fix the issue so reverting the
  deadlock fix for now"

* tag 'platform-drivers-x86-v6.7-7' of git://git.kernel.org/pub/scm/linux/kernel/git/pdx86/platform-drivers-x86:
  Revert "platform/x86: p2sb: Allow p2sb_bar() calls during PCI device probe"

8 months agoMerge tag 'sound-6.7-final' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai...
Linus Torvalds [Thu, 4 Jan 2024 18:51:16 +0000 (10:51 -0800)]
Merge tag 'sound-6.7-final' of git://git./linux/kernel/git/tiwai/sound

Pull sound fixes from Takashi Iwai:
 "It became more than wished, partly because of vacations. But all
  changes are fairly device-specific and should be safe to apply:

   - A regression fix for Oops at ASoC HD-audio probe

   - A series of TAS2781 HD-audio codec fixes

   - A random build regression fix with SPI helpers

   - Minor endianness fix for USB-audio mixer code

   - ASoC FSL driver error handling fix

   - ASoC Mediatek driver register fix

   - A series of ASoC meson g12a driver fixes

   - A few usual HD-audio oneliner quirks"

* tag 'sound-6.7-final' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound:
  ALSA: hda/realtek: Fix mute and mic-mute LEDs for HP ProBook 440 G6
  ASoC: meson: g12a-tohdmitx: Fix event generation for S/PDIF mux
  ASoC: meson: g12a-toacodec: Fix event generation
  ASoC: meson: g12a-tohdmitx: Validate written enum values
  ASoC: meson: g12a-toacodec: Validate written enum values
  ASoC: SOF: Intel: hda-codec: Delay the codec device registration
  ALSA: hda: cs35l41: fix building without CONFIG_SPI
  ALSA: hda/realtek: fix mute/micmute LEDs for a HP ZBook
  ALSA: hda/realtek: enable SND_PCI_QUIRK for hp pavilion 14-ec1xxx series
  ASoC: mediatek: mt8186: fix AUD_PAD_TOP register and offset
  ALSA: scarlett2: Convert meter levels from little-endian
  ALSA: hda/tas2781: remove sound controls in unbind
  ALSA: hda/tas2781: move set_drv_data outside tasdevice_init
  ALSA: hda/tas2781: fix typos in comment
  ALSA: hda/tas2781: do not use regcache
  ASoC: fsl_rpmsg: Fix error handler with pm_runtime_enable

8 months agoMerge tag 'drm-fixes-2024-01-04' of git://anongit.freedesktop.org/drm/drm
Linus Torvalds [Thu, 4 Jan 2024 18:48:03 +0000 (10:48 -0800)]
Merge tag 'drm-fixes-2024-01-04' of git://anongit.freedesktop.org/drm/drm

Pull drm fixes from Dave Airlie:
 "These were from over the holiday period, mainly i915, a couple of
  qaic, bridge and an mgag200.

  qaic:
   - fix GEM import
   - add quirk for soc version

  bridge:
   - parade-ps8640, ti-sn65dsi86: fix aux reads bounds

  mgag200:
   - fix gamma LUT init

  i915:
   - Fix bogus DPCD rev usage for DP phy test pattern setup
   - Fix handling of MMIO triggered reports in the OA buffer"

* tag 'drm-fixes-2024-01-04' of git://anongit.freedesktop.org/drm/drm:
  drm/i915/perf: Update handling of MMIO triggered reports
  drm/i915/dp: Fix passing the correct DPCD_REV for drm_dp_set_phy_test_pattern
  drm/mgag200: Fix gamma lut not initialized for G200ER, G200EV, G200SE
  drm/bridge: ps8640: Fix size mismatch warning w/ len
  drm/bridge: ti-sn65dsi86: Never store more than msg->size bytes in AUX xfer
  drm/bridge: parade-ps8640: Never store more than msg->size bytes in AUX xfer
  accel/qaic: Implement quirk for SOC_HW_VERSION
  accel/qaic: Fix GEM import path code

8 months agonet/tcp: Only produce AO/MD5 logs if there are any keys
Dmitry Safonov [Thu, 4 Jan 2024 13:42:39 +0000 (13:42 +0000)]
net/tcp: Only produce AO/MD5 logs if there are any keys

User won't care about inproper hash options in the TCP header if they
don't use neither TCP-AO nor TCP-MD5. Yet, those logs can add up in
syslog, while not being a real concern to the host admin:
> kernel: TCP: TCP segment has incorrect auth options set for XX.20.239.12.54681->XX.XX.90.103.80 [S]

Keep silent and avoid logging when there aren't any keys in the system.

Side-note: I also defined static_branch_tcp_*() helpers to avoid more
ifdeffery, going to remove more ifdeffery further with their help.

Reported-by: Christian Kujau <lists@nerdbynature.de>
Closes: https://lore.kernel.org/all/f6b59324-1417-566f-a976-ff2402718a8d@nerdbynature.de/
Signed-off-by: Dmitry Safonov <dima@arista.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Fixes: 2717b5adea9e ("net/tcp: Add tcp_hash_fail() ratelimited logs")
Link: https://lore.kernel.org/r/20240104-tcp_hash_fail-logs-v1-1-ff3e1f6f9e72@arista.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
8 months agoMerge branch '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue
Jakub Kicinski [Thu, 4 Jan 2024 16:19:14 +0000 (08:19 -0800)]
Merge branch '40GbE' of git://git./linux/kernel/git/tnguy/net-queue

Tony Nguyen says:

====================
Intel Wired LAN Driver Updates 2024-01-03 (i40e, ice, igc)

This series contains updates to i40e, ice, and igc drivers.

Ke Xiao fixes use after free for unicast filters on i40e.

Andrii restores VF MSI-X flag after PCI reset on i40e.

Paul corrects admin queue link status structure to fulfill firmware
expectations for ice.

Rodrigo Cataldo corrects value used for hicredit calculations on igc.

* '40GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue:
  igc: Fix hicredit calculation
  ice: fix Get link status data length
  i40e: Restore VF MSI-X state during PCI reset
  i40e: fix use-after-free in i40e_aqc_add_filters()
====================

Link: https://lore.kernel.org/r/20240103193254.822968-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
8 months agonet: Implement missing SO_TIMESTAMPING_NEW cmsg support
Thomas Lange [Thu, 4 Jan 2024 08:57:44 +0000 (09:57 +0100)]
net: Implement missing SO_TIMESTAMPING_NEW cmsg support

Commit 9718475e6908 ("socket: Add SO_TIMESTAMPING_NEW") added the new
socket option SO_TIMESTAMPING_NEW. However, it was never implemented in
__sock_cmsg_send thus breaking SO_TIMESTAMPING cmsg for platforms using
SO_TIMESTAMPING_NEW.

Fixes: 9718475e6908 ("socket: Add SO_TIMESTAMPING_NEW")
Link: https://lore.kernel.org/netdev/6a7281bf-bc4a-4f75-bb88-7011908ae471@app.fastmail.com/
Signed-off-by: Thomas Lange <thomas@corelatus.se>
Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://lore.kernel.org/r/20240104085744.49164-1-thomas@corelatus.se
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
8 months agoRevert "platform/x86: p2sb: Allow p2sb_bar() calls during PCI device probe"
Shin'ichiro Kawasaki [Thu, 4 Jan 2024 11:40:50 +0000 (20:40 +0900)]
Revert "platform/x86: p2sb: Allow p2sb_bar() calls during PCI device probe"

This reverts commit b28ff7a7c3245d7f62acc20f15b4361292fe4117.

The commit introduced P2SB device scan and resource cache during the
boot process to avoid deadlock. But it caused detection failure of
IDE controllers on old systems [1]. The IDE controllers on old systems
and P2SB devices on newer systems have same PCI DEVFN. It is suspected
the confusion between those two is the failure cause. Revert the change
at this moment until the proper solution gets ready.

Link: https://lore.kernel.org/platform-driver-x86/CABq1_vjfyp_B-f4LAL6pg394bP6nDFyvg110TOLHHb0x4aCPeg@mail.gmail.com/T/#m07b30468d9676fc5e3bb2122371121e4559bb383
Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Link: https://lore.kernel.org/r/20240104114050.3142690-1-shinichiro.kawasaki@wdc.com
Reviewed-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
8 months agoKVM: x86/pmu: fix masking logic for MSR_CORE_PERF_GLOBAL_CTRL
Paolo Bonzini [Thu, 4 Jan 2024 15:15:17 +0000 (16:15 +0100)]
KVM: x86/pmu: fix masking logic for MSR_CORE_PERF_GLOBAL_CTRL

When commit c59a1f106f5c ("KVM: x86/pmu: Add IA32_PEBS_ENABLE
MSR emulation for extended PEBS") switched the initialization of
cpuc->guest_switch_msrs to use compound literals, it screwed up
the boolean logic:

+ u64 pebs_mask = cpuc->pebs_enabled & x86_pmu.pebs_capable;
...
- arr[0].guest = intel_ctrl & ~cpuc->intel_ctrl_host_mask;
- arr[0].guest &= ~(cpuc->pebs_enabled & x86_pmu.pebs_capable);
+               .guest = intel_ctrl & (~cpuc->intel_ctrl_host_mask | ~pebs_mask),

Before the patch, the value of arr[0].guest would have been intel_ctrl &
~cpuc->intel_ctrl_host_mask & ~pebs_mask.  The intent is to always treat
PEBS events as host-only because, while the guest runs, there is no way
to tell the processor about the virtual address where to put PEBS records
intended for the host.

Unfortunately, the new expression can be expanded to

(intel_ctrl & ~cpuc->intel_ctrl_host_mask) | (intel_ctrl & ~pebs_mask)

which makes no sense; it includes any bit that isn't *both* marked as
exclude_guest and using PEBS.  So, reinstate the old logic.  Another
way to write it could be "intel_ctrl & ~(cpuc->intel_ctrl_host_mask |
pebs_mask)", presumably the intention of the author of the faulty.
However, I personally find the repeated application of A AND NOT B to
be a bit more readable.

This shows up as guest failures when running concurrent long-running
perf workloads on the host, and was reported to happen with rcutorture.
All guests on a given host would die simultaneously with something like an
instruction fault or a segmentation violation.

Reported-by: Paul E. McKenney <paulmck@kernel.org>
Analyzed-by: Sean Christopherson <seanjc@google.com>
Tested-by: Paul E. McKenney <paulmck@kernel.org>
Cc: stable@vger.kernel.org
Fixes: c59a1f106f5c ("KVM: x86/pmu: Add IA32_PEBS_ENABLE MSR emulation for extended PEBS")
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
8 months agodrm/amd/pm: Use gpu_metrics_v1_5 for SMUv13.0.6
Asad Kamal [Wed, 20 Dec 2023 09:32:48 +0000 (17:32 +0800)]
drm/amd/pm: Use gpu_metrics_v1_5 for SMUv13.0.6

Use gpu_metrics_v1_5 for SMUv13.0.6 to fill
gpu metric info

Signed-off-by: Asad Kamal <asad.kamal@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Le Ma <le.ma@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
8 months agodrm/amd/pm: Add gpu_metrics_v1_5
Asad Kamal [Wed, 20 Dec 2023 09:21:48 +0000 (17:21 +0800)]
drm/amd/pm: Add gpu_metrics_v1_5

Add new gpu_metrics_v1_5 to acquire vcn/jpeg activity
& pcie nak error counters

Signed-off-by: Asad Kamal <asad.kamal@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Le Ma <le.ma@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
8 months agodrm/amd/pm: Add mem_busy_percent for GCv9.4.3 apu
Asad Kamal [Fri, 22 Dec 2023 10:24:20 +0000 (18:24 +0800)]
drm/amd/pm: Add mem_busy_percent for GCv9.4.3 apu

Expose sysfs entry mem_busy_percent for GC version
9.4.3 APU system

Signed-off-by: Asad Kamal <asad.kamal@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
8 months agodrm/amd/display: Fix sending VSC (+ colorimetry) packets for DP/eDP displays without PSR
Joshua Ashton [Mon, 1 Jan 2024 18:28:22 +0000 (18:28 +0000)]
drm/amd/display: Fix sending VSC (+ colorimetry) packets for DP/eDP displays without PSR

The check for sending the vsc infopacket to the display was gated behind
PSR (Panel Self Refresh) being enabled.

The vsc infopacket also contains the colorimetry (specifically the
container color gamut) information for the stream on modern DP.

PSR is typically only supported on mobile phone eDP displays, thus this
was not getting sent for typical desktop monitors or TV screens.

This functionality is needed for proper HDR10 functionality on DP as it
wants BT2020 RGB/YCbCr for the container color space.

Cc: stable@vger.kernel.org
Cc: Harry Wentland <harry.wentland@amd.com>
Cc: Xaver Hugl <xaver.hugl@gmail.com>
Cc: Melissa Wen <mwen@igalia.com>
Fixes: 15f9dfd545a1 ("drm/amd/display: Register Colorspace property for DP and HDMI")
Tested-by: Simon Berz <simon@berz.me>
Tested-by: Xaver Hugl <xaver.hugl@kde.org>
Signed-off-by: Joshua Ashton <joshua@froggi.es>
Signed-off-by: Hamza Mahfooz <hamza.mahfooz@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
8 months agodrm/amdgpu: skip gpu_info fw loading on navi12
Alex Deucher [Wed, 20 Dec 2023 17:36:08 +0000 (12:36 -0500)]
drm/amdgpu: skip gpu_info fw loading on navi12

It's no longer required.

Link: https://gitlab.freedesktop.org/drm/amd/-/issues/2318
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
8 months agodrm/amd/display: add nv12 bounding box
Alex Deucher [Wed, 20 Dec 2023 17:33:45 +0000 (12:33 -0500)]
drm/amd/display: add nv12 bounding box

This was included in gpu_info firmware, move it into the
driver for consistency with other nv1x parts.

Link: https://gitlab.freedesktop.org/drm/amd/-/issues/2318
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
8 months agobnxt_en: Remove mis-applied code from bnxt_cfg_ntp_filters()
Michael Chan [Thu, 4 Jan 2024 00:59:24 +0000 (16:59 -0800)]
bnxt_en: Remove mis-applied code from bnxt_cfg_ntp_filters()

The 2 lines to check for the BNXT_HWRM_PF_UNLOAD_SP_EVENT bit was
mis-applied to bnxt_cfg_ntp_filters() and should have been applied to
bnxt_sp_task().

Fixes: 19241368443f ("bnxt_en: Send PF driver unload notification to all VFs.")
Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 months agonet: ravb: Wait for operating mode to be applied
Claudiu Beznea [Wed, 3 Jan 2024 08:13:53 +0000 (10:13 +0200)]
net: ravb: Wait for operating mode to be applied

CSR.OPS bits specify the current operating mode and (according to
documentation) they are updated by HW when the operating mode change
request is processed. To comply with this check CSR.OPS before proceeding.

Commit introduces ravb_set_opmode() that does all the necessities for
setting the operating mode (set CCC.OPC (and CCC.GAC, CCC.CSEL, if any) and
wait for CSR.OPS) and call it where needed. This should comply with all the
HW manuals requirements as different manual variants specify that different
modes need to be checked in CSR.OPS when setting CCC.OPC.

If gPTP active in config mode is supported and it needs to be enabled, the
CCC.GAC and CCC.CSEL needs to be configured along with CCC.OPC in the same
write access. For this, ravb_set_opmode() allows passing GAC and CSEL as
part of opmode and the function updates accordingly CCC register.

Fixes: c156633f1353 ("Renesas Ethernet AVB driver proper")
Signed-off-by: Claudiu Beznea <claudiu.beznea.uj@bp.renesas.com>
Reviewed-by: Sergey Shtylyov <s.shtylyov@omp.ru>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 months agoasix: Add check for usbnet_get_endpoints
Chen Ni [Wed, 3 Jan 2024 03:35:34 +0000 (03:35 +0000)]
asix: Add check for usbnet_get_endpoints

Add check for usbnet_get_endpoints() and return the error if it fails
in order to transfer the error.

Fixes: 16626b0cc3d5 ("asix: Add a new driver for the AX88172A")
Signed-off-by: Chen Ni <nichen@iscas.ac.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 months agoocteontx2-af: Re-enable MAC TX in otx2_stop processing
Naveen Mamindlapalli [Tue, 2 Jan 2024 14:14:00 +0000 (19:44 +0530)]
octeontx2-af: Re-enable MAC TX in otx2_stop processing

During QoS scheduling testing with multiple strict priority flows, the
netdev tx watchdog timeout routine is invoked when a low priority QoS
queue doesn't get a chance to transmit the packets because other high
priority flows are completely subscribing the transmit link. The netdev
tx watchdog timeout routine will stop MAC RX and TX functionality in
otx2_stop() routine before cleanup of HW TX queues which results in SMQ
flush errors because the packets belonging to low priority queues will
never gets flushed since MAC TX is disabled. This patch fixes the issue
by re-enabling MAC TX to ensure the packets in HW pipeline gets flushed
properly.

Fixes: a7faa68b4e7f ("octeontx2-af: Start/Stop traffic in CGX along with NPC")
Signed-off-by: Naveen Mamindlapalli <naveenm@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 months agoocteontx2-af: Always configure NIX TX link credits based on max frame size
Naveen Mamindlapalli [Tue, 2 Jan 2024 09:56:43 +0000 (15:26 +0530)]
octeontx2-af: Always configure NIX TX link credits based on max frame size

Currently the NIX TX link credits are initialized based on the max frame
size that can be transmitted on a link but when the MTU is changed, the
NIX TX link credits are reprogrammed by the SW based on the new MTU value.
Since SMQ max packet length is programmed to max frame size by default,
there is a chance that NIX TX may stall while sending a max frame sized
packet on the link with insufficient credits to send the packet all at
once. This patch avoids stall issue by not changing the link credits
dynamically when the MTU is changed.

Fixes: 1c74b89171c3 ("octeontx2-af: Wait for TX link idle for credits change")
Signed-off-by: Naveen Mamindlapalli <naveenm@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
Signed-off-by: Nithin Kumar Dabilpuram <ndabilpuram@marvell.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
8 months agoALSA: hda/realtek: Fix mute and mic-mute LEDs for HP ProBook 440 G6
Siddhesh Dharme [Thu, 4 Jan 2024 06:07:36 +0000 (11:37 +0530)]
ALSA: hda/realtek: Fix mute and mic-mute LEDs for HP ProBook 440 G6

LEDs in 'HP ProBook 440 G6' laptop are controlled by ALC236 codec.
Enable already existing quirk 'ALC236_FIXUP_HP_MUTE_LED_MICMUTE_VREF'
to fix mute and mic-mute LEDs.

Signed-off-by: Siddhesh Dharme <siddheshdharme18@gmail.com>
Cc: <stable@vger.kernel.org>
Link: https://lore.kernel.org/r/20240104060736.5149-1-siddheshdharme18@gmail.com
Signed-off-by: Takashi Iwai <tiwai@suse.de>
8 months agoMerge tag 'asoc-fix-v6.7-rc8' of https://git.kernel.org/pub/scm/linux/kernel/git...
Takashi Iwai [Thu, 4 Jan 2024 07:58:30 +0000 (08:58 +0100)]
Merge tag 'asoc-fix-v6.7-rc8' of https://git./linux/kernel/git/broonie/sound into for-linus

ASoC: Fixes for v6.7

I recently got a LibreTech Sapphire board for my CI and while
integrating it found and fixed some issues, including crashes for the
enum validation.  There's also a couple of patches adding quirks for
another x86 laptop from Hans and an error handling fix for the Freescale
rpmsg driver.

8 months agox86/kprobes: fix incorrect return address calculation in kprobe_emulate_call_indirect
Jinghao Jia [Tue, 2 Jan 2024 23:33:45 +0000 (17:33 -0600)]
x86/kprobes: fix incorrect return address calculation in kprobe_emulate_call_indirect

kprobe_emulate_call_indirect currently uses int3_emulate_call to emulate
indirect calls. However, int3_emulate_call always assumes the size of
the call to be 5 bytes when calculating the return address. This is
incorrect for register-based indirect calls in x86, which can be either
2 or 3 bytes depending on whether REX prefix is used. At kprobe runtime,
the incorrect return address causes control flow to land onto the wrong
place after return -- possibly not a valid instruction boundary. This
can lead to a panic like the following:

[    7.308204][    C1] BUG: unable to handle page fault for address: 000000000002b4d8
[    7.308883][    C1] #PF: supervisor read access in kernel mode
[    7.309168][    C1] #PF: error_code(0x0000) - not-present page
[    7.309461][    C1] PGD 0 P4D 0
[    7.309652][    C1] Oops: 0000 [#1] SMP
[    7.309929][    C1] CPU: 1 PID: 0 Comm: swapper/1 Not tainted 6.7.0-rc5-trace-for-next #6
[    7.310397][    C1] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.0-20220807_005459-localhost 04/01/2014
[    7.311068][    C1] RIP: 0010:__common_interrupt+0x52/0xc0
[    7.311349][    C1] Code: 01 00 4d 85 f6 74 39 49 81 fe 00 f0 ff ff 77 30 4c 89 f7 4d 8b 5e 68 41 ba 91 76 d8 42 45 03 53 fc 74 02 0f 0b cc ff d3 65 48 <8b> 05 30 c7 ff 7e 65 4c 89 3d 28 c7 ff 7e 5b 41 5c 41 5e 41 5f c3
[    7.312512][    C1] RSP: 0018:ffffc900000e0fd0 EFLAGS: 00010046
[    7.312899][    C1] RAX: 0000000000000001 RBX: 0000000000000023 RCX: 0000000000000001
[    7.313334][    C1] RDX: 00000000000003cd RSI: 0000000000000001 RDI: ffff888100d302a4
[    7.313702][    C1] RBP: 0000000000000001 R08: 0ef439818636191f R09: b1621ff338a3b482
[    7.314146][    C1] R10: ffffffff81e5127b R11: ffffffff81059810 R12: 0000000000000023
[    7.314509][    C1] R13: 0000000000000000 R14: ffff888100d30200 R15: 0000000000000000
[    7.314951][    C1] FS:  0000000000000000(0000) GS:ffff88813bc80000(0000) knlGS:0000000000000000
[    7.315396][    C1] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[    7.315691][    C1] CR2: 000000000002b4d8 CR3: 0000000003028003 CR4: 0000000000370ef0
[    7.316153][    C1] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[    7.316508][    C1] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[    7.316948][    C1] Call Trace:
[    7.317123][    C1]  <IRQ>
[    7.317279][    C1]  ? __die_body+0x64/0xb0
[    7.317482][    C1]  ? page_fault_oops+0x248/0x370
[    7.317712][    C1]  ? __wake_up+0x96/0xb0
[    7.317964][    C1]  ? exc_page_fault+0x62/0x130
[    7.318211][    C1]  ? asm_exc_page_fault+0x22/0x30
[    7.318444][    C1]  ? __cfi_native_send_call_func_single_ipi+0x10/0x10
[    7.318860][    C1]  ? default_idle+0xb/0x10
[    7.319063][    C1]  ? __common_interrupt+0x52/0xc0
[    7.319330][    C1]  common_interrupt+0x78/0x90
[    7.319546][    C1]  </IRQ>
[    7.319679][    C1]  <TASK>
[    7.319854][    C1]  asm_common_interrupt+0x22/0x40
[    7.320082][    C1] RIP: 0010:default_idle+0xb/0x10
[    7.320309][    C1] Code: 4c 01 c7 4c 29 c2 e9 72 ff ff ff cc cc cc cc 90 90 90 90 90 90 90 90 90 90 90 b8 0c 67 40 a5 66 90 0f 00 2d 09 b9 3b 00 fb f4 <fa> c3 0f 1f 00 90 90 90 90 90 90 90 90 90 90 90 b8 0c 67 40 a5 e9
[    7.321449][    C1] RSP: 0018:ffffc9000009bee8 EFLAGS: 00000256
[    7.321808][    C1] RAX: ffff88813bca8b68 RBX: 0000000000000001 RCX: 000000000001ef0c
[    7.322227][    C1] RDX: 0000000000000000 RSI: 0000000000000001 RDI: 000000000001ef0c
[    7.322656][    C1] RBP: ffffc9000009bef8 R08: 8000000000000000 R09: 00000000000008c2
[    7.323083][    C1] R10: 0000000000000000 R11: ffffffff81058e70 R12: 0000000000000000
[    7.323530][    C1] R13: ffff8881002b30c0 R14: 0000000000000000 R15: 0000000000000000
[    7.323948][    C1]  ? __cfi_lapic_next_deadline+0x10/0x10
[    7.324239][    C1]  default_idle_call+0x31/0x50
[    7.324464][    C1]  do_idle+0xd3/0x240
[    7.324690][    C1]  cpu_startup_entry+0x25/0x30
[    7.324983][    C1]  start_secondary+0xb4/0xc0
[    7.325217][    C1]  secondary_startup_64_no_verify+0x179/0x17b
[    7.325498][    C1]  </TASK>
[    7.325641][    C1] Modules linked in:
[    7.325906][    C1] CR2: 000000000002b4d8
[    7.326104][    C1] ---[ end trace 0000000000000000 ]---
[    7.326354][    C1] RIP: 0010:__common_interrupt+0x52/0xc0
[    7.326614][    C1] Code: 01 00 4d 85 f6 74 39 49 81 fe 00 f0 ff ff 77 30 4c 89 f7 4d 8b 5e 68 41 ba 91 76 d8 42 45 03 53 fc 74 02 0f 0b cc ff d3 65 48 <8b> 05 30 c7 ff 7e 65 4c 89 3d 28 c7 ff 7e 5b 41 5c 41 5e 41 5f c3
[    7.327570][    C1] RSP: 0018:ffffc900000e0fd0 EFLAGS: 00010046
[    7.327910][    C1] RAX: 0000000000000001 RBX: 0000000000000023 RCX: 0000000000000001
[    7.328273][    C1] RDX: 00000000000003cd RSI: 0000000000000001 RDI: ffff888100d302a4
[    7.328632][    C1] RBP: 0000000000000001 R08: 0ef439818636191f R09: b1621ff338a3b482
[    7.329223][    C1] R10: ffffffff81e5127b R11: ffffffff81059810 R12: 0000000000000023
[    7.329780][    C1] R13: 0000000000000000 R14: ffff888100d30200 R15: 0000000000000000
[    7.330193][    C1] FS:  0000000000000000(0000) GS:ffff88813bc80000(0000) knlGS:0000000000000000
[    7.330632][    C1] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[    7.331050][    C1] CR2: 000000000002b4d8 CR3: 0000000003028003 CR4: 0000000000370ef0
[    7.331454][    C1] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[    7.331854][    C1] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[    7.332236][    C1] Kernel panic - not syncing: Fatal exception in interrupt
[    7.332730][    C1] Kernel Offset: disabled
[    7.333044][    C1] ---[ end Kernel panic - not syncing: Fatal exception in interrupt ]---

The relevant assembly code is (from objdump, faulting address
highlighted):

ffffffff8102ed9d:       41 ff d3                  call   *%r11
ffffffff8102eda0:       65 48 <8b> 05 30 c7 ff    mov    %gs:0x7effc730(%rip),%rax

The emulation incorrectly sets the return address to be ffffffff8102ed9d
+ 0x5 = ffffffff8102eda2, which is the 8b byte in the middle of the next
mov. This in turn causes incorrect subsequent instruction decoding and
eventually triggers the page fault above.

Instead of invoking int3_emulate_call, perform push and jmp emulation
directly in kprobe_emulate_call_indirect. At this point we can obtain
the instruction size from p->ainsn.size so that we can calculate the
correct return address.

Link: https://lore.kernel.org/all/20240102233345.385475-1-jinghao7@illinois.edu/
Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step")
Cc: stable@vger.kernel.org
Signed-off-by: Jinghao Jia <jinghao7@illinois.edu>
Signed-off-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
8 months agoMerge tag 'nf-24-01-03' of git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf
Jakub Kicinski [Thu, 4 Jan 2024 02:05:23 +0000 (18:05 -0800)]
Merge tag 'nf-24-01-03' of git://git./linux/kernel/git/netfilter/nf

Pablo Neira Ayuso says:

====================
Netfilter fixes for net

The following patchset contains Netfilter fixes for net:

1) Fix nat packets in the related state in OVS, from Brad Cowie.

2) Drop chain reference counter on error path in case chain binding
   fails.

* tag 'nf-24-01-03' of git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf:
  netfilter: nft_immediate: drop chain reference counter on error
  netfilter: nf_nat: fix action not being set for all ct states
====================

Link: https://lore.kernel.org/r/20240103113001.137936-1-pablo@netfilter.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
8 months agoMerge tag 'drm-misc-fixes-2024-01-03' of git://anongit.freedesktop.org/drm/drm-misc...
Dave Airlie [Thu, 4 Jan 2024 01:18:31 +0000 (11:18 +1000)]
Merge tag 'drm-misc-fixes-2024-01-03' of git://anongit.freedesktop.org/drm/drm-misc into drm-fixes

drm-misc-fixes for v6.7 final:
- 2 small qaic fixes.
- Fixes for overflow in aux xfer.
- Fix uninitialised gamma lut in gmag200.
- Small compiler warning fix for backports of a ps8640 fix.

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/9ba866b4-3144-47a9-a2c0-7313c67249d7@linux.intel.com
8 months agoMerge branch '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue
Jakub Kicinski [Thu, 4 Jan 2024 01:02:45 +0000 (17:02 -0800)]
Merge branch '1GbE' of git://git./linux/kernel/git/tnguy/net-queue

Tony Nguyen says:

====================
Intel Wired LAN Driver Updates 2023-12-27 (igc)

This series contains updates to igc driver only.

Kurt Kanzenbach resolves issues around VLAN ntuple rules; correctly
reporting back added rules and checking for valid values.

* '1GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue:
  igc: Check VLAN EtherType mask
  igc: Check VLAN TCI mask
  igc: Report VLAN EtherType matching back to user
====================

Link: https://lore.kernel.org/r/20231227210041.3035055-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
8 months agoMerge branch '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net...
Jakub Kicinski [Thu, 4 Jan 2024 01:00:15 +0000 (17:00 -0800)]
Merge branch '100GbE' of git://git./linux/kernel/git/tnguy/net-queue

Tony Nguyen says:

====================
Intel Wired LAN Driver Updates 2023-12-27 (ice, i40e)

This series contains updates to ice and i40e drivers.

Katarzyna changes message to no longer be reported as error under
certain conditions as it can be expected on ice.

Ngai-Mint ensures VSI is always closed when stopping interface to
prevent NULL pointer dereference for ice.

Arkadiusz corrects reporting of phase offset value for ice.

Sudheer corrects checking on ADQ filters to prevent invalid values on
i40e.

* '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue:
  i40e: Fix filter input checks to prevent config with invalid values
  ice: dpll: fix phase offset value
  ice: Shut down VSI with "link-down-on-close" enabled
  ice: Fix link_down_on_close message
====================

Link: https://lore.kernel.org/r/20231227182541.3033124-1-anthony.l.nguyen@intel.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>