Now, the epoll only use wake_up() interface to wake up task.
However, sometimes, there are epoll users which want to use
the synchronous wakeup flag to hint the scheduler, such as
Android binder driver.
So add a wake_up_sync() define, and use the wake_up_sync()
when the sync is true in ep_poll_callback().
Co-developed-by: Jing Xia <jing.xia@unisoc.com> Signed-off-by: Jing Xia <jing.xia@unisoc.com> Signed-off-by: Xuewen Yan <xuewen.yan@unisoc.com> Link: https://lore.kernel.org/r/20240426080548.8203-1-xuewen.yan@unisoc.com Tested-by: Brian Geffon <bgeffon@google.com> Reviewed-by: Brian Geffon <bgeffon@google.com> Reported-by: Benoit Lize <lizeb@google.com> Signed-off-by: Christian Brauner <brauner@kernel.org> Cc: Brian Geffon <bgeffon@google.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When directory's last extent has more that one block and its length is
not multiple of a block side, the code wrongly decided to move to the
next extent instead of processing the last partial block. This led to
directory corruption. Fix the rounding issue.
Signed-off-by: Jan Kara <jack@suse.cz> Cc: Shreenidhi Shedi <yesshedi@gmail.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
It becomes a path component, so it shouldn't exceed NAME_MAX
characters. This was hardened in commit c152737be22b ("ceph: Use
strscpy() instead of strcpy() in __get_snap_name()"), but no actual
check was put in place.
__of_get_dma_parent() returns OF device node @args.np, but the node's
refcount is increased twice, by both of_parse_phandle_with_args() and
of_node_get(), so causes refcount leakage for the node.
Fix by directly returning the node got by of_parse_phandle_with_args().
Fixes: f83a6e5dea6c ("of: address: Add support for the parent DMA bus") Cc: stable@vger.kernel.org Signed-off-by: Zijun Hu <quic_zijuhu@quicinc.com> Link: https://lore.kernel.org/r/20241206-of_core_fix-v1-4-dc28ed56bec3@quicinc.com Signed-off-by: Rob Herring (Arm) <robh@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The current code uses some 'goto put;' to cancel the parsing operation
and can lead to a return code value of 0 even on error cases.
Indeed, some goto calls are done from a loop without setting the ret
value explicitly before the goto call and so the ret value can be set to
0 due to operation done in previous loop iteration. For instance match
can be set to 0 in the previous loop iteration (leading to a new
iteration) but ret can also be set to 0 it the of_property_read_u32()
call succeed. In that case if no match are found or if an error is
detected the new iteration, the return value can be wrongly 0.
Avoid those cases setting the ret value explicitly before the goto
calls.
Fixes: bd6f2fd5a1d5 ("of: Support parsing phandle argument lists through a nexus node") Cc: stable@vger.kernel.org Signed-off-by: Herve Codina <herve.codina@bootlin.com> Link: https://lore.kernel.org/r/20241202165819.158681-1-herve.codina@bootlin.com Signed-off-by: Rob Herring (Arm) <robh@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When F_SEAL_FUTURE_WRITE was introduced, it was overlooked that udmabuf
must reject memfds with this flag, just like ones with F_SEAL_WRITE.
Fix it by adding F_SEAL_FUTURE_WRITE to SEALS_DENIED.
Fixes: ab3948f58ff8 ("mm/memfd: add an F_SEAL_FUTURE_WRITE seal to memfd") Cc: stable@vger.kernel.org Acked-by: Vivek Kasireddy <vivek.kasireddy@intel.com> Signed-off-by: Jann Horn <jannh@google.com> Reviewed-by: Joel Fernandes (Google) <joel@joelfernandes.org> Signed-off-by: Vivek Kasireddy <vivek.kasireddy@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20241204-udmabuf-fixes-v2-2-23887289de1c@google.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Because the inode bitmap is corrupted, an inode with an inode number that
should exist as a ".nilfs" file was reassigned by nilfs_mkdir for "file0",
causing an inode duplication during execution. And this causes an
underflow of i_nlink in rmdir operations.
The inode is used twice by the same task to unmount and remove directories
".nilfs" and "file0", it trigger warning in nilfs_rmdir.
Avoid to this issue, check i_nlink in nilfs_iget(), if it is 0, it means
that this inode has been deleted, and iput is executed to reclaim it.
When block_invalidatepage was converted to block_invalidate_folio, the
fallback to block_invalidatepage in folio_invalidate() if the
address_space_operations method invalidatepage (currently
invalidate_folio) was not set, was removed.
Unfortunately, some pseudo-inodes in nilfs2 use empty_aops set by
inode_init_always_gfp() as is, or explicitly set it to
address_space_operations. Therefore, with this change,
block_invalidatepage() is no longer called from folio_invalidate(), and as
a result, the buffer_head structures attached to these pages/folios are no
longer freed via try_to_free_buffers().
Thus, these buffer heads are now leaked by truncate_inode_pages(), which
cleans up the page cache from inode evict(), etc.
Three types of caches use empty_aops: gc inode caches and the DAT shadow
inode used by GC, and b-tree node caches. Of these, b-tree node caches
explicitly call invalidate_mapping_pages() during cleanup, which involves
calling try_to_free_buffers(), so the leak was not visible during normal
operation but worsened when GC was performed.
Fix this issue by using address_space_operations with invalidate_folio set
to block_invalidate_folio instead of empty_aops, which will ensure the
same behavior as before.
On a malformed interrupt-map property which is shorter than expected by
1 cell, we may read bogus data past the end of the property instead of
returning an error in of_irq_parse_imap_parent().
Decrement the remaining length when skipping over the interrupt parent
phandle cell.
Fixes: 935df1bd40d4 ("of/irq: Factor out parsing of interrupt-map parent phandle+args from of_irq_parse_raw()") Cc: stable@vger.kernel.org Signed-off-by: Zijun Hu <quic_zijuhu@quicinc.com> Link: https://lore.kernel.org/r/20241209-of_irq_fix-v1-1-782f1419c8a1@quicinc.com
[rh: reword commit msg] Signed-off-by: Rob Herring (Arm) <robh@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When the server is recalling a layout, we should ignore the count of
outstanding layoutget calls, since the server is expected to return
either NFS4ERR_RECALLCONFLICT or NFS4ERR_RETURNCONFLICT for as long as
the recall is outstanding.
Currently, we may end up livelocking, causing the layout to eventually
be forcibly revoked.
Fixes: bf0291dd2267 ("pNFS: Ensure LAYOUTGET and LAYOUTRETURN are properly serialised") Cc: stable@vger.kernel.org Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
task work can be executed after the task has gone through io_uring
termination, whether it's the final task_work run or the fallback path.
In this case, task work will find ->io_wq being already killed and
null'ed, which is a problem if it then tries to forward the request to
io_queue_iowq(). Make io_queue_iowq() fail requests in this case.
Note that it also checks PF_KTHREAD, because the user can first close
a DEFER_TASKRUN ring and shortly after kill the task, in which case
->iowq check would race.
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 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>
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:
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>
[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>
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.
On SH, devm_clk_get_optional_enabled() fails with -EINVAL if the clock
is not found. This happens because __devm_clk_get() assumes it can pass
a NULL clock pointer (as returned by clk_get_optional()) to the init()
function (clk_prepare_enable() in this case), while the SH
implementation of clk_enable() considers that an error.
Fix this by making the SH clk_enable() implementation return zero
instead, like the Common Clock Framework does.
Reported-by: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de> Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be> Tested-by: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de> Acked-by: Stephen Boyd <sboyd@kernel.org> Link: https://lore.kernel.org/r/b53e6b557b4240579933b3359dda335ff94ed5af.1675354849.git.geert+renesas@glider.be Signed-off-by: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
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.
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>
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>
-------------------------------
| 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.
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>
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>
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>
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>
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>
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>
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>
When receiving clc msg, the field length in smc_clc_msg_hdr indicates the
length of msg should be received from network and the value should not be
fully trusted as it is from the network. Once the value of length exceeds
the value of buflen in function smc_clc_wait_msg it may run into deadloop
when trying to drain the remaining data exceeding buflen.
This patch checks the return value of sock_recvmsg when draining data in
case of deadloop in draining.
Fixes: fb4f79264c0f ("net/smc: tolerate future SMCD versions") Signed-off-by: Guangguan Wang <guangguan.wang@linux.alibaba.com> Reviewed-by: Wen Gu <guwen@linux.alibaba.com> Reviewed-by: D. Wythe <alibuda@linux.alibaba.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
When receiving proposal msg in server, the field smcd_v2_ext_offset in
proposal msg is from the remote client and can not be fully trusted.
Once the value of smcd_v2_ext_offset exceed the max value, there has
the chance to access wrong address, and crash may happen.
This patch checks the value of smcd_v2_ext_offset before using it.
Fixes: 5c21c4ccafe8 ("net/smc: determine accepted ISM devices") Signed-off-by: Guangguan Wang <guangguan.wang@linux.alibaba.com> Reviewed-by: Wen Gu <guwen@linux.alibaba.com> Reviewed-by: D. Wythe <alibuda@linux.alibaba.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
When receiving proposal msg in server, the field iparea_offset
and the field ipv6_prefixes_cnt in proposal msg are from the
remote client and can not be fully trusted. Especially the
field iparea_offset, once exceed the max value, there has the
chance to access wrong address, and crash may happen.
This patch checks iparea_offset and ipv6_prefixes_cnt before using them.
Fixes: e7b7a64a8493 ("smc: support variable CLC proposal messages") Signed-off-by: Guangguan Wang <guangguan.wang@linux.alibaba.com> Reviewed-by: Wen Gu <guwen@linux.alibaba.com> Reviewed-by: D. Wythe <alibuda@linux.alibaba.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
When application sending data more than sndbuf_space, there have chances
application will sleep in epoll_wait, and will never be wakeup again. This
is caused by a race between smc_poll and smc_cdc_tx_handler.
application tasklet
smc_tx_sendmsg(len > sndbuf_space) |
epoll_wait for EPOLL_OUT,timeout=0 |
smc_poll |
if (!smc->conn.sndbuf_space) |
| smc_cdc_tx_handler
| atomic_add sndbuf_space
| smc_tx_sndbuf_nonfull
| if (!test_bit SOCK_NOSPACE)
| do not sk_write_space;
set_bit SOCK_NOSPACE; |
return mask=0; |
Application will sleep in epoll_wait as smc_poll returns 0. And
smc_cdc_tx_handler will not call sk_write_space because the SOCK_NOSPACE
has not be set. If there is no inflight cdc msg, sk_write_space will not be
called any more, and application will sleep in epoll_wait forever.
So check sndbuf_space again after NOSPACE flag is set to break the race.
Fixes: 8dce2786a290 ("net/smc: smc_poll improvements") Signed-off-by: Guangguan Wang <guangguan.wang@linux.alibaba.com> Suggested-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
link down work may be scheduled before lgr freed but execute
after lgr freed, which may result in crash. So it is need to
hold a reference before shedule link down work, and put the
reference after work executed or canceled.
Fixes: 541afa10c126 ("net/smc: add smcr_port_err() and smcr_link_down() processing") Signed-off-by: Guangguan Wang <guangguan.wang@linux.alibaba.com> Reviewed-by: Tony Lu <tonylu@linux.alibaba.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
The cxl_port_setup_targets() algorithm fails to identify valid target list
ordering in the presence of 4-way and above switches resulting in
'cxl create-region' failures of the form:
$ cxl create-region -d decoder0.0 -g 1024 -s 2G -t ram -w 8 -m mem4 mem1 mem6 mem3 mem2 mem5 mem7 mem0
cxl region: create_region: region0: failed to set target7 to mem0
cxl region: cmd_create_region: created 0 regions
Fix the method to calculate @distance by iterativeley multiplying the
number of targets per switch port. This also follows the algorithm
recommended here [1].
Since commit f63b94be6942 ("i2c: pnx: Fix potential deadlock warning
from del_timer_sync() call in isr") jiffies are stored in
i2c_pnx_algo_data.timeout, but wait_timeout and wait_reset are still
using it as milliseconds. Convert jiffies back to milliseconds to wait
for the expected amount of time.
Fixes: f63b94be6942 ("i2c: pnx: Fix potential deadlock warning from del_timer_sync() call in isr") Signed-off-by: Vladimir Riabchun <ferr.lambarginio@gmail.com> Signed-off-by: Andi Shyti <andi.shyti@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
When drivers access P2SB device resources, it calls p2sb_bar(). Before
the commit 5913320eb0b3 ("platform/x86: p2sb: Allow p2sb_bar() calls
during PCI device probe"), p2sb_bar() obtained the resources and then
called pci_stop_and_remove_bus_device() for clean up. Then the P2SB
device disappeared. The commit 5913320eb0b3 introduced the P2SB device
resource cache feature in the boot process. During the resource cache,
pci_stop_and_remove_bus_device() is called for the P2SB device, then the
P2SB device disappears regardless of whether p2sb_bar() is called or
not. Such P2SB device disappearance caused a confusion [1]. To avoid the
confusion, avoid the pci_stop_and_remove_bus_device() call when the BIOS
does not hide the P2SB device.
For that purpose, cache the P2SB device resources only if the BIOS hides
the P2SB device. Call p2sb_scan_and_cache() only if p2sb_hidden_by_bios
is true. This allows removing two branches from p2sb_scan_and_cache().
When p2sb_bar() is called, get the resources from the cache if the P2SB
device is hidden. Otherwise, read the resources from the unhidden P2SB
device.
Reported-by: Daniel Walker (danielwa) <danielwa@cisco.com> Closes: https://lore.kernel.org/lkml/ZzTI+biIUTvFT6NC@goliath/ [1] Fixes: 5913320eb0b3 ("platform/x86: p2sb: Allow p2sb_bar() calls during PCI device probe") Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com> Reviewed-by: Hans de Goede <hdegoede@redhat.com> Link: https://lore.kernel.org/r/20241128002836.373745-5-shinichiro.kawasaki@wdc.com Reviewed-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com> Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
To prepare for the following fix, move the code to hide and unhide the
P2SB device from p2sb_cache_resources() to p2sb_scan_and_cache().
Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com> Reviewed-by: Hans de Goede <hdegoede@redhat.com> Link: https://lore.kernel.org/r/20241128002836.373745-4-shinichiro.kawasaki@wdc.com Reviewed-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com> Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
Stable-dep-of: 360c400d0f56 ("p2sb: Do not scan and remove the P2SB device when it is unhidden") Signed-off-by: Sasha Levin <sashal@kernel.org>
To prepare for the following fix, introduce the global flag
p2sb_hidden_by_bios. Check if the BIOS hides the P2SB device and store
the result in the flag. This allows to refer to the check result across
functions.
Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com> Reviewed-by: Hans de Goede <hdegoede@redhat.com> Link: https://lore.kernel.org/r/20241128002836.373745-3-shinichiro.kawasaki@wdc.com Reviewed-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com> Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
Stable-dep-of: 360c400d0f56 ("p2sb: Do not scan and remove the P2SB device when it is unhidden") Signed-off-by: Sasha Levin <sashal@kernel.org>
To prepare for the following fix, factor out the code to read the P2SB
resource from the cache to the new function p2sb_read_from_cache().
Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com> Reviewed-by: Hans de Goede <hdegoede@redhat.com> Link: https://lore.kernel.org/r/20241128002836.373745-2-shinichiro.kawasaki@wdc.com Reviewed-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com> Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
Stable-dep-of: 360c400d0f56 ("p2sb: Do not scan and remove the P2SB device when it is unhidden") Signed-off-by: Sasha Levin <sashal@kernel.org>
p2sb_get_devfn() always succeeds, make it return void and
remove error checking from its callers.
Reviewed-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com> Signed-off-by: Hans de Goede <hdegoede@redhat.com> Link: https://lore.kernel.org/r/20240305094500.23778-1-hdegoede@redhat.com
Stable-dep-of: 360c400d0f56 ("p2sb: Do not scan and remove the P2SB device when it is unhidden") Signed-off-by: Sasha Levin <sashal@kernel.org>
Introduce pci_resource_n() and replace open-coded implementations of it
in pci.h.
Link: https://lore.kernel.org/r/20230330162434.35055-3-andriy.shevchenko@linux.intel.com Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com> Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Stable-dep-of: 360c400d0f56 ("p2sb: Do not scan and remove the P2SB device when it is unhidden") Signed-off-by: Sasha Levin <sashal@kernel.org>
When using dma_map_sg() to map the scatterlist with iommu enabled,
the entries in the scatterlist can be mergerd into less but longer
entries in the function __finalise_sg(). So that the number of
valid mapped entries is actually smaller than ureq->num_reqs,and
there are still some invalid entries in the scatterlist with
dma_addr=0xffffffff and len=0. Writing these invalid sg entries
into the dma_desc can cause a data transmission error.
The function dma_map_sg() returns the number of valid map entries
and the return value is assigned to usb_request::num_mapped_sgs in
function usb_gadget_map_request_by_dev(). So that just write valid
mapped entries into dma_desc according to the usb_request::num_mapped_sgs,
and set the IOC bit if it's the last valid mapped entry.
This patch poses no risk to no-iommu situation, cause
ureq->num_mapped_sgs equals ureq->num_sgs while using dma_direct_map_sg()
to map the scatterlist whith iommu disabled.
Add it to silent warning:
arch/mips/boot/dts/loongson/ls7a-pch.dtsi:68.16-416.5: Warning (interrupt_provider): /bus@10000000/pci@1a000000: '#interrupt-cells' found, but node is not an interrupt provider
arch/mips/boot/dts/loongson/loongson64g_4core_ls7a.dts:32.31-40.4: Warning (interrupt_provider): /bus@10000000/msi-controller@2ff00000: Missing '#interrupt-cells' in interrupt provider
arch/mips/boot/dts/loongson/loongson64g_4core_ls7a.dtb: Warning (interrupt_map): Failed prerequisite 'interrupt_provider'
Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com> Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de> Signed-off-by: Sasha Levin <sashal@kernel.org>
The Broadcom BCM5760X NIC may be a multi-function device.
While it does not advertise an ACS capability, peer-to-peer transactions
are not possible between the individual functions. So it is ok to treat
them as fully isolated.
Add an ACS quirk for this device so the functions can be in independent
IOMMU groups and attached individually to userspace applications using
VFIO.
Some platforms (e.g. ti,j721e-usb, ti,am64-usb) require
this bit to be set to workaround a lockup issue with PHY
short suspend intervals [1]. Add a platform quirk flag
to indicate if Suspend Residency should be enabled.
[1] - https://www.ti.com/lit/er/sprz457h/sprz457h.pdf
i2409 - USB: USB2 PHY locks up due to short suspend
Signed-off-by: Roger Quadros <rogerq@kernel.org> Signed-off-by: Ravi Gunasekaran <r-gunasekaran@ti.com> Acked-by: Peter Chen <peter.chen@kernel.org> Link: https://lore.kernel.org/r/20240516044537.16801-2-r-gunasekaran@ti.com Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
The vmd driver creates a "domain" symlink in sysfs for each VMD bridge.
Previously this symlink was created after pci_bus_add_devices() added
devices below the VMD bridge and emitted udev events to announce them to
userspace.
This led to a race between userspace consumers of the udev events and the
kernel creation of the symlink. One such consumer is mdadm, which
assembles block devices into a RAID array, and for devices below a VMD
bridge, mdadm depends on the "domain" symlink.
If mdadm loses the race, it may be unable to assemble a RAID array, which
may cause a boot failure or other issues, with complaints like this:
(udev-worker)[2149]: nvme1n1: '/sbin/mdadm -I /dev/nvme1n1'(err) 'mdadm: Unable to get real path for '/sys/bus/pci/drivers/vmd/0000:c7:00.5/domain/device''
(udev-worker)[2149]: nvme1n1: '/sbin/mdadm -I /dev/nvme1n1'(err) 'mdadm: /dev/nvme1n1 is not attached to Intel(R) RAID controller.'
(udev-worker)[2149]: nvme1n1: '/sbin/mdadm -I /dev/nvme1n1'(err) 'mdadm: No OROM/EFI properties for /dev/nvme1n1'
(udev-worker)[2149]: nvme1n1: '/sbin/mdadm -I /dev/nvme1n1'(err) 'mdadm: no RAID superblock on /dev/nvme1n1.'
(udev-worker)[2149]: nvme1n1: Process '/sbin/mdadm -I /dev/nvme1n1' failed with exit code 1.
This symptom prevents the OS from booting successfully.
After a NVMe disk is probed/added by the nvme driver, udevd invokes mdadm
to detect if there is a mdraid associated with this NVMe disk, and mdadm
determines if a NVMe device is connected to a particular VMD domain by
checking the "domain" symlink. For example:
Thread A Thread B Thread mdadm
vmd_enable_domain
pci_bus_add_devices
__driver_probe_device
...
work_on_cpu
schedule_work_on
: wakeup Thread B
nvme_probe
: wakeup scan_work
to scan nvme disk
and add nvme disk
then wakeup udevd
: udevd executes
mdadm command
flush_work main
: wait for nvme_probe done ...
__driver_probe_device find_driver_devices
: probe next nvme device : 1) Detect domain symlink
... 2) Find domain symlink
... from vmd sysfs
... 3) Domain symlink not
... created yet; failed
sysfs_create_link
: create domain symlink
Create the VMD "domain" symlink before invoking pci_bus_add_devices() to
avoid this race.
Suggested-by: Adrian Huang <ahuang12@lenovo.com> Link: https://lore.kernel.org/linux-pci/20240605124844.24293-1-sjiwei@163.com Signed-off-by: Jiwei Sun <sunjw10@lenovo.com> Signed-off-by: Krzysztof Wilczyński <kwilczynski@kernel.org>
[bhelgaas: commit log] Signed-off-by: Bjorn Helgaas <bhelgaas@google.com> Reviewed-by: Nirmal Patel <nirmal.patel@linux.intel.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Use preserve_config in place of checking for PCI_PROBE_ONLY flag to enable
support for "linux,pci-probe-only" on a per host bridge basis.
This also obviates the use of adding PCI_REASSIGN_ALL_BUS flag if
!PCI_PROBE_ONLY, as pci_assign_unassigned_root_bus_resources() takes care
of reassigning the resources that are not already claimed.
If the link is powered off during suspend, electrical noise may cause
errors that are logged via AER. If the AER interrupt is enabled and shares
an IRQ with PME, that causes a spurious wakeup during suspend.
Disable the AER interrupt during suspend to prevent this. Clear error
status before re-enabling IRQ interrupts during resume so we don't get an
interrupt for errors that occurred during the suspend/resume process.
Changes to sch->q.qlen around qdisc_tree_reduce_backlog() need to happen
_before_ a call to said function because otherwise it may fail to notify
parent qdiscs when the child is about to become empty.
Signed-off-by: Lion Ackermann <nnamrec@gmail.com> Acked-by: Toke Høiland-Jørgensen <toke@toke.dk> Signed-off-by: David S. Miller <davem@davemloft.net> Cc: Artem Metla <ametla@google.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In 32-bit x86 builds CONFIG_STATIC_CALL_INLINE isn't set, leading to
static_call_initialized not being available.
Define it as "0" in that case.
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au> Fixes: 0ef8047b737d ("x86/static-call: provide a way to do very early static-call updates") Signed-off-by: Juergen Gross <jgross@suse.com> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The usb_get_descriptor() function does DMA so we're not allowed
to use a stack buffer for that. Doing DMA to the stack is not portable
all architectures. Move the "new_device_descriptor" from being stored
on the stack and allocate it with kmalloc() instead.
The hypercall page is no longer needed. It can be removed, as from the
Xen perspective it is optional.
But, from Linux's perspective, it removes naked RET instructions that
escape the speculative protections that Call Depth Tracking and/or
Untrain Ret are trying to achieve.
This is part of XSA-466 / CVE-2024-53241.
Reported-by: Andrew Cooper <andrew.cooper3@citrix.com> Signed-off-by: Juergen Gross <jgross@suse.com> Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com> Reviewed-by: Jan Beulich <jbeulich@suse.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Add generic hypercall functions usable for all normal (i.e. not iret)
hypercalls. Depending on the guest type and the processor vendor
different functions need to be used due to the to be used instruction
for entering the hypervisor:
- PV guests need to use syscall
- HVM/PVH guests on Intel need to use vmcall
- HVM/PVH guests on AMD and Hygon need to use vmmcall
As PVH guests need to issue hypercalls very early during boot, there
is a 4th hypercall function needed for HVM/PVH which can be used on
Intel and AMD processors. It will check the vendor type and then set
the Intel or AMD specific function to use via static_call().
This is part of XSA-466 / CVE-2024-53241.
Reported-by: Andrew Cooper <andrew.cooper3@citrix.com> Signed-off-by: Juergen Gross <jgross@suse.com> Co-developed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The syscall instruction is used in Xen PV mode for doing hypercalls.
Allow syscall to be used in the kernel in case it is tagged with an
unwind hint for objtool.
This is part of XSA-466 / CVE-2024-53241.
Reported-by: Andrew Cooper <andrew.cooper3@citrix.com> Signed-off-by: Juergen Gross <jgross@suse.com> Co-developed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
In order to be able to differentiate between AMD and Intel based
systems for very early hypercalls without having to rely on the Xen
hypercall page, make get_cpu_vendor() non-static.
Refactor early_cpu_init() for the same reason by splitting out the
loop initializing cpu_devs() into an externally callable function.
This is part of XSA-466 / CVE-2024-53241.
Reported-by: Andrew Cooper <andrew.cooper3@citrix.com> Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When removing a netfront device directly after a suspend/resume cycle
it might happen that the queues have not been setup again, causing a
crash during the attempt to stop the queues another time.
Fix that by checking the queues are existing before trying to stop
them.
This is XSA-465 / CVE-2024-53240.
Reported-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com> Fixes: d50b7914fae0 ("xen-netfront: Fix NULL sring after live migration") Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Nikolay Kuratov [Mon, 16 Dec 2024 11:19:23 +0000 (14:19 +0300)]
tracing/kprobes: Skip symbol counting logic for module symbols in create_local_trace_kprobe()
commit b022f0c7e404 ("tracing/kprobes: Return EADDRNOTAVAIL when func matches several symbols")
avoids checking number_of_same_symbols() for module symbol in
__trace_kprobe_create(), but create_local_trace_kprobe() should avoid this
check too. Doing this check leads to ENOENT for module_name:symbol_name
constructions passed over perf_event_open.
No bug in newer kernels as it was fixed more generally by
commit 9d8616034f16 ("tracing/kprobes: Add symbol counting check when module loads")
Range propagation must not affect subreg_def marks, otherwise the
following example is rewritten by verifier incorrectly when
BPF_F_TEST_RND_HI32 flag is set:
(or zero extension of w1 at (2) is missing for architectures that
require zero extension for upper register half).
The following happens w/o this patch:
- r0 is marked as not a subreg at (0);
- w1 is marked as subreg at (2);
- w1 subreg_def is overridden at (3) by copy_register_state();
- w1 is read at (5) but mark_insn_zext() does not mark (2)
for zero extension, because w1 subreg_def is not set;
- because of BPF_F_TEST_RND_HI32 flag verifier inserts random
value for hi32 bits of (2) (marked (r));
- this random value is read at (5).
Fixes: 75748837b7e5 ("bpf: Propagate scalar ranges through register assignments.") Reported-by: Lonial Con <kongln9170@gmail.com> Signed-off-by: Lonial Con <kongln9170@gmail.com> Signed-off-by: Eduard Zingerman <eddyz87@gmail.com> Signed-off-by: Andrii Nakryiko <andrii@kernel.org> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Acked-by: Daniel Borkmann <daniel@iogearbox.net> Closes: https://lore.kernel.org/bpf/7e2aa30a62d740db182c170fdd8f81c596df280d.camel@gmail.com Link: https://lore.kernel.org/bpf/20240924210844.1758441-1-eddyz87@gmail.com
[ shung-hsi.yu: sync_linked_regs() was called find_equal_scalars() before commit 4bf79f9be434 ("bpf: Track equal scalars history on per-instruction level"), and
modification is done because there is only a single call to
copy_register_state() before commit 98d7ca374ba4 ("bpf: Track delta between
"linked" registers."). ] Signed-off-by: Shung-Hsi Yu <shung-hsi.yu@suse.com> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
After a recent change to clamp() and its variants [1] that increases the
coverage of the check that high is greater than low because it can be
done through inlining, certain build configurations (such as s390
defconfig) fail to build with clang with:
block/blk-iocost.c:1101:11: error: call to '__compiletime_assert_557' declared with 'error' attribute: clamp() low limit 1 greater than high limit active
1101 | inuse = clamp_t(u32, inuse, 1, active);
| ^
include/linux/minmax.h:218:36: note: expanded from macro 'clamp_t'
218 | #define clamp_t(type, val, lo, hi) __careful_clamp(type, val, lo, hi)
| ^
include/linux/minmax.h:195:2: note: expanded from macro '__careful_clamp'
195 | __clamp_once(type, val, lo, hi, __UNIQUE_ID(v_), __UNIQUE_ID(l_), __UNIQUE_ID(h_))
| ^
include/linux/minmax.h:188:2: note: expanded from macro '__clamp_once'
188 | BUILD_BUG_ON_MSG(statically_true(ulo > uhi), \
| ^
__propagate_weights() is called with an active value of zero in
ioc_check_iocgs(), which results in the high value being less than the
low value, which is undefined because the value returned depends on the
order of the comparisons.
The purpose of this expression is to ensure inuse is not more than
active and at least 1. This could be written more simply with a ternary
expression that uses min(inuse, active) as the condition so that the
value of that condition can be used if it is not zero and one if it is.
Do this conversion to resolve the error and add a comment to deter
people from turning this back into clamp().
The voice setting is used by sco_connect() or sco_conn_defer_accept()
after being set by sco_sock_setsockopt().
The PCM part of the voice setting is used for offload mode through PCM
chipset port.
This commits add support for mSBC 16 bits offloading, i.e. audio data
not transported over HCI.
The BCM4349B1 supports 16 bits transparent data on its I2S port.
If BT_VOICE_TRANSPARENT is used when accepting a SCO connection, this
gives only garbage audio while using BT_VOICE_TRANSPARENT_16BIT gives
correct audio.
This has been tested with connection to iPhone 14 and Samsung S24.
Fixes: ad10b1a48754 ("Bluetooth: Add Bluetooth socket voice option") Signed-off-by: Frédéric Danis <frederic.danis@collabora.com> Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
This updates iso_sock_accept to use nested locking for the parent
socket, to avoid lockdep warnings caused because the parent and
child sockets are locked by the same thread:
[ 41.585683] ============================================
[ 41.585688] WARNING: possible recursive locking detected
[ 41.585694] 6.12.0-rc6+ #22 Not tainted
[ 41.585701] --------------------------------------------
[ 41.585705] iso-tester/3139 is trying to acquire lock:
[ 41.585711] ffff988b29530a58 (sk_lock-AF_BLUETOOTH)
at: bt_accept_dequeue+0xe3/0x280 [bluetooth]
[ 41.585905]
but task is already holding lock:
[ 41.585909] ffff988b29533a58 (sk_lock-AF_BLUETOOTH)
at: iso_sock_accept+0x61/0x2d0 [bluetooth]
[ 41.586064]
other info that might help us debug this:
[ 41.586069] Possible unsafe locking scenario:
This bug was first introduced in c27f3d011b08, where the author of the
patch probably meant to do DeleteMutex instead of ReleaseMutex. The
mutex leak was noticed later on and fixed in e4dfe108371, but the bogus
MutexRelease line was never removed, so do it now.
Similar to bonding driver, add NETIF_F_GSO_ENCAP_ALL to TEAM_VLAN_FEATURES
in order to support slave devices which propagate NETIF_F_GSO_UDP_TUNNEL &
NETIF_F_GSO_UDP_TUNNEL_CSUM as vlan_features.
Fixes: 3625920b62c3 ("teaming: fix vlan_features computing") Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Cc: Nikolay Aleksandrov <razor@blackwall.org> Cc: Ido Schimmel <idosch@idosch.org> Cc: Jiri Pirko <jiri@nvidia.com> Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org> Reviewed-by: Hangbin Liu <liuhangbin@gmail.com> Link: https://patch.msgid.link/20241210141245.327886-5-daniel@iogearbox.net Signed-off-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Drivers like mlx5 expose NIC's vlan_features such as
NETIF_F_GSO_UDP_TUNNEL & NETIF_F_GSO_UDP_TUNNEL_CSUM which are
later not propagated when the underlying devices are bonded and
a vlan device created on top of the bond.
Right now, the more cumbersome workaround for this is to create
the vlan on top of the mlx5 and then enslave the vlan devices
to a bond.
To fix this, add NETIF_F_GSO_ENCAP_ALL to BOND_VLAN_FEATURES
such that bond_compute_features() can probe and propagate the
vlan_features from the slave devices up to the vlan device.
# ethtool -k enp2s0f0np0 | grep udp
tx-udp_tnl-segmentation: on
tx-udp_tnl-csum-segmentation: on
tx-udp-segmentation: on
rx-udp_tunnel-port-offload: on
rx-udp-gro-forwarding: off
# ethtool -k enp2s0f1np1 | grep udp
tx-udp_tnl-segmentation: on
tx-udp_tnl-csum-segmentation: on
tx-udp-segmentation: on
rx-udp_tunnel-port-offload: on
rx-udp-gro-forwarding: off
# ethtool -k bond0 | grep udp
tx-udp_tnl-segmentation: on
tx-udp_tnl-csum-segmentation: on
tx-udp-segmentation: on
rx-udp_tunnel-port-offload: off [fixed]
rx-udp-gro-forwarding: off
Before:
# ethtool -k bond0.100 | grep udp
tx-udp_tnl-segmentation: off [requested on]
tx-udp_tnl-csum-segmentation: off [requested on]
tx-udp-segmentation: on
rx-udp_tunnel-port-offload: off [fixed]
rx-udp-gro-forwarding: off
After:
# ethtool -k bond0.100 | grep udp
tx-udp_tnl-segmentation: on
tx-udp_tnl-csum-segmentation: on
tx-udp-segmentation: on
rx-udp_tunnel-port-offload: off [fixed]
rx-udp-gro-forwarding: off
Various users have run into this reporting performance issues when
configuring Cilium in vxlan tunneling mode and having the combination
of bond & vlan for the core devices connecting the Kubernetes cluster
to the outside world.
Fixes: a9b3ace44c7d ("bonding: fix vlan_features computing") Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Cc: Nikolay Aleksandrov <razor@blackwall.org> Cc: Ido Schimmel <idosch@idosch.org> Cc: Jiri Pirko <jiri@nvidia.com> Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org> Reviewed-by: Hangbin Liu <liuhangbin@gmail.com> Link: https://patch.msgid.link/20241210141245.327886-3-daniel@iogearbox.net Signed-off-by: Paolo Abeni <pabeni@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
In general, 'qlen' of any classful qdisc should keep track of the
number of packets that the qdisc itself and all of its children holds.
In case of netem, 'qlen' only accounts for the packets in its internal
tfifo. When netem is used with a child qdisc, the child qdisc can use
'qdisc_tree_reduce_backlog' to inform its parent, netem, about created
or dropped SKBs. This function updates 'qlen' and the backlog statistics
of netem, but netem does not account for changes made by a child qdisc.
'qlen' then indicates the wrong number of packets in the tfifo.
If a child qdisc creates new SKBs during enqueue and informs its parent
about this, netem's 'qlen' value is increased. When netem dequeues the
newly created SKBs from the child, the 'qlen' in netem is not updated.
If 'qlen' reaches the configured sch->limit, the enqueue function stops
working, even though the tfifo is not full.
Reproduce the bug:
Ensure that the sender machine has GSO enabled. Configure netem as root
qdisc and tbf as its child on the outgoing interface of the machine
as follows:
$ tc qdisc add dev <oif> root handle 1: netem delay 100ms limit 100
$ tc qdisc add dev <oif> parent 1:0 tbf rate 50Mbit burst 1542 latency 50ms
Send bulk TCP traffic out via this interface, e.g., by running an iPerf3
client on the machine. Check the qdisc statistics:
$ tc -s qdisc show dev <oif>
Statistics after 10s of iPerf3 TCP test before the fix (note that
netem's backlog > limit, netem stopped accepting packets):
qdisc netem 1: root refcnt 2 limit 1000 delay 100ms
Sent 2767766 bytes 1848 pkt (dropped 652, overlimits 0 requeues 0)
backlog 4294528236b 1155p requeues 0
qdisc tbf 10: parent 1:1 rate 50Mbit burst 1537b lat 50ms
Sent 2767766 bytes 1848 pkt (dropped 327, overlimits 7601 requeues 0)
backlog 0b 0p requeues 0
tbf segments the GSO SKBs (tbf_segment) and updates the netem's 'qlen'.
The interface fully stops transferring packets and "locks". In this case,
the child qdisc and tfifo are empty, but 'qlen' indicates the tfifo is at
its limit and no more packets are accepted.
This patch adds a counter for the entries in the tfifo. Netem's 'qlen' is
only decreased when a packet is returned by its dequeue function, and not
during enqueuing into the child qdisc. External updates to 'qlen' are thus
accounted for and only the behavior of the backlog statistics changes. As
in other qdiscs, 'qlen' then keeps track of how many packets are held in
netem and all of its children. As before, sch->limit remains as the
maximum number of packets in the tfifo. The same applies to netem's
backlog statistics.
Fixes: 50612537e9ab ("netem: fix classful handling") Signed-off-by: Martin Ottens <martin.ottens@fau.de> Acked-by: Jamal Hadi Salim <jhs@mojatatu.com> Link: https://patch.msgid.link/20241210131412.1837202-1-martin.ottens@fau.de Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
tc qdisc replace dev $send_if parent root handle 100 taprio \
num_tc 8 queues 1@0 1@1 1@2 1@3 1@4 1@5 1@6 1@7 \
map 0 1 2 3 4 5 6 7 \
base-time 0 cycle-time 10000 \
sched-entry S 01 1250 \
sched-entry S 02 1250 \
sched-entry S 04 1250 \
sched-entry S 08 1250 \
sched-entry S 10 1250 \
sched-entry S 20 1250 \
sched-entry S 40 1250 \
sched-entry S 80 1250 \
flags 2
ptp4l would fail to take TX timestamps of Pdelay_Resp messages like this:
increasing tx_timestamp_timeout may correct this issue, but it is likely caused by a driver bug
ptp4l[4134.168]: port 2: send peer delay response failed
It turns out that the driver can't take their TX timestamps because it
can't transmit them in the first place. And there's nothing special
about the Pdelay_Resp packets - they're just regular 68 byte packets.
But with this taprio configuration, the switch would refuse to send even
the ETH_ZLEN minimum packet size.
This should have definitely not been the case. When applying the taprio
config, the driver prints:
mscc_felix 0000:00:00.5: port 0 tc 0 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 132 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 1 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 132 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 2 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 132 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 3 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 132 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 4 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 132 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 5 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 132 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 6 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 132 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 7 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 132 octets including FCS
and thus, everything under 132 bytes - ETH_FCS_LEN should have been sent
without problems. Yet it's not.
For the forwarding path, the configuration is fine, yet packets injected
from Linux get stuck with this schedule no matter what.
The first hint that the static guard bands are the cause of the problem
is that reverting Michael Walle's commit 297c4de6f780 ("net: dsa: felix:
re-enable TAS guard band mode") made things work. It must be that the
guard bands are calculated incorrectly.
I remembered that there is a magic constant in the driver, set to 33 ns
for no logical reason other than experimentation, which says "never let
the static guard bands get so large as to leave less than this amount of
remaining space in the time slot, because the queue system will refuse
to schedule packets otherwise, and they will get stuck". I had a hunch
that my previous experimentally-determined value was only good for
packets coming from the forwarding path, and that the CPU injection path
needed more.
I came to the new value of 35 ns through binary search, after seeing
that with 544 ns (the bit time required to send the Pdelay_Resp packet
at gigabit) it works. Again, this is purely experimental, there's no
logic and the manual doesn't say anything.
The new driver prints for this schedule look like this:
mscc_felix 0000:00:00.5: port 0 tc 0 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 131 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 1 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 131 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 2 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 131 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 3 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 131 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 4 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 131 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 5 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 131 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 6 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 131 octets including FCS
mscc_felix 0000:00:00.5: port 0 tc 7 min gate length 1250 ns not enough for max frame size 1526 at 1000 Mbps, dropping frames over 131 octets including FCS
So yes, the maximum MTU is now even smaller by 1 byte than before.
This is maybe counter-intuitive, but makes more sense with a diagram of
one time slot.
Before:
Gate open Gate close
| |
v 1250 ns total time slot duration v
<---------------------------------------------------->
<----><---------------------------------------------->
33 ns 1217 ns static guard band
useful
Gate open Gate close
| |
v 1250 ns total time slot duration v
<---------------------------------------------------->
<-----><--------------------------------------------->
35 ns 1215 ns static guard band
useful
The static guard band implemented by this switch hardware directly
determines the maximum allowable MTU for that traffic class. The larger
it is, the earlier the switch will stop scheduling frames for
transmission, because otherwise they might overrun the gate close time
(and avoiding that is the entire purpose of Michael's patch).
So, we now have guard bands smaller by 2 ns, thus, in this particular
case, we lose a byte of the maximum MTU.
Fixes: 11afdc6526de ("net: dsa: felix: tc-taprio intervals smaller than MTU should send at least one packet") Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com> Reviewed-by: Michael Walle <mwalle@kernel.org> Link: https://patch.msgid.link/20241210132640.3426788-1-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
