Currently, io_uring_unreg_ringfd() (which cleans up registered rings) is
only called on exit, but __io_uring_free (which frees the tctx in which the
registered ring pointers are stored) is also called on execve (via
begin_new_exec -> io_uring_task_cancel -> __io_uring_cancel ->
io_uring_cancel_generic -> __io_uring_free).
This means: A process going through execve while having registered rings
will leak references to the rings' `struct file`.
Fix it by zapping registered rings on execve(). This is implemented by
moving the io_uring_unreg_ringfd() from io_uring_files_cancel() into its
callee __io_uring_cancel(), which is called from io_uring_task_cancel() on
execve.
This could probably be exploited *on 32-bit kernels* by leaking 2^32
references to the same ring, because the file refcount is stored in a
pointer-sized field and get_file() doesn't have protection against
refcount overflow, just a WARN_ONCE(); but on 64-bit it should have no
impact beyond a memory leak.
Cc: stable@vger.kernel.org Fixes: e7a6c00dc77a ("io_uring: add support for registering ring file descriptors") Signed-off-by: Jann Horn <jannh@google.com> Link: https://lore.kernel.org/r/20241218-uring-reg-ring-cleanup-v1-1-8f63e999045b@google.com Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Currently, LoongArch LLVM does not support the constraint "o" and no plan
to support it, it only supports the similar constraint "m", so change the
constraints from "nor" in the "else" case to arch-specific "nmr" to avoid
the build error such as "unexpected asm memory constraint" for LoongArch.
The sysctl tests for vm.memfd_noexec rely on the kernel to support PID
namespaces (i.e. the kernel is built with CONFIG_PID_NS=y). If the
kernel the test runs on does not support PID namespaces, the first sysctl
test will fail when attempting to spawn a new thread in a new PID
namespace, abort the test, preventing the remaining tests from being run.
This is not desirable, as not all kernels need PID namespaces, but can
still use the other features provided by memfd. Therefore, only run the
sysctl tests if the kernel supports PID namespaces. Otherwise, skip those
tests and emit an informative message to let the user know why the sysctl
tests are not being run.
Link: https://lkml.kernel.org/r/20241205192943.3228757-1-isaacmanjarres@google.com Fixes: 11f75a01448f ("selftests/memfd: add tests for MFD_NOEXEC_SEAL MFD_EXEC") Signed-off-by: Isaac J. Manjarres <isaacmanjarres@google.com> Reviewed-by: Jeff Xu <jeffxu@google.com> Cc: Suren Baghdasaryan <surenb@google.com> Cc: Kalesh Singh <kaleshsingh@google.com> Cc: <stable@vger.kernel.org> [6.6+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The TP_printk() portion of a trace event is executed at the time a event
is read from the trace. This can happen seconds, minutes, hours, days,
months, years possibly later since the event was recorded. If the print
format contains a dereference to a string via "%s", and that string was
allocated, there's a chance that string could be freed before it is read
by the trace file.
To protect against such bugs, there are two functions that verify the
event. The first one is test_event_printk(), which is called when the
event is created. It reads the TP_printk() format as well as its arguments
to make sure nothing may be dereferencing a pointer that was not copied
into the ring buffer along with the event. If it is, it will trigger a
WARN_ON().
For strings that use "%s", it is not so easy. The string may not reside in
the ring buffer but may still be valid. Strings that are static and part
of the kernel proper which will not be freed for the life of the running
system, are safe to dereference. But to know if it is a pointer to a
static string or to something on the heap can not be determined until the
event is triggered.
This brings us to the second function that tests for the bad dereferencing
of strings, trace_check_vprintf(). It would walk through the printf format
looking for "%s", and when it finds it, it would validate that the pointer
is safe to read. If not, it would produces a WARN_ON() as well and write
into the ring buffer "[UNSAFE-MEMORY]".
The problem with this is how it used va_list to have vsnprintf() handle
all the cases that it didn't need to check. Instead of re-implementing
vsnprintf(), it would make a copy of the format up to the %s part, and
call vsnprintf() with the current va_list ap variable, where the ap would
then be ready to point at the string in question.
For architectures that passed va_list by reference this was possible. For
architectures that passed it by copy it was not. A test_can_verify()
function was used to differentiate between the two, and if it wasn't
possible, it would disable it.
Even for architectures where this was feasible, it was a stretch to rely
on such a method that is undocumented, and could cause issues later on
with new optimizations of the compiler.
Instead, the first function test_event_printk() was updated to look at
"%s" as well. If the "%s" argument is a pointer outside the event in the
ring buffer, it would find the field type of the event that is the problem
and mark the structure with a new flag called "needs_test". The event
itself will be marked by TRACE_EVENT_FL_TEST_STR to let it be known that
this event has a field that needs to be verified before the event can be
printed using the printf format.
When the event fields are created from the field type structure, the
fields would copy the field type's "needs_test" value.
Finally, before being printed, a new function ignore_event() is called
which will check if the event has the TEST_STR flag set (if not, it
returns false). If the flag is set, it then iterates through the events
fields looking for the ones that have the "needs_test" flag set.
Then it uses the offset field from the field structure to find the pointer
in the ring buffer event. It runs the tests to make sure that pointer is
safe to print and if not, it triggers the WARN_ON() and also adds to the
trace output that the event in question has an unsafe memory access.
The ignore_event() makes the trace_check_vprintf() obsolete so it is
removed.
Link: https://lore.kernel.org/all/CAHk-=wh3uOnqnZPpR0PeLZZtyWbZLboZ7cHLCKRWsocvs9Y7hQ@mail.gmail.com/ Cc: stable@vger.kernel.org Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Al Viro <viro@ZenIV.linux.org.uk> Cc: Linus Torvalds <torvalds@linux-foundation.org> Link: https://lore.kernel.org/20241217024720.848621576@goodmis.org Fixes: 5013f454a352c ("tracing: Add check of trace event print fmts for dereferencing pointers") Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The test_event_printk() code makes sure that when a trace event is
registered, any dereferenced pointers in from the event's TP_printk() are
pointing to content in the ring buffer. But currently it does not handle
"%s", as there's cases where the string pointer saved in the ring buffer
points to a static string in the kernel that will never be freed. As that
is a valid case, the pointer needs to be checked at runtime.
Currently the runtime check is done via trace_check_vprintf(), but to not
have to replicate everything in vsnprintf() it does some logic with the
va_list that may not be reliable across architectures. In order to get rid
of that logic, more work in the test_event_printk() needs to be done. Some
of the strings can be validated at this time when it is obvious the string
is valid because the string will be saved in the ring buffer content.
Do all the validation of strings in the ring buffer at boot in
test_event_printk(), and make sure that the field of the strings that
point into the kernel are accessible. This will allow adding checks at
runtime that will validate the fields themselves and not rely on paring
the TP_printk() format at runtime.
Cc: stable@vger.kernel.org Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Al Viro <viro@ZenIV.linux.org.uk> Cc: Linus Torvalds <torvalds@linux-foundation.org> Link: https://lore.kernel.org/20241217024720.685917008@goodmis.org Fixes: 5013f454a352c ("tracing: Add check of trace event print fmts for dereferencing pointers") Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The process_pointer() helper function looks to see if various trace event
macros are used. These macros are for storing data in the event. This
makes it safe to dereference as the dereference will then point into the
event on the ring buffer where the content of the data stays with the
event itself.
The test_event_printk() analyzes print formats of trace events looking for
cases where it may dereference a pointer that is not in the ring buffer
which can possibly be a bug when the trace event is read from the ring
buffer and the content of that pointer no longer exists.
The function needs to accurately go from one print format argument to the
next. It handles quotes and parenthesis that may be included in an
argument. When it finds the start of the next argument, it uses a simple
"c = strstr(fmt + i, ',')" to find the end of that argument!
In order to include "%s" dereferencing, it needs to process the entire
content of the print format argument and not just the content of the first
',' it finds. As there may be content like:
Which is an example of a full argument of an existing event. As the code
already handles finding the next print format argument, process the
argument at the end of it and not the start of it. This way it has both
the start of the argument as well as the end of it.
Add a helper function "process_pointer()" that will do the processing during
the loop as well as at the end. It also makes the code cleaner and easier
to read.
Cc: stable@vger.kernel.org Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Al Viro <viro@ZenIV.linux.org.uk> Cc: Linus Torvalds <torvalds@linux-foundation.org> Link: https://lore.kernel.org/20241217024720.362271189@goodmis.org Fixes: 5013f454a352c ("tracing: Add check of trace event print fmts for dereferencing pointers") Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Move pm_runtime_set_active() to ivpu_pm_init() so when
ivpu_ipc_send_receive_internal() is executed before ivpu_pm_enable()
it already has correct runtime state, even if last resume was
not successful.
Fixes: 8ed520ff4682 ("accel/ivpu: Move set autosuspend delay to HW specific code") Cc: stable@vger.kernel.org # v6.7+ Reviewed-by: Karol Wachowski <karol.wachowski@intel.com> Reviewed-by: Jeffrey Hugo <quic_jhugo@quicinc.com> Signed-off-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20241210130939.1575610-4-jacek.lawrynowicz@linux.intel.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Commit ef7134c7fc48 ("smb: client: Fix use-after-free of network namespace.")
fixed a netns UAF by manually enabled socket refcounting
(sk->sk_net_refcnt=1 and sock_inuse_add(net, 1)).
The reason the patch worked for that bug was because we now hold
references to the netns (get_net_track() gets a ref internally)
and they're properly released (internally, on __sk_destruct()),
but only because sk->sk_net_refcnt was set.
Problem:
(this happens regardless of CONFIG_NET_NS_REFCNT_TRACKER and regardless
if init_net or other)
Setting sk->sk_net_refcnt=1 *manually* and *after* socket creation is not
only out of cifs scope, but also technically wrong -- it's set conditionally
based on user (=1) vs kernel (=0) sockets. And net/ implementations
seem to base their user vs kernel space operations on it.
e.g. upon TCP socket close, the TCP timers are not cleared because
sk->sk_net_refcnt=1:
(cf. commit 151c9c724d05 ("tcp: properly terminate timers for kernel sockets"))
net/ipv4/tcp.c:
void tcp_close(struct sock *sk, long timeout)
{
lock_sock(sk);
__tcp_close(sk, timeout);
release_sock(sk);
if (!sk->sk_net_refcnt)
inet_csk_clear_xmit_timers_sync(sk);
sock_put(sk);
}
Which will throw a lockdep warning and then, as expected, deadlock on
tcp_write_timer().
A way to reproduce this is by running the reproducer from ef7134c7fc48
and then 'rmmod cifs'. A few seconds later, the deadlock/lockdep
warning shows up.
Fix:
We shouldn't mess with socket internals ourselves, so do not set
sk_net_refcnt manually.
Also change __sock_create() to sock_create_kern() for explicitness.
As for non-init_net network namespaces, we deal with it the best way
we can -- hold an extra netns reference for server->ssocket and drop it
when it's released. This ensures that the netns still exists whenever
we need to create/destroy server->ssocket, but is not directly tied to
it.
Fixes: ef7134c7fc48 ("smb: client: Fix use-after-free of network namespace.") Cc: stable@vger.kernel.org Signed-off-by: Enzo Matsumiya <ematsumiya@suse.de> Signed-off-by: Steve French <stfrench@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Use is_64_bit_hypercall() instead of is_64_bit_mode() to detect a 64-bit
hypercall when completing said hypercall. For guests with protected state,
e.g. SEV-ES and SEV-SNP, KVM must assume the hypercall was made in 64-bit
mode as the vCPU state needed to detect 64-bit mode is unavailable.
Hacking the sev_smoke_test selftest to generate a KVM_HC_MAP_GPA_RANGE
hypercall via VMGEXIT trips the WARN:
read_hv_sched_clock_tsc() assumes that the Hyper-V clock counter is
bigger than the variable hv_sched_clock_offset, which is cached during
early boot, but depending on the timing this assumption may be false
when a hibernated VM starts again (the clock counter starts from 0
again) and is resuming back (Note: hv_init_tsc_clocksource() is not
called during hibernation/resume); consequently,
read_hv_sched_clock_tsc() may return a negative integer (which is
interpreted as a huge positive integer since the return type is u64)
and new kernel messages are prefixed with huge timestamps before
read_hv_sched_clock_tsc() grows big enough (which typically takes
several seconds).
Fix the issue by saving the Hyper-V clock counter just before the
suspend, and using it to correct the hv_sched_clock_offset in
resume. This makes hv tsc page based sched_clock continuous and ensures
that post resume, it starts from where it left off during suspend.
Override x86_platform.save_sched_clock_state and
x86_platform.restore_sched_clock_state routines to correct this as soon
as possible.
Note: if Invariant TSC is available, the issue doesn't happen because
1) we don't register read_hv_sched_clock_tsc() for sched clock:
See commit e5313f1c5404 ("clocksource/drivers/hyper-v: Rework
clocksource and sched clock setup");
2) the common x86 code adjusts TSC similarly: see
__restore_processor_state() -> tsc_verify_tsc_adjust(true) and
x86_platform.restore_sched_clock_state().
Cc: stable@vger.kernel.org Fixes: 1349401ff1aa ("clocksource/drivers/hyper-v: Suspend/resume Hyper-V clocksource for hibernation") Co-developed-by: Dexuan Cui <decui@microsoft.com> Signed-off-by: Dexuan Cui <decui@microsoft.com> Signed-off-by: Naman Jain <namjain@linux.microsoft.com> Reviewed-by: Michael Kelley <mhklinux@outlook.com> Link: https://lore.kernel.org/r/20240917053917.76787-1-namjain@linux.microsoft.com Signed-off-by: Wei Liu <wei.liu@kernel.org>
Message-ID: <20240917053917.76787-1-namjain@linux.microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
hv_fcopy_uio_daemon.c:436:53: warning: '%s' directive output may be truncated
writing up to 14 bytes into a region of size 10 [-Wformat-truncation=]
436 | snprintf(uio_dev_path, sizeof(uio_dev_path), "/dev/%s", uio_name);
Also added 'static' for the array 'desc[]'.
Fixes: 82b0945ce2c2 ("tools: hv: Add new fcopy application based on uio driver") Cc: stable@vger.kernel.org # 6.10+ Signed-off-by: Dexuan Cui <decui@microsoft.com> Reviewed-by: Saurabh Sengar <ssengar@linux.microsoft.com> Link: https://lore.kernel.org/r/20240910004433.50254-1-decui@microsoft.com Signed-off-by: Wei Liu <wei.liu@kernel.org>
Message-ID: <20240910004433.50254-1-decui@microsoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This can happen because the KVP/VSS channel callback can be invoked
even before the channel is fully opened:
1) as soon as hv_kvp_init() -> hvutil_transport_init() creates
/dev/vmbus/hv_kvp, the kvp daemon can open the device file immediately and
register itself to the driver by writing a message KVP_OP_REGISTER1 to the
file (which is handled by kvp_on_msg() ->kvp_handle_handshake()) and
reading the file for the driver's response, which is handled by
hvt_op_read(), which calls hvt->on_read(), i.e. kvp_register_done().
2) the problem with kvp_register_done() is that it can cause the
channel callback to be called even before the channel is fully opened,
and when the channel callback is starting to run, util_probe()->
vmbus_open() may have not initialized the ringbuffer yet, so the
callback can hit the panic of NULL pointer dereference.
To reproduce the panic consistently, we can add a "ssleep(10)" for KVP in
__vmbus_open(), just before the first hv_ringbuffer_init(), and then we
unload and reload the driver hv_utils, and run the daemon manually within
the 10 seconds.
Fix the panic by reordering the steps in util_probe() so the char dev
entry used by the KVP or VSS daemon is not created until after
vmbus_open() has completed. This reordering prevents the race condition
from happening.
Reported-by: Dexuan Cui <decui@microsoft.com> Fixes: e0fa3e5e7df6 ("Drivers: hv: utils: fix a race on userspace daemons registration") Cc: stable@vger.kernel.org Signed-off-by: Michael Kelley <mhklinux@outlook.com> Acked-by: Wei Liu <wei.liu@kernel.org> Link: https://lore.kernel.org/r/20241106154247.2271-3-mhklinux@outlook.com Signed-off-by: Wei Liu <wei.liu@kernel.org>
Message-ID: <20241106154247.2271-3-mhklinux@outlook.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
A bug was discovered where the idle shadow stacks were not initialized
for offline CPUs when starting function graph tracer, and when they came
online they were not traced due to the missing shadow stack. To fix
this, the idle task shadow stack initialization was moved to using the
CPU hotplug callbacks. But it removed the initialization when the
function graph was enabled. The problem here is that the hotplug
callbacks are called when the CPUs come online, but the idle shadow
stack initialization only happens if function graph is currently
active. This caused the online CPUs to not get their shadow stack
initialized.
The idle shadow stack initialization still needs to be done when the
function graph is registered, as they will not be allocated if function
graph is not registered.
Use the helper function rather than reading it directly.
Reviewed-by: Yang Wang <kevinyang.wang@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 63bfd24088b42c6f55c2096bfc41b50213d419b2) Cc: stable@vger.kernel.org Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Use the helper function rather than reading it directly.
Reviewed-by: Yang Wang <kevinyang.wang@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit f1fd1d0f40272948aa6ab82a3a82ecbbc76dff53) Cc: stable@vger.kernel.org Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Use the helper function rather than reading it directly.
Reviewed-by: Yang Wang <kevinyang.wang@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 0ec43fbece784215d3c4469973e4556d70bce915) Cc: stable@vger.kernel.org Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
With uncoupling of physical and virtual address spaces population of
the identity mapping was changed to use the type POPULATE_IDENTITY
instead of POPULATE_DIRECT. This breaks DirectMap accounting:
Adjust all locations of update_page_count() in vmem.c to use
POPULATE_IDENTITY instead of POPULATE_DIRECT as well. With this
accounting is correct again:
[BUG]
There is a bug report in the mailing list where btrfs_run_delayed_refs()
failed to drop the ref count for logical 25870311358464 num_bytes 2113536.
[CAUSE]
There is no concrete evidence yet, but considering 0 -> 1 is also a
single bit flipped, it's possible that hardware memory bitflip is
involved, causing the on-disk extent tree to be corrupted.
[FIX]
To prevent us reading such corrupted extent item, or writing such
damaged extent item back to disk, enhance the handling of
BTRFS_EXTENT_DATA_REF_KEY and BTRFS_SHARED_DATA_REF_KEY keys for both
inlined and key items, to detect such 0 ref count and reject them.
CC: stable@vger.kernel.org # 5.4+ Link: https://lore.kernel.org/linux-btrfs/7c69dd49-c346-4806-86e7-e6f863a66f48@app.fastmail.com/ Reported-by: Frankie Fisher <frankie@terrorise.me.uk> Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Qu Wenruo <wqu@suse.com> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
to make decisions about whether or not to visit a node during snapshot
delete. This is because for normal subvolumes this is set to 0, and for
snapshots it's set to the creation generation. The idea being that if
the generation of the node is less than or equal to our creation
generation then we don't need to visit that node, because it doesn't
belong to us, we can simply drop our reference and move on.
However reloc roots don't have their generation stored in
root->root_key.offset, instead that is the objectid of their
corresponding fs root. This means we can incorrectly not walk into
nodes that need to be dropped when deleting a reloc root.
There are a variety of consequences to making the wrong choice in two
distinct areas.
visit_node_for_delete()
1. False positive. We think we are newer than the block when we really
aren't. We don't visit the node and drop our reference to the node
and carry on. This would result in leaked space.
2. False negative. We do decide to walk down into a block that we
should have just dropped our reference to. However this means that
the child node will have refs > 1, so we will switch to
UPDATE_BACKREF, and then the subsequent walk_down_proc() will notice
that btrfs_header_owner(node) != root->root_key.objectid and it'll
break out of the loop, and then walk_up_proc() will drop our reference,
so this appears to be ok.
do_walk_down()
1. False positive. We are in UPDATE_BACKREF and incorrectly decide that
we are done and don't need to update the backref for our lower nodes.
This is another case that simply won't happen with relocation, as we
only have to do UPDATE_BACKREF if the node below us was shared and
didn't have FULL_BACKREF set, and since we don't own that node
because we're a reloc root we actually won't end up in this case.
2. False negative. Again this is tricky because as described above, we
simply wouldn't be here from relocation, because we don't own any of
the nodes because we never set btrfs_header_owner() to the reloc root
objectid, and we always use FULL_BACKREF, we never actually need to
set FULL_BACKREF on any children.
Having spent a lot of time stressing relocation/snapshot delete recently
I've not seen this pop in practice. But this is objectively incorrect,
so fix this to get the correct starting generation based on the root
we're dropping to keep me from thinking there's a problem here.
CC: stable@vger.kernel.org Reviewed-by: Filipe Manana <fdmanana@suse.com> Signed-off-by: Josef Bacik <josef@toxicpanda.com> Signed-off-by: David Sterba <dsterba@suse.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Btrfs like other file systems can't really deal with I/O not aligned to
it's internal block size (which strangely is called sector size in
btrfs, for historical reasons), but the block layer split helper doesn't
even know about that.
Round down the split boundary so that all I/Os are aligned.
Fixes: d5e4377d5051 ("btrfs: split zone append bios in btrfs_submit_bio") CC: stable@vger.kernel.org # 6.12 Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Signed-off-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Damien Le Moal <dlemoal@kernel.org> Reviewed-by: David Sterba <dsterba@suse.com> Signed-off-by: David Sterba <dsterba@suse.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
It was recently noticed that set_codetag_empty() might be used not only to
mark NULL alloctag references as empty to avoid warnings but also to reset
valid tags (in clear_page_tag_ref()). Since set_codetag_empty() is
defined as NOOP for CONFIG_MEM_ALLOC_PROFILING_DEBUG=n, such use of
set_codetag_empty() leads to subtle bugs. Fix set_codetag_empty() for
CONFIG_MEM_ALLOC_PROFILING_DEBUG=n to reset the tag reference.
Link: https://lkml.kernel.org/r/20241130001423.1114965-2-surenb@google.com Fixes: a8fc28dad6d5 ("alloc_tag: introduce clear_page_tag_ref() helper function") Signed-off-by: Suren Baghdasaryan <surenb@google.com> Reported-by: David Wang <00107082@163.com> Closes: https://lore.kernel.org/lkml/20241124074318.399027-1-00107082@163.com/ Cc: David Wang <00107082@163.com> Cc: Kent Overstreet <kent.overstreet@linux.dev> Cc: Mike Rapoport (Microsoft) <rppt@kernel.org> Cc: Pasha Tatashin <pasha.tatashin@soleen.com> Cc: Sourav Panda <souravpanda@google.com> Cc: Yu Zhao <yuzhao@google.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Add a check before the calculation to avoid this problem.
syzbot reported this as a slab-out-of-bounds in __rb_map_vma:
BUG: KASAN: slab-out-of-bounds in __rb_map_vma+0x9ab/0xae0 kernel/trace/ring_buffer.c:7058
Read of size 8 at addr ffff8880767dd2b8 by task syz-executor187/5836
In split_large_buddy(), we might call pfn_to_page() on a PFN that might
not exist. In corner cases, such as when freeing the highest pageblock in
the last memory section, this could result with CONFIG_SPARSEMEM &&
!CONFIG_SPARSEMEM_EXTREME in __pfn_to_section() returning NULL and and
__section_mem_map_addr() dereferencing that NULL pointer.
Let's fix it, and avoid doing a pfn_to_page() call for the first
iteration, where we already have the page.
So far this was found by code inspection, but let's just CC stable as the
fix is easy.
Link: https://lkml.kernel.org/r/20241210093437.174413-1-david@redhat.com Fixes: fd919a85cd55 ("mm: page_isolation: prepare for hygienic freelists") Signed-off-by: David Hildenbrand <david@redhat.com> Reported-by: Vlastimil Babka <vbabka@suse.cz> Closes: https://lkml.kernel.org/r/e1a898ba-a717-4d20-9144-29df1a6c8813@suse.cz Reviewed-by: Vlastimil Babka <vbabka@suse.cz> Reviewed-by: Zi Yan <ziy@nvidia.com> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Cc: Yu Zhao <yuzhao@google.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If the caller of vmap() specifies VM_MAP_PUT_PAGES (currently only the
i915 driver), we will decrement nr_vmalloc_pages and MEMCG_VMALLOC in
vfree(). These counters are incremented by vmalloc() but not by vmap() so
this will cause an underflow. Check the VM_MAP_PUT_PAGES flag before
decrementing either counter.
Link: https://lkml.kernel.org/r/20241211202538.168311-1-willy@infradead.org Fixes: b944afc9d64d ("mm: add a VM_MAP_PUT_PAGES flag for vmap") Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Reviewed-by: Shakeel Butt <shakeel.butt@linux.dev> Reviewed-by: Balbir Singh <balbirs@nvidia.com> Acked-by: Michal Hocko <mhocko@suse.com> Cc: Christoph Hellwig <hch@lst.de> Cc: Muchun Song <muchun.song@linux.dev> Cc: Roman Gushchin <roman.gushchin@linux.dev> Cc: "Uladzislau Rezki (Sony)" <urezki@gmail.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Setting backing device is done before ZRAM initialization. If we set the
backing device, then remove the ZRAM module without initializing the
device, the backing device reference will be leaked and the device will be
hold forever.
Fix this by always reset the ZRAM fully on rmmod or reset store.
Patch series "zram: fix backing device setup issue", v2.
This series fixes two bugs of backing device setting:
- ZRAM should reject using a zero sized (or the uninitialized ZRAM
device itself) as the backing device.
- Fix backing device leaking when removing a uninitialized ZRAM
device.
This patch (of 2):
Setting a zero sized block device as backing device is pointless, and one
can easily create a recursive loop by setting the uninitialized ZRAM
device itself as its own backing device by (zram0 is uninitialized):
echo /dev/zram0 > /sys/block/zram0/backing_dev
It's definitely a wrong config, and the module will pin itself, kernel
should refuse doing so in the first place.
By refusing to use zero sized device we avoided misuse cases including
this one above.
Use the helper function rather than reading it directly.
Reviewed-by: Yang Wang <kevinyang.wang@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 8f2cd1067afe68372a1723e05e19b68ed187676a) Cc: stable@vger.kernel.org Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Use the helper function rather than reading it directly.
Reviewed-by: Yang Wang <kevinyang.wang@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 22b9555bc90df22b585bdd1f161b61584b13af51) Cc: stable@vger.kernel.org Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Use the helper function rather than reading it directly.
Reviewed-by: Yang Wang <kevinyang.wang@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 2c8eeaaa0fe5841ccf07a0eb51b1426f34ef39f7) Cc: stable@vger.kernel.org Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The TP_printk() of a TRACE_EVENT() is a generic printf format that any
developer can create for their event. It may include pointers to strings
and such. A boot mapped buffer may contain data from a previous kernel
where the strings addresses are different.
One solution is to copy the event content and update the pointers by the
recorded delta, but a simpler solution (for now) is to just use the
print_fields() function to print these events. The print_fields() function
just iterates the fields and prints them according to what type they are,
and ignores the TP_printk() format from the event itself.
To understand the difference, when printing via TP_printk() the output
looks like this:
The values returned by the driver after processing the contents of the
Temperature Result and the Temperature Limit Registers do not correspond to
the TMP512/TMP513 specifications. A raw register value is converted to a
signed integer value by a sign extension in accordance with the algorithm
provided in the specification, but due to the off-by-one error in the sign
bit index, the result is incorrect.
According to the TMP512 and TMP513 datasheets, the Temperature Result (08h
to 0Bh) and Limit (11h to 14h) Registers are 13-bit two's complement
integer values, shifted left by 3 bits. The value is scaled by 0.0625
degrees Celsius per bit. E.g., if regval = 1 1110 0111 0000 000, the
output should be -25 degrees, but the driver will return +487 degrees.
Found by Linux Verification Center (linuxtesting.org) with SVACE.
Fixes: 59dfa75e5d82 ("hwmon: Add driver for Texas Instruments TMP512/513 sensor chips.") Signed-off-by: Murad Masimov <m.masimov@maxima.ru> Link: https://lore.kernel.org/r/20241216173648.526-4-m.masimov@maxima.ru
[groeck: fixed description line length] Signed-off-by: Guenter Roeck <linux@roeck-us.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
The value returned by the driver after processing the contents of the
Current Register does not correspond to the TMP512/TMP513 specifications.
A raw register value is converted to a signed integer value by a sign
extension in accordance with the algorithm provided in the specification,
but due to the off-by-one error in the sign bit index, the result is
incorrect. Moreover, negative values will be reported as large positive
due to missing sign extension from u32 to long.
According to the TMP512 and TMP513 datasheets, the Current Register (07h)
is a 16-bit two's complement integer value. E.g., if regval = 1000 0011
0000 0000, then the value must be (-32000 * lsb), but the driver will
return (33536 * lsb).
Fix off-by-one bug, and also cast data->curr_lsb_ua (which is of type u32)
to long to prevent incorrect cast for negative values.
Found by Linux Verification Center (linuxtesting.org) with SVACE.
Fixes: 59dfa75e5d82 ("hwmon: Add driver for Texas Instruments TMP512/513 sensor chips.") Signed-off-by: Murad Masimov <m.masimov@maxima.ru> Link: https://lore.kernel.org/r/20241216173648.526-3-m.masimov@maxima.ru
[groeck: Fixed description line length] Signed-off-by: Guenter Roeck <linux@roeck-us.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
The values returned by the driver after processing the contents of the
Shunt Voltage Register and the Shunt Limit Registers do not correspond to
the TMP512/TMP513 specifications. A raw register value is converted to a
signed integer value by a sign extension in accordance with the algorithm
provided in the specification, but due to the off-by-one error in the sign
bit index, the result is incorrect. Moreover, the PGA shift calculated with
the tmp51x_get_pga_shift function is relevant only to the Shunt Voltage
Register, but is also applied to the Shunt Limit Registers.
According to the TMP512 and TMP513 datasheets, the Shunt Voltage Register
(04h) is 13 to 16 bit two's complement integer value, depending on the PGA
setting. The Shunt Positive (0Ch) and Negative (0Dh) Limit Registers are
16-bit two's complement integer values. Below are some examples:
* Shunt Voltage Register
If PGA = 8, and regval = 1000 0011 0000 0000, then the decimal value must
be -32000, but the value calculated by the driver will be 33536.
* Shunt Limit Register
If regval = 1000 0011 0000 0000, then the decimal value must be -32000, but
the value calculated by the driver will be 768, if PGA = 1.
Fix sign bit index, and also correct misleading comment describing the
tmp51x_get_pga_shift function.
Found by Linux Verification Center (linuxtesting.org) with SVACE.
Since 2320c9e6a768 ("drm/sched: memset() 'job' in drm_sched_job_init()")
accessing job->base.sched can produce unexpected results as the initialisation
of (*job)->base.sched done in amdgpu_job_alloc is overwritten by the
memset.
This commit fixes an issue when a CS would fail validation and would
be rejected after job->num_ibs is incremented. In this case,
amdgpu_ib_free(ring->adev, ...) will be called, which would crash the
machine because the ring value is bogus.
To fix this, pass a NULL pointer to amdgpu_ib_free(): we can do this
because the device is actually not used in this function.
The next commit will remove the ring argument completely.
Fixes: 2320c9e6a768 ("drm/sched: memset() 'job' in drm_sched_job_init()") Signed-off-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com> Reviewed-by: Alex Deucher <alexander.deucher@amd.com> Reviewed-by: Christian König <christian.koenig@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 2ae520cb12831d264ceb97c61f72c59d33c0dbd7) Signed-off-by: Sasha Levin <sashal@kernel.org>
On gt reset, if a context is running, then accumulate it's active time
into the busyness counter since there will be no chance for the context
to switch out and update it's run time.
Active busyness of an engine is calculated using gt timestamp and the
context switch in time. While capturing the gt timestamp, it's possible
that the context switches out. This race could result in an active
busyness value that is greater than the actual context runtime value by a
small amount. This leads to a negative delta and throws off busyness
calculations for the user.
If a subsequent count is smaller than the previous one, just return the
previous one, since we expect the busyness to catch up.
On GT reset, we store total busyness counts for all engines and
re-register the utilization buffer with GuC. At that time we should
reset the buffer, so that we don't get spurious busyness counts on
subsequent queries.
To repro this issue, run igt@perf_pmu@busy-hang followed by
igt@perf_pmu@most-busy-idle-check-all for a couple iterations.
The shutdown function can be called when the display is already
unprepared. For example during reboot this trigger a kernel
backlog. Calling the drm_panel_unprepare, allow us to avoid
to trigger the kernel warning.
The DSI host must be enabled for the panel to be initialized in
prepare(). Set the prepare_prev_first flag to guarantee this.
This fixes the panel operation on NXP i.MX8MP SoC / Samsung DSIM
DSI host.
The arguments for __dma_buf_debugfs_list_del do not match for both the
CONFIG_DEBUG_FS case and the !CONFIG_DEBUG_FS case. The !CONFIG_DEBUG_FS
case should take a struct dma_buf *, but it's currently struct file *.
This can lead to the build error:
error: passing argument 1 of ‘__dma_buf_debugfs_list_del’ from
incompatible pointer type [-Werror=incompatible-pointer-types]
dma-buf.c:63:53: note: expected ‘struct file *’ but argument is of
type ‘struct dma_buf *’
63 | static void __dma_buf_debugfs_list_del(struct file *file)
Fixes: bfc7bc539392 ("dma-buf: Do not build debugfs related code when !CONFIG_DEBUG_FS") Signed-off-by: T.J. Mercier <tjmercier@google.com> Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com> Signed-off-by: Sumit Semwal <sumit.semwal@linaro.org> Link: https://patchwork.freedesktop.org/patch/msgid/20241117170326.1971113-1-tjmercier@google.com Signed-off-by: Sasha Levin <sashal@kernel.org>
In export_udmabuf(), if dma_buf_fd() fails because the FD table is full, a
dma_buf owning the udmabuf has already been created; but the error handling
in udmabuf_create() will tear down the udmabuf without doing anything about
the containing dma_buf.
This leaves a dma_buf in memory that contains a dangling pointer; though
that doesn't seem to lead to anything bad except a memory leak.
Fix it by moving the dma_buf_fd() call out of export_udmabuf() so that we
can give it different error handling.
Note that the shape of this code changed a lot in commit 5e72b2b41a21
("udmabuf: convert udmabuf driver to use folios"); but the memory leak
seems to have existed since the introduction of udmabuf.
Fixes: d3116756a710 ("drm/ttm: rename bo->mem and make it a pointer") Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3837 Reviewed-by: Christian König <christian.koenig@amd.com> Signed-off-by: Michel Dänzer <mdaenzer@redhat.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 695c2c745e5dff201b75da8a1d237ce403600d04) Cc: stable@vger.kernel.org Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The VM pointer might already be outdated when that function is called.
Use the PASID instead to gather the information instead.
Signed-off-by: Christian König <christian.koenig@amd.com> Reviewed-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 57f812d171af4ba233d3ed7c94dfa5b8e92dcc04) Cc: stable@vger.kernel.org Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drm_mode_vrefresh() is trying to avoid divide by zero
by checking whether htotal or vtotal are zero. But we may
still end up with a div-by-zero of vtotal*htotal*...
Cc: stable@vger.kernel.org Reported-by: syzbot+622bba18029bcde672e1@syzkaller.appspotmail.com Closes: https://syzkaller.appspot.com/bug?extid=622bba18029bcde672e1 Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20241129042629.18280-2-ville.syrjala@linux.intel.com Reviewed-by: Jani Nikula <jani.nikula@intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Instead of returning a generic NULL on error from
drm_dp_tunnel_mgr_create(), use error pointers with informative codes
to align the function with stub that is executed when
CONFIG_DRM_DISPLAY_DP_TUNNEL is unset. This will also trigger IS_ERR()
in current caller (intel_dp_tunnerl_mgr_init()) instead of bypassing it
via NULL pointer.
v2: use error codes inside drm_dp_tunnel_mgr_create() instead of handling
on caller's side (Michal, Imre)
v3: fixup commit message and add "CC"/"Fixes" lines (Andi),
mention aligning function code with stub
Fixes: 91888b5b1ad2 ("drm/i915/dp: Add support for DP tunnel BW allocation") Cc: Imre Deak <imre.deak@intel.com> Cc: <stable@vger.kernel.org> # v6.9+ Signed-off-by: Krzysztof Karas <krzysztof.karas@intel.com> Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com> Signed-off-by: Imre Deak <imre.deak@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/7q4fpnmmztmchczjewgm6igy55qt6jsm7tfd4fl4ucfq6yg2oy@q4lxtsu6445c Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When USB-C monitor is connected directly to Intel Barlow Ridge host, it
goes into "redrive" mode that basically routes the DisplayPort signals
directly from the GPU to the USB-C monitor without any tunneling needed.
However, the host router must be powered on for this to work. Aaron
reported that there are a couple of cases where this will not work with
the current code:
- Booting with USB-C monitor plugged in.
- Plugging in USB-C monitor when the host router is in sleep state
(runtime suspended).
- Plugging in USB-C device while the system is in system sleep state.
In all these cases once the host router is runtime suspended the picture
on the connected USB-C display disappears too. This is certainly not
what the user expected.
For this reason improve the redrive mode handling to keep the host
router from runtime suspending when detect that any of the above cases
is happening.
Fixes: a75e0684efe5 ("thunderbolt: Keep the domain powered when USB4 port is in redrive mode") Reported-by: Aaron Rainbolt <arainbolt@kfocus.org> Closes: https://lore.kernel.org/linux-usb/20241009220118.70bfedd0@kf-ir16/ Cc: stable@vger.kernel.org Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
xHC hosts from several vendors have the same issue where endpoints start
so slowly that a later queued 'Stop Endpoint' command may complete before
endpoint is up and running.
The 'Stop Endpoint' command fails with context state error as the endpoint
still appears as stopped.
See commit 42b758137601 ("usb: xhci: Limit Stop Endpoint retries") for
details
-------------------------------
| If Number | Function |
-------------------------------
| 2 | USB AP Log Port |
-------------------------------
| 3 | USB AP GNSS Port|
-------------------------------
| 4 | USB AP META Port|
-------------------------------
| 5 | ADB port |
-------------------------------
| 6 | USB MD AT Port |
------------------------------
| 7 | USB MD META Port|
-------------------------------
| 8 | USB NTZ Port |
-------------------------------
| 9 | USB Debug port |
-------------------------------
Signed-off-by: Jack Wu <wojackbb@gmail.com> Cc: stable@vger.kernel.org Signed-off-by: Johan Hovold <johan@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Tested successfully connecting to the Internet via rndis interface after
dialing via AT commands on If#=3 or If#=4.
Not sure of the purpose of the other serial interfaces.
Signed-off-by: Michal Hrusecky <michal.hrusecky@turris.com> Cc: stable@vger.kernel.org Signed-off-by: Johan Hovold <johan@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The Hexagon-specific constant extender optimization in LLVM may crash on
Linux kernel code [1], such as fs/bcache/btree_io.c after
commit 32ed4a620c54 ("bcachefs: Btree path tracepoints") in 6.12:
clang: llvm/lib/Target/Hexagon/HexagonConstExtenders.cpp:745: bool (anonymous namespace)::HexagonConstExtenders::ExtRoot::operator<(const HCE::ExtRoot &) const: Assertion `ThisB->getParent() == OtherB->getParent()' failed.
Stack dump:
0. Program arguments: clang --target=hexagon-linux-musl ... fs/bcachefs/btree_io.c
1. <eof> parser at end of file
2. Code generation
3. Running pass 'Function Pass Manager' on module 'fs/bcachefs/btree_io.c'.
4. Running pass 'Hexagon constant-extender optimization' on function '@__btree_node_lock_nopath'
Without assertions enabled, there is just a hang during compilation.
This has been resolved in LLVM main (20.0.0) [2] and backported to LLVM
19.1.0 but the kernel supports LLVM 13.0.1 and newer, so disable the
constant expander optimization using the '-mllvm' option when using a
toolchain that is not fixed.
When looking up a non-existent file, efivarfs returns -EINVAL if the
file does not conform to the NAME-GUID format and -ENOENT if it does.
This is caused by efivars_d_hash() returning -EINVAL if the name is not
formatted correctly. This error is returned before simple_lookup()
returns a negative dentry, and is the error value that the user sees.
Fix by removing this check. If the file does not exist, simple_lookup()
will return a negative dentry leading to -ENOENT and efivarfs_create()
already has a validity check before it creates an entry (and will
correctly return -EINVAL)
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com> Cc: <stable@vger.kernel.org>
[ardb: make efivarfs_valid_name() static] Signed-off-by: Ard Biesheuvel <ardb@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Currently, the RIIC driver may run the I2C bus faster than requested,
which may cause subtle failures. E.g. Biju reported a measured bus
speed of 450 kHz instead of the expected maximum of 400 kHz on RZ/G2L.
The initial calculation of the bus period uses DIV_ROUND_UP(), to make
sure the actual bus speed never becomes faster than the requested bus
speed. However, the subsequent division-by-two steps do not use
round-up, which may lead to a too-small period, hence a too-fast and
possible out-of-spec bus speed. E.g. on RZ/Five, requesting a bus speed
of 100 resp. 400 kHz will yield too-fast target bus speeds of 100806
resp. 403226 Hz instead of 97656 resp. 390625 Hz.
Fix this by using DIV_ROUND_UP() in the subsequent divisions, too.
Commit be26ba96421a ("block: Fix potential deadlock while freezing queue and
acquiring sysfs_loc") actually reverts commit 22465bbac53c ("blk-mq: move cpuhp
callback registering out of q->sysfs_lock"), and causes the original resctrl
lockdep warning.
So revert it and we need to fix the issue in another way.
Cc: Nilay Shroff <nilay@linux.ibm.com> Fixes: be26ba96421a ("block: Fix potential deadlock while freezing queue and acquiring sysfs_loc") Signed-off-by: Ming Lei <ming.lei@redhat.com> Link: https://lore.kernel.org/r/20241218101617.3275704-2-ming.lei@redhat.com Signed-off-by: Jens Axboe <axboe@kernel.dk> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Currently, we don't use the return value from sock_queue_rcv_skb, which
means we may leak skbs if a message is not successfully queued to a
socket.
Instead, ensure that we're freeing the skb where the sock hasn't
otherwise taken ownership of the skb by adding checks on the
sock_queue_rcv_skb() to invoke a kfree on failure.
In doing so, rather than using the 'rc' value to trigger the
kfree_skb(), use the skb pointer itself, which is more explicit.
Also, add a kunit test for the sock delivery failure cases.
The "gl->tot_len" variable is controlled by the user. It comes from
process_responses(). On 32bit systems, the "gl->tot_len +
sizeof(struct cpl_pass_accept_req) + sizeof(struct rss_header)" addition
could have an integer wrapping bug. Use size_add() to prevent this.
Fixes: a08943947873 ("crypto: chtls - Register chtls with net tls") Cc: stable@vger.kernel.org Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org> Reviewed-by: Simon Horman <horms@kernel.org> Link: https://patch.msgid.link/c6bfb23c-2db2-4e1b-b8ab-ba3925c82ef5@stanley.mountain Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Snapshot the output of CPUID.0xD.[1..n] during kvm.ko initiliaization to
avoid the overead of CPUID during runtime. The offset, size, and metadata
for CPUID.0xD.[1..n] sub-leaves does not depend on XCR0 or XSS values, i.e.
is constant for a given CPU, and thus can be cached during module load.
On Intel's Emerald Rapids, CPUID is *wildly* expensive, to the point where
recomputing XSAVE offsets and sizes results in a 4x increase in latency of
nested VM-Enter and VM-Exit (nested transitions can trigger
xstate_required_size() multiple times per transition), relative to using
cached values. The issue is easily visible by running `perf top` while
triggering nested transitions: kvm_update_cpuid_runtime() shows up at a
whopping 50%.
As measured via RDTSC from L2 (using KVM-Unit-Test's CPUID VM-Exit test
and a slightly modified L1 KVM to handle CPUID in the fastpath), a nested
roundtrip to emulate CPUID on Skylake (SKX), Icelake (ICX), and Emerald
Rapids (EMR) takes:
SKX 11650
ICX 22350
EMR 28850
Using cached values, the latency drops to:
SKX 6850
ICX 9000
EMR 7900
The underlying issue is that CPUID itself is slow on ICX, and comically
slow on EMR. The problem is exacerbated on CPUs which support XSAVES
and/or XSAVEC, as KVM invokes xstate_required_size() twice on each
runtime CPUID update, and because there are more supported XSAVE features
(CPUID for supported XSAVE feature sub-leafs is significantly slower).
Note, updating runtime CPUID information multiple times per nested
transition is itself a flaw, especially since CPUID is a mandotory
intercept on both Intel and AMD. E.g. KVM doesn't need to ensure emulated
CPUID state is up-to-date while running L2. That flaw will be fixed in a
future patch, as deferring runtime CPUID updates is more subtle than it
appears at first glance, the benefits aren't super critical to have once
the XSAVE issue is resolved, and caching CPUID output is desirable even if
KVM's updates are deferred.
Cc: Jim Mattson <jmattson@google.com> Cc: stable@vger.kernel.org Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-ID: <20241211013302.1347853-2-seanjc@google.com> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Catalin reports that a hypervisor lying to a guest about the size
of the ASID field may result in unexpected issues:
- if the underlying HW does only supports 8 bit ASIDs, the ASID
field in a TLBI VAE1* operation is only 8 bits, and the HW will
ignore the other 8 bits
- if on the contrary the HW is 16 bit capable, the ASID field
in the same TLBI operation is always 16 bits, irrespective of
the value of TCR_ELx.AS.
This could lead to missed invalidations if the guest was lead to
assume that the HW had 8 bit ASIDs while they really are 16 bit wide.
In order to avoid any potential disaster that would be hard to debug,
prenent the migration between a host with 8 bit ASIDs to one with
wider ASIDs (the converse was obviously always forbidden). This is
also consistent with what we already do for VMIDs.
If it becomes absolutely mandatory to support such a migration path
in the future, we will have to trap and emulate all TLBIs, something
that nobody should look forward to.
Fixes: d5a32b60dc18 ("KVM: arm64: Allow userspace to change ID_AA64MMFR{0-2}_EL1") Reported-by: Catalin Marinas <catalin.marinas@arm.com> Signed-off-by: Marc Zyngier <maz@kernel.org> Cc: stable@vger.kernel.org Cc: Will Deacon <will@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Marc Zyngier <maz@kernel.org> Cc: James Morse <james.morse@arm.com> Cc: Oliver Upton <oliver.upton@linux.dev> Acked-by: Catalin Marinas <catalin.marinas@arm.com> Link: https://lore.kernel.org/r/20241203190236.505759-1-maz@kernel.org Signed-off-by: Oliver Upton <oliver.upton@linux.dev> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The intent of the check is to see whether at least one UMC has ECC
enabled. So do that instead of tracking which ones are enabled in masks
which are too small in size anyway and lead to not loading the driver on
Zen4 machines with UMCs enabled over UMC8.
Fixes: e2be5955a886 ("EDAC/amd64: Add support for AMD Family 19h Models 10h-1Fh and A0h-AFh") Reported-by: Avadhut Naik <avadhut.naik@amd.com> Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de> Tested-by: Avadhut Naik <avadhut.naik@amd.com> Reviewed-by: Avadhut Naik <avadhut.naik@amd.com> Cc: <stable@kernel.org> Link: https://lore.kernel.org/r/20241210212054.3895697-1-avadhut.naik@amd.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
It appears that the relatively popular RK3399 SoC has been put together
using a large amount of illicit substances, as experiments reveal that its
integration of GIC500 exposes the *secure* programming interface to
non-secure.
This has some pretty bad effects on the way priorities are handled, and
results in a dead machine if booting with pseudo-NMI enabled
(irqchip.gicv3_pseudo_nmi=1) if the kernel contains 18fdb6348c480 ("arm64:
irqchip/gic-v3: Select priorities at boot time"), which relies on the
priorities being programmed using the NS view.
Let's restore some sanity by going one step further and disable security
altogether in this case. This is not any worse, and puts us in a mode where
priorities actually make some sense.
Huge thanks to Mark Kettenis who initially identified this issue on
OpenBSD, and to Chen-Yu Tsai who reported the problem in Linux.
Fixes: 18fdb6348c480 ("arm64: irqchip/gic-v3: Select priorities at boot time") Reported-by: Mark Kettenis <mark.kettenis@xs4all.nl> Reported-by: Chen-Yu Tsai <wens@csie.org> Signed-off-by: Marc Zyngier <maz@kernel.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Chen-Yu Tsai <wens@csie.org> Cc: stable@vger.kernel.org Link: https://lore.kernel.org/all/20241213141037.3995049-1-maz@kernel.org Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Current implementation leaves pdev->dev as a wakeup source. Add a
device_init_wakeup(&pdev->dev, false) call in the .remove() function and
in the error path of the .probe() function.
Signed-off-by: Joe Hattori <joe@pf.is.s.u-tokyo.ac.jp> Fixes: 527f36f5efa4 ("mmc: mediatek: add support for SDIO eint wakup IRQ") Cc: stable@vger.kernel.org
Message-ID: <20241203023442.2434018-1-joe@pf.is.s.u-tokyo.ac.jp> Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Value 0 in ADMA length descriptor is interpreted as 65536 on new Tegra
chips, remove SDHCI_QUIRK_BROKEN_ADMA_ZEROLEN_DESC quirk to make sure max
ADMA2 length is 65536.
fwnode_find_mii_timestamper() calls of_parse_phandle_with_fixed_args()
but does not decrement the refcount of the obtained OF node. Add an
of_node_put() call before returning from the function.
This bug was detected by an experimental static analysis tool that I am
developing.
Fixes: bc1bee3b87ee ("net: mdiobus: Introduce fwnode_mdiobus_register_phy()") Signed-off-by: Joe Hattori <joe@pf.is.s.u-tokyo.ac.jp> Reviewed-by: Andrew Lunn <andrew@lunn.ch> Link: https://patch.msgid.link/20241218035106.1436405-1-joe@pf.is.s.u-tokyo.ac.jp Signed-off-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Empty netlink responses from do() are not correct (as opposed to
dump() where not dumping anything is perfectly fine).
We should return an error if the target object does not exist,
in this case if the netdev is down it has no queues.
Fixes: 6b6171db7fc8 ("netdev-genl: Add netlink framework functions for queue") Reported-by: syzbot+0a884bc2d304ce4af70f@syzkaller.appspotmail.com Reviewed-by: Eric Dumazet <edumazet@google.com> Reviewed-by: Joe Damato <jdamato@fastly.com> Link: https://patch.msgid.link/20241218022508.815344-1-kuba@kernel.org Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Robert Hodaszi reports that locally terminated traffic towards
VLAN-unaware bridge ports is broken with ocelot-8021q. He is describing
the same symptoms as for commit 1f9fc48fd302 ("net: dsa: sja1105: fix
reception from VLAN-unaware bridges").
For context, the set merged as "VLAN fixes for Ocelot driver":
https://lore.kernel.org/netdev/20240815000707.2006121-1-vladimir.oltean@nxp.com/
was developed in a slightly different form earlier this year, in January.
Initially, the switch was unconditionally configured to set OCELOT_ES0_TAG
when using ocelot-8021q, regardless of port operating mode.
This led to the situation where VLAN-unaware bridge ports would always
push their PVID - see ocelot_vlan_unaware_pvid() - a negligible value
anyway - into RX packets. To strip this in software, we would have needed
DSA to know what private VID the switch chose for VLAN-unaware bridge
ports, and pushed into the packets. This was implemented downstream, and
a remnant of it remains in the form of a comment mentioning
ds->ops->get_private_vid(), as something which would maybe need to be
considered in the future.
However, for upstream, it was deemed inappropriate, because it would
mean introducing yet another behavior for stripping VLAN tags from
VLAN-unaware bridge ports, when one already existed (ds->untag_bridge_pvid).
The latter has been marked as obsolete along with an explanation why it
is logically broken, but still, it would have been confusing.
So, for upstream, felix_update_tag_8021q_rx_rule() was developed, which
essentially changed the state of affairs from "Felix with ocelot-8021q
delivers all packets as VLAN-tagged towards the CPU" into "Felix with
ocelot-8021q delivers all packets from VLAN-aware bridge ports towards
the CPU". This was done on the premise that in VLAN-unaware mode,
there's nothing useful in the VLAN tags, and we can avoid introducing
ds->ops->get_private_vid() in the DSA receive path if we configure the
switch to not push those VLAN tags into packets in the first place.
Unfortunately, and this is when the trainwreck started, the selftests
developed initially and posted with the series were not re-ran.
dsa_software_vlan_untag() was initially written given the assumption
that users of this feature would send _all_ traffic as VLAN-tagged.
It was only partially adapted to the new scheme, by removing
ds->ops->get_private_vid(), which also used to be necessary in
standalone ports mode.
Where the trainwreck became even worse is that I had a second opportunity
to think about this, when the dsa_software_vlan_untag() logic change
initially broke sja1105, in commit 1f9fc48fd302 ("net: dsa: sja1105: fix
reception from VLAN-unaware bridges"). I did not connect the dots that
it also breaks ocelot-8021q, for pretty much the same reason that not
all received packets will be VLAN-tagged.
To be compatible with the optimized Felix control path which runs
felix_update_tag_8021q_rx_rule() to only push VLAN tags when useful (in
VLAN-aware mode), we need to restore the old dsa_software_vlan_untag()
logic. The blamed commit introduced the assumption that
dsa_software_vlan_untag() will see only VLAN-tagged packets, assumption
which is false. What corrupts RX traffic is the fact that we call
skb_vlan_untag() on packets which are not VLAN-tagged in the first
place.
Fixes: 93e4649efa96 ("net: dsa: provide a software untagging function on RX for VLAN-aware bridges") Reported-by: Robert Hodaszi <robert.hodaszi@digi.com> Closes: https://lore.kernel.org/netdev/20241215163334.615427-1-robert.hodaszi@digi.com/ Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com> Link: https://patch.msgid.link/20241216135059.1258266-1-vladimir.oltean@nxp.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Fix the way tcpdump is executed by:
- Using the right variable for the namespace. Currently the use of the
empty "ns" makes the command fail.
- Waiting until it starts to capture to ensure the interesting traffic
is caught on slow systems.
- Using line-buffered output to ensure logs are available when the test
is paused with "-p". Otherwise the last chunk of data might only be
written when tcpdump is killed.
With CONFIG_PROVE_LOCKING, when creating a set of type bitmap:ip, adding
it to a set of type list:set and populating it from iptables SET target
triggers a kernel warning:
| WARNING: possible recursive locking detected
| 6.12.0-rc7-01692-g5e9a28f41134-dirty #594 Not tainted
| --------------------------------------------
| ping/4018 is trying to acquire lock:
| ffff8881094a6848 (&set->lock){+.-.}-{2:2}, at: ip_set_add+0x28c/0x360 [ip_set]
|
| but task is already holding lock:
| ffff88811034c048 (&set->lock){+.-.}-{2:2}, at: ip_set_add+0x28c/0x360 [ip_set]
This is a false alarm: ipset does not allow nested list:set type, so the
loop in list_set_kadd() can never encounter the outer set itself. No
other set type supports embedded sets, so this is the only case to
consider.
To avoid the false report, create a distinct lock class for list:set
type ipset locks.
Fixes: f830837f0eed ("netfilter: ipset: list:set set type support") Signed-off-by: Phil Sutter <phil@nwl.cc> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
The 'max_avail' value is calculated from the system memory
size using order_base_2().
order_base_2(x) is defined as '(x) ? fn(x) : 0'.
The compiler generates two copies of the code that follows
and then expands clamp(max, min, PAGE_SHIFT - 12) (11 on 32bit).
This triggers a compile-time assert since min is 5.
In reality a system would have to have less than 512MB memory
for the bounds passed to clamp to be reversed.
Swap the order of the arguments to clamp() to avoid the warning.
Replace the clamp_val() on the line below with clamp().
clamp_val() is just 'an accident waiting to happen' and not needed here.
Detected by compile time checks added to clamp(), specifically:
minmax.h: use BUILD_BUG_ON_MSG() for the lo < hi test in clamp()
The interrupt line of PCI devices is interpreted as edge-triggered,
however the interrupt signal of the m_can controller integrated in Intel
Elkhart Lake CPUs appears to be generated level-triggered.
Consider the following sequence of events:
- IR register is read, interrupt X is set
- A new interrupt Y is triggered in the m_can controller
- IR register is written to acknowledge interrupt X. Y remains set in IR
As at no point in this sequence no interrupt flag is set in IR, the
m_can interrupt line will never become deasserted, and no edge will ever
be observed to trigger another run of the ISR. This was observed to
result in the TX queue of the EHL m_can to get stuck under high load,
because frames were queued to the hardware in m_can_start_xmit(), but
m_can_finish_tx() was never run to account for their successful
transmission.
On an Elkhart Lake based board with the two CAN interfaces connected to
each other, the following script can reproduce the issue:
ip link set can0 up type can bitrate 1000000
ip link set can1 up type can bitrate 1000000
To fix the issue, repeatedly read and acknowledge interrupts at the
start of the ISR until no interrupt flags are set, so the next incoming
interrupt will also result in an edge on the interrupt line.
While we have received a report that even with this patch, the TX queue
can become stuck under certain (currently unknown) circumstances on the
Elkhart Lake, this patch completely fixes the issue with the above
reproducer, and it is unclear whether the remaining issue has a similar
cause at all.
While an m_can controller usually already has the init flag from a
hardware reset, no such reset happens on the integrated m_can_pci of the
Intel Elkhart Lake. If the CAN controller is found in an active state,
m_can_dev_setup() would fail because m_can_niso_supported() calls
m_can_cccr_update_bits(), which refuses to modify any other configuration
bits when CCCR_INIT is not set.
To avoid this issue, set CCCR_INIT before attempting to modify any other
configuration flags.
There are two skb pointers to manage tx skb's enqueued from n/w stack.
waiting_tx_skb pointer points to the tx skb which needs to be processed
and ongoing_tx_skb pointer points to the tx skb which is being processed.
SPI thread prepares the tx data chunks from the tx skb pointed by the
ongoing_tx_skb pointer. When the tx skb pointed by the ongoing_tx_skb is
processed, the tx skb pointed by the waiting_tx_skb is assigned to
ongoing_tx_skb and the waiting_tx_skb pointer is assigned with NULL.
Whenever there is a new tx skb from n/w stack, it will be assigned to
waiting_tx_skb pointer if it is NULL. Enqueuing and processing of a tx skb
handled in two different threads.
Consider a scenario where the SPI thread processed an ongoing_tx_skb and
it moves next tx skb from waiting_tx_skb pointer to ongoing_tx_skb pointer
without doing any NULL check. At this time, if the waiting_tx_skb pointer
is NULL then ongoing_tx_skb pointer is also assigned with NULL. After
that, if a new tx skb is assigned to waiting_tx_skb pointer by the n/w
stack and there is a chance to overwrite the tx skb pointer with NULL in
the SPI thread. Finally one of the tx skb will be left as unhandled,
resulting packet missing and memory leak.
- Consider the below scenario where the TXC reported from the previous
transfer is 10 and ongoing_tx_skb holds an tx ethernet frame which can be
transported in 20 TXCs and waiting_tx_skb is still NULL.
tx_credits = 10; /* 21 are filled in the previous transfer */
ongoing_tx_skb = 20;
waiting_tx_skb = NULL; /* Still NULL */
- So, (tc6->ongoing_tx_skb || tc6->waiting_tx_skb) becomes true.
- After oa_tc6_prepare_spi_tx_buf_for_tx_skbs()
ongoing_tx_skb = 10;
waiting_tx_skb = NULL; /* Still NULL */
- Perform SPI transfer.
- Process SPI rx buffer to get the TXC from footers.
- Now let's assume previously filled 21 TXCs are freed so we are good to
transport the next remaining 10 tx chunks from ongoing_tx_skb.
tx_credits = 21;
ongoing_tx_skb = 10;
waiting_tx_skb = NULL;
- So, (tc6->ongoing_tx_skb || tc6->waiting_tx_skb) becomes true again.
- In the oa_tc6_prepare_spi_tx_buf_for_tx_skbs()
ongoing_tx_skb = NULL;
waiting_tx_skb = NULL;
- Now the below bad case might happen,
Thread1 (oa_tc6_start_xmit) Thread2 (oa_tc6_spi_thread_handler)
--------------------------- -----------------------------------
- if waiting_tx_skb is NULL
- if ongoing_tx_skb is NULL
- ongoing_tx_skb = waiting_tx_skb
- waiting_tx_skb = skb
- waiting_tx_skb = NULL
...
- ongoing_tx_skb = NULL
- if waiting_tx_skb is NULL
- waiting_tx_skb = skb
To overcome the above issue, protect the moving of tx skb reference from
waiting_tx_skb pointer to ongoing_tx_skb pointer and assigning new tx skb
to waiting_tx_skb pointer, so that the other thread can't access the
waiting_tx_skb pointer until the current thread completes moving the tx
skb reference safely.
Fixes: 53fbde8ab21e ("net: ethernet: oa_tc6: implement transmit path to transfer tx ethernet frames") Signed-off-by: Parthiban Veerasooran <parthiban.veerasooran@microchip.com> Reviewed-by: Larysa Zaremba <larysa.zaremba@intel.com> Signed-off-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
SPI thread wakes up to perform SPI transfer whenever there is an TX skb
from n/w stack or interrupt from MAC-PHY. Ethernet frame from TX skb is
transferred based on the availability tx credits in the MAC-PHY which is
reported from the previous SPI transfer. Sometimes there is a possibility
that TX skb is available to transmit but there is no tx credits from
MAC-PHY. In this case, there will not be any SPI transfer but the thread
will be running in an endless loop until tx credits available again.
So checking the availability of tx credits along with TX skb will prevent
the above infinite loop. When the tx credits available again that will be
notified through interrupt which will trigger the SPI transfer to get the
available tx credits.
Fixes: 53fbde8ab21e ("net: ethernet: oa_tc6: implement transmit path to transfer tx ethernet frames") Reviewed-by: Jacob Keller <jacob.e.keller@intel.com> Signed-off-by: Parthiban Veerasooran <parthiban.veerasooran@microchip.com> Signed-off-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
There is a check for NULL at the start of create_txqs() and
create_rxqs() which tess if "nic_dev->txqs" is non-NULL. The
intention is that if the device is already open and the queues
are already created then we don't create them a second time.
However, the bug is that if we have an error in the create_txqs()
then the pointer doesn't get set back to NULL. The NULL check
at the start of the function will say that it's already open when
it's not and the device can't be used.
Set ->txqs back to NULL on cleanup on error.
Fixes: c3e79baf1b03 ("net-next/hinic: Add logical Txq and Rxq") Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org> Reviewed-by: Simon Horman <horms@kernel.org> Link: https://patch.msgid.link/0cc98faf-a0ed-4565-a55b-0fa2734bc205@stanley.mountain Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Small follow-up to align this to an equivalent behavior as the bond driver.
The change in 3625920b62c3 ("teaming: fix vlan_features computing") removed
the netdevice vlan_features when there is no team port attached, yet it
leaves the full set of enc_features intact.
Instead, leave the default features as pre 3625920b62c3, and recompute once
we do have ports attached. Also, similarly as in bonding case, call the
netdev_base_features() helper on the enc_features.
Fixes: 3625920b62c3 ("teaming: fix vlan_features computing") Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org> Link: https://patch.msgid.link/20241213123657.401868-1-daniel@iogearbox.net Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
The context is supposed to record the next queue to dump,
not last dumped. If the dump doesn't fit we will restart
from the already-dumped queue, duplicating the message.
Before this fix and with the selftest improvements later
in this series we see:
# ./run_kselftest.sh -t drivers/net:stats.py
timeout set to 45
selftests: drivers/net: stats.py
KTAP version 1
1..5
ok 1 stats.check_pause
ok 2 stats.check_fec
ok 3 stats.pkt_byte_sum
# Check| At /root/ksft-net-drv/drivers/net/./stats.py, line 125, in qstat_by_ifindex:
# Check| ksft_eq(len(queues[qtype]), len(set(queues[qtype])),
# Check failed 45 != 44 repeated queue keys
# Check| At /root/ksft-net-drv/drivers/net/./stats.py, line 127, in qstat_by_ifindex:
# Check| ksft_eq(len(queues[qtype]), max(queues[qtype]) + 1,
# Check failed 45 != 44 missing queue keys
# Check| At /root/ksft-net-drv/drivers/net/./stats.py, line 125, in qstat_by_ifindex:
# Check| ksft_eq(len(queues[qtype]), len(set(queues[qtype])),
# Check failed 45 != 44 repeated queue keys
# Check| At /root/ksft-net-drv/drivers/net/./stats.py, line 127, in qstat_by_ifindex:
# Check| ksft_eq(len(queues[qtype]), max(queues[qtype]) + 1,
# Check failed 45 != 44 missing queue keys
# Check| At /root/ksft-net-drv/drivers/net/./stats.py, line 125, in qstat_by_ifindex:
# Check| ksft_eq(len(queues[qtype]), len(set(queues[qtype])),
# Check failed 103 != 100 repeated queue keys
# Check| At /root/ksft-net-drv/drivers/net/./stats.py, line 127, in qstat_by_ifindex:
# Check| ksft_eq(len(queues[qtype]), max(queues[qtype]) + 1,
# Check failed 103 != 100 missing queue keys
# Check| At /root/ksft-net-drv/drivers/net/./stats.py, line 125, in qstat_by_ifindex:
# Check| ksft_eq(len(queues[qtype]), len(set(queues[qtype])),
# Check failed 102 != 100 repeated queue keys
# Check| At /root/ksft-net-drv/drivers/net/./stats.py, line 127, in qstat_by_ifindex:
# Check| ksft_eq(len(queues[qtype]), max(queues[qtype]) + 1,
# Check failed 102 != 100 missing queue keys
not ok 4 stats.qstat_by_ifindex
ok 5 stats.check_down
# Totals: pass:4 fail:1 xfail:0 xpass:0 skip:0 error:0
With the fix:
# ./ksft-net-drv/run_kselftest.sh -t drivers/net:stats.py
timeout set to 45
selftests: drivers/net: stats.py
KTAP version 1
1..5
ok 1 stats.check_pause
ok 2 stats.check_fec
ok 3 stats.pkt_byte_sum
ok 4 stats.qstat_by_ifindex
ok 5 stats.check_down
# Totals: pass:5 fail:0 xfail:0 xpass:0 skip:0 error:0
Fixes: ab63a2387cb9 ("netdev: add per-queue statistics") Reviewed-by: Joe Damato <jdamato@fastly.com> Link: https://patch.msgid.link/20241213152244.3080955-3-kuba@kernel.org Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
The context is supposed to record the next queue to dump,
not last dumped. If the dump doesn't fit we will restart
from the already-dumped queue, duplicating the message.
Before this fix and with the selftest improvements later
in this series we see:
# ./run_kselftest.sh -t drivers/net:queues.py
timeout set to 45
selftests: drivers/net: queues.py
KTAP version 1
1..2
# Check| At /root/ksft-net-drv/drivers/net/./queues.py, line 32, in get_queues:
# Check| ksft_eq(queues, expected)
# Check failed 102 != 100
# Check| At /root/ksft-net-drv/drivers/net/./queues.py, line 32, in get_queues:
# Check| ksft_eq(queues, expected)
# Check failed 101 != 100
not ok 1 queues.get_queues
ok 2 queues.addremove_queues
# Totals: pass:1 fail:1 xfail:0 xpass:0 skip:0 error:0
not ok 1 selftests: drivers/net: queues.py # exit=1
With the fix:
# ./ksft-net-drv/run_kselftest.sh -t drivers/net:queues.py
timeout set to 45
selftests: drivers/net: queues.py
KTAP version 1
1..2
ok 1 queues.get_queues
ok 2 queues.addremove_queues
# Totals: pass:2 fail:0 xfail:0 xpass:0 skip:0 error:0
Fixes: 6b6171db7fc8 ("netdev-genl: Add netlink framework functions for queue") Reviewed-by: Joe Damato <jdamato@fastly.com> Link: https://patch.msgid.link/20241213152244.3080955-2-kuba@kernel.org Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Since commit 0a77d947f599 ("ksmbd: check outstanding simultaneous SMB
operations"), ksmbd enforces a maximum number of simultaneous operations
for a connection. The problem is that reaching the limit causes ksmbd to
close the socket, and the client has no indication that it should have
slowed down.
This behaviour can be reproduced by setting "smb2 max credits = 128" (or
lower), and transferring a large file (25GB).
Windows clients fail with 0x8007003b (with smaller files even).
Fix this by delaying reading from the socket until there's room to
allocate a request. This effectively applies backpressure on the client,
so the transfer completes, albeit at a slower rate.
This changes the semantics of req_running to count all in-flight
requests on a given connection, rather than the number of elements
in the conn->request list. The latter is used only in smb2_cancel,
and the counter is not used
Signed-off-by: Marios Makassikis <mmakassikis@freebox.fr> Acked-by: Namjae Jeon <linkinjeon@kernel.org> Signed-off-by: Steve French <stfrench@microsoft.com>
Stable-dep-of: 43fb7bce8866 ("ksmbd: fix broken transfers when exceeding max simultaneous operations") Signed-off-by: Sasha Levin <sashal@kernel.org>
The existing linked list based implementation of how ts tags are
assigned and managed is unsafe against concurrency and corner cases:
- element addition in tx processing can race against element removal
in ts queue completion,
- element removal in ts queue completion can race against element
removal in device close,
- if a large number of frames gets added to tx queue without ts queue
completions in between, elements with duplicate tag values can get
added.
Use a different implementation, based on per-port used tags bitmaps and
saved skb arrays.
Safety for addition in tx processing vs removal in ts completion is
provided by:
tag = find_first_zero_bit(...);
smp_mb();
<write rdev->ts_skb[tag]>
set_bit(...);
Some calls into ionic_get_module_eeprom() don't use a single
full buffer size, but instead multiple calls with an offset.
Teach our driver to use the offset correctly so we can
respond appropriately to the caller.
Fixes: 4d03e00a2140 ("ionic: Add initial ethtool support") Signed-off-by: Shannon Nelson <shannon.nelson@amd.com> Reviewed-by: Jacob Keller <jacob.e.keller@intel.com> Link: https://patch.msgid.link/20241212213157.12212-4-shannon.nelson@amd.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
There are some FW error handling paths that can cause us to
try to destroy the workqueue more than once, so let's be sure
we're checking for that.
The case where this popped up was in an AER event where the
handlers got called in such a way that ionic_reset_prepare()
and thus ionic_dev_teardown() got called twice in a row.
The second time through the workqueue was already destroyed,
and destroy_workqueue() choked on the bad wq pointer.
We didn't hit this in AER handler testing before because at
that time we weren't using a private workqueue. Later we
replaced the use of the system workqueue with our own private
workqueue but hadn't rerun the AER handler testing since then.
Fixes: 9e25450da700 ("ionic: add private workqueue per-device") Signed-off-by: Shannon Nelson <shannon.nelson@amd.com> Reviewed-by: Jacob Keller <jacob.e.keller@intel.com> Link: https://patch.msgid.link/20241212213157.12212-3-shannon.nelson@amd.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
If register_netdev() fails, then the driver leaks the netdev notifier.
Fix this by calling ionic_lif_unregister() on register_netdev()
failure. This will also call ionic_lif_unregister_phc() if it has
already been registered.
Fixes: 30b87ab4c0b3 ("ionic: remove lif list concept") Signed-off-by: Brett Creeley <brett.creeley@amd.com> Signed-off-by: Shannon Nelson <shannon.nelson@amd.com> Reviewed-by: Jacob Keller <jacob.e.keller@intel.com> Link: https://patch.msgid.link/20241212213157.12212-2-shannon.nelson@amd.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Use the correct attribute space for sub-message key lookup in nested
attributes when adding attributes. This fixes rt_link where the "kind"
key and "data" sub-message are nested attributes in "linkinfo".
Packets injected by the CPU should have a SRC_PORT field equal to the
CPU port module index in the Analyzer block (ocelot->num_phys_ports).
The blamed commit copied the ocelot_ifh_set_basic() call incorrectly
from ocelot_xmit_common() in net/dsa/tag_ocelot.c. Instead of calling
with "x", it calls with BIT_ULL(x), but the field is not a port mask,
but rather a single port index.
[ side note: this is the technical debt of code duplication :( ]
The error used to be silent and doesn't appear to have other
user-visible manifestations, but with new changes in the packing
library, it now fails loudly as follows:
------------[ cut here ]------------
Cannot store 0x40 inside bits 46-43 - will truncate
sja1105 spi2.0: xmit timed out
WARNING: CPU: 1 PID: 102 at lib/packing.c:98 __pack+0x90/0x198
sja1105 spi2.0: timed out polling for tstamp
CPU: 1 UID: 0 PID: 102 Comm: felix_xmit
Tainted: G W N 6.13.0-rc1-00372-gf706b85d972d-dirty #2605
Call trace:
__pack+0x90/0x198 (P)
__pack+0x90/0x198 (L)
packing+0x78/0x98
ocelot_ifh_set_basic+0x260/0x368
ocelot_port_inject_frame+0xa8/0x250
felix_port_deferred_xmit+0x14c/0x258
kthread_worker_fn+0x134/0x350
kthread+0x114/0x138
The code path pertains to the ocelot switchdev driver and to the felix
secondary DSA tag protocol, ocelot-8021q. Here seen with ocelot-8021q.
The messenger (packing) is not really to blame, so fix the original
commit instead.
Fixes: e1b9e80236c5 ("net: mscc: ocelot: fix QoS class for injected packets with "ocelot-8021q"") Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com> Reviewed-by: Simon Horman <horms@kernel.org> Link: https://patch.msgid.link/20241212165546.879567-1-vladimir.oltean@nxp.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
projects / thirdparty / kernel / stable.git / log
