Upon experiencing a PCI error, fbnic reset the device to recover from
the failure. Reset the hardware stats as part of the device reset to
ensure accurate stats reporting.
Note that the reset is not really resetting the aggregate value to 0,
which may result in a spike for a system collecting deltas in stats.
Rather, the reset re-latches the current value as previous, in case HW
got reset.
Signed-off-by: Mohsin Bashir <mohsin.bashr@gmail.com> Reviewed-by: Jacob Keller <jacob.e.keller@intel.com> Link: https://patch.msgid.link/20250825200206.2357713-3-kuba@kernel.org Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
This patch fixes an issue where two different flows on the same RXq
produce the same hash resulting in continuous flow overwrites.
Flow #1: A packet for Flow #1 comes in, kernel calls the steering
function. The driver gives back a filter id. The kernel saves
this filter id in the selected slot. Later, the driver's
service task checks if any filters have expired and then
installs the rule for Flow #1.
Flow #2: A packet for Flow #2 comes in. It goes through the same steps.
But this time, the chosen slot is being used by Flow #1. The
driver gives a new filter id and the kernel saves it in the
same slot. When the driver's service task runs, it runs through
all the flows, checks if Flow #1 should be expired, the kernel
returns True as the slot has a different filter id, and then
the driver installs the rule for Flow #2.
Flow #1: Another packet for Flow #1 comes in. The same thing repeats.
The slot is overwritten with a new filter id for Flow #1.
This causes a repeated cycle of flow programming for missed packets,
wasting CPU cycles while not improving performance. This problem happens
at higher rates when the RPS table is small, but tests show it still
happens even with 12,000 connections and an RPS size of 16K per queue
(global table size = 144x16K = 64K).
This patch prevents overwriting an rps_dev_flow entry if it is active.
The intention is that it is better to do aRFS for the first flow instead
of hurting all flows on the same hash. Without this, two (or more) flows
on one RX queue with the same hash can keep overwriting each other. This
causes the driver to reprogram the flow repeatedly.
Changes:
1. Add a new 'hash' field to struct rps_dev_flow.
2. Add rps_flow_is_active(): a helper function to check if a flow is
active or not, extracted from rps_may_expire_flow(). It is further
simplified as per reviewer feedback.
3. In set_rps_cpu():
- Avoid overwriting by programming a new filter if:
- The slot is not in use, or
- The slot is in use but the flow is not active, or
- The slot has an active flow with the same hash, but target CPU
differs.
- Save the hash in the rps_dev_flow entry.
4. rps_may_expire_flow(): Use earlier extracted rps_flow_is_active().
Testing & results:
- Driver: ice (E810 NIC), Kernel: net-next
- #CPUs = #RXq = 144 (1:1)
- Number of flows: 12K
- Eight RPS settings from 256 to 32768. Though RPS=256 is not ideal,
it is still sufficient to cover 12K flows (256*144 rx-queues = 64K
global table slots)
- Global Table Size = 144 * RPS (effectively equal to 256 * RPS)
- Each RPS test duration = 8 mins (org code) + 8 mins (new code).
- Metrics captured on client
Legend for following tables:
Steer-C: #times ndo_rx_flow_steer() was Called by set_rps_cpu()
Steer-L: #times ice_arfs_flow_steer() Looped over aRFS entries
Add: #times driver actually programmed aRFS (ice_arfs_build_entry())
Del: #times driver deleted the flow (ice_arfs_del_flow_rules())
Units: K = 1,000 times, M = 1 million times
Some observations:
1. Overall Latency improved: (1790.19-1634.94)/1790.19*100 = 8.67%
2. Overall CPU increased: (777.32-751.49)/751.45*100 = 3.44%
3. Flow Management (add/delete) remained almost constant at ~11K
compared to values in millions.
Add a new _fini() routine on the GT TLB invalidation
side to handle this worker cleanup on driver teardown.
v2: Move the TLB teardown to the gt fini() routine called during
gt_init rather than in gt_alloc. This way the GT structure stays
alive for while we reset the TLB state.
[WHY&HOW]
The sink max slice width limitation should be considered for DSC, but
was removed in "refactor DSC cap calculations".
This patch adds it back and takes the valid minimum between the sink and
source.
Signed-off-by: Dillon Varone <Dillon.Varone@amd.com> Signed-off-by: Aurabindo Pillai <aurabindo.pillai@amd.com> Reviewed-by: Wenjing Liu <Wenjing.Liu@amd.com> Tested-by: Dan Wheeler <daniel.wheeler@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
When KFD asks CP to preempt queues, other than preempt CP queues, CP
also requests SDMA to preempt SDMA queues with UNMAP_LATENCY timeout.
Currently queue_preemption_timeout_ms is 9000 ms by default but can be
configured via module parameter. KFD_UNMAP_LATENCY_MS is hard coded as
4000 ms though. This patch ties KFD_UNMAP_LATENCY_MS to
queue_preemption_timeout_ms so in a slow system such as emulator, both
CP and SDMA slowness are taken into account.
Signed-off-by: Amber Lin <Amber.Lin@amd.com> Reviewed-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
[Why]
For the HW cursor, its current position in the pipe_ctx->stream struct is
not affected by the 180 rotation, i. e. the top left corner is still at
0,0. However, the DPP & HUBP set_cursor_position functions require rotated
position.
The current approach is hard-coded for ODM 2:1, thus it's failing for
ODM 4:1, resulting in a double cursor.
[How]
Instead of calculating the new cursor position relatively to the
viewports, we calculate it using a viewavable clip_rect of each plane.
The clip_rects are first offset and scaled to the same space as the
src_rect, i. e. Stream space -> Plane space.
In case of a pipe split, which divides the plane into 2 or more viewports,
the clip_rect is the union of all the viewports of the given plane.
With the assumption that the viewports in HUBP's set_cursor_position are
in the Plane space as well, it should produce a correct cursor position
for any number of pipe splits.
Reviewed-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com> Signed-off-by: Ivan Lipski <ivan.lipski@amd.com> Signed-off-by: Leo Li <sunpeng.li@amd.com> Signed-off-by: Aurabindo Pillai <aurabindo.pillai@amd.com> Tested-by: Dan Wheeler <daniel.wheeler@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
When it only allocates vram without va, which is 0, and a
SVM range allocated stays in this range, the vram allocation
returns failure. It should be skipped for this case from
SVM usage check.
Signed-off-by: Eric Huang <jinhuieric.huang@amd.com> Reviewed-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Signed-off-by: Ce Sun <cesun102@amd.com> Reviewed-by: Yang Wang <kevinyang.wang@amd.com> Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
By polling, poll ACA bank count to ensure that valid
ACA bank reg info can be obtained
v2: add corresponding delay before send msg to SMU to query mca bank info
(Stanley)
v3: the loop cannot exit. (Thomas)
v4: remove amdgpu_aca_clear_bank_count. (Kevin)
v5: continuously inject ce. If a creation interruption
occurs at this time, bank reg info will be lost. (Thomas)
v5: each cycle is delayed by 100ms. (Tao)
Signed-off-by: Ce Sun <cesun102@amd.com> Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
The pci_endpoint_test tests the 32-bit MSI range. However, the device might
not have all vectors configured. For example, if msi_interrupts is 8 in the
ep function space or if the MSI Multiple Message Capable value is
configured as 4 (maximum 16 vectors).
In this case, do not attempt to run the test to avoid timeouts and directly
return the error value.
This patch ensures the gt will be awake for the entire duration
of the resume sequences until GuCRC takes over and GT-C6 gets
re-enabled.
Before suspending GT-C6 is kept enabled, but upon resume, GuCRC
is not yet alive to properly control the exits and some cases of
instability and corruption related to GT-C6 can be observed.
The original commit be2ff42c5d6e ("fuse: Use hash table to link
processing request") converted fuse_pqueue->processing to a hash table,
but virtio_fs_enqueue_req() was not updated to use it correctly.
So use fuse_pqueue->processing as a hash table, this make the code
more coherent
Co-developed-by: Fushuai Wang <wangfushuai@baidu.com> Signed-off-by: Fushuai Wang <wangfushuai@baidu.com> Signed-off-by: Li RongQing <lirongqing@baidu.com> Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com> Signed-off-by: Miklos Szeredi <mszeredi@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
The test currently modifies the HDS settings and doesn't restore them.
This may cause subsequent tests to fail (or pass when they should not).
Add defer()ed reset handling.
fixed_phy_register() creates and registers the phy_device. To be
symmetric, we should not only unregister, but also free the phy_device
in fixed_phy_unregister(). This allows to simplify code in users.
Note wrt of_phy_deregister_fixed_link():
put_device(&phydev->mdio.dev) and phy_device_free(phydev) are identical.
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com> Reviewed-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk> Link: https://patch.msgid.link/ad8dda9a-10ed-4060-916b-3f13bdbb899d@gmail.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
The commit 9ab440a9d042 ("drm/xe/ptl: L3bank mask is not
available on the media GT") added a workaround to ignore
the fuse register that L3 bank availability as it did not
contain valid values. Same is true for WCL therefore extend
the workaround to cover it.
Device can be unbound, so driver must also release memory for the wakeup
source. Do not use devm interface, because it would change the order of
cleanup.
Eliminate the use of static variables within the log pull implementation
to resolve a race condition and prevent data gaps when pulling logs from
multiple controllers in parallel, ensuring each operation is properly
isolated.
Firmware can enter a transient fault while creating operational queues.
The driver fails the load immediately.
Add a retry loop that checks controller status and history bit after
queue creation. If either indicates a fault, retry init up to a set
limit before failing.
When Wi-Fi is scanning at 2.4GHz, PTA will abort almost all the BT request.
Once the Wi-Fi slot stay too long, BT audio device can not get enough data,
audio glitch will happened. This patch limit 2.4Ghz Wi-Fi slot to 80ms
while Wi-Fi is scanning to avoid audio glitch.
During enclosure reboot or expander reset, firmware may report a link
speed of 0 in "Device Add" events while the link is still coming up.
The driver drops such devices, leaving them missing even after the link
recovers.
Fix this by treating link speed 0 as 1.5 Gbps during device addition so
the device is exposed to the OS. The actual link speed will be updated
later when link-up events arrive.
Currently, ip_extract_route_hint uses RTN_BROADCAST to decide
whether to use the route dst hint mechanism.
This check is too strict, as it prevents directed broadcast
routes from using the hint, resulting in poor performance
during bursts of directed broadcast traffic.
Fix this in ip_extract_route_hint and modify ip_route_use_hint
to preserve the intended behaviour.
Module aliases are used by userspace to identify the correct module to
load for a detected hardware. The currently supported RPMSG device IDs for
this module include "rpmsg-raw", but the module alias is "rpmsg_chrdev".
Use the helper macro MODULE_DEVICE_TABLE(rpmsg) to export the correct
supported IDs. And while here, to keep backwards compatibility we also add
the other ID "rpmsg_chrdev" so that it is also still exported as an alias.
This has the side benefit of adding support for some legacy firmware
which still uses the original "rpmsg_chrdev" ID. This was the ID used for
this driver before it was upstreamed (as reflected by the module alias).
Signed-off-by: Andrew Davis <afd@ti.com> Acked-by: Hari Nagalla <hnagalla@ti.com> Tested-by: Hari Nagalla <hnagalla@ti.com> Link: https://lore.kernel.org/r/20250619205722.133827-1-afd@ti.com Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
GPD devices originally used BMI160 sensors with the "BMI0160" PNP ID.
When they switched to BMI260 sensors in newer hardware, they reused
the existing Windows driver which accepts both "BMI0160" and "BMI0260"
IDs. Consequently, they kept "BMI0160" in DSDT tables for new BMI260
devices, causing driver mismatches in Linux.
1. GPD updated BIOS v0.40+[1] for newer devices to report "BMI0260" for
BMI260 sensors to avoid loading the bmi160 driver on Linux. While this
isn't Bosch's VID;
2. Bosch's official Windows driver uses "BMI0260" as a compatible ID
3. We're seeing real devices shipping with "BMI0260" in DSDT
The DSDT excerpt of GPD G1619-04 with BIOS v0.40:
Scope (_SB.I2CC)
{
Device (BMA2)
{
Name (_ADR, Zero) // _ADR: Address
Name (_HID, "BMI0260") // _HID: Hardware ID
Name (_CID, "BMI0260") // _CID: Compatible ID
Name (_DDN, "Accelerometer") // _DDN: DOS Device Name
Name (_UID, One) // _UID: Unique ID
Method (_CRS, 0, NotSerialized) // _CRS: Current Resource Settings
{
Name (RBUF, ResourceTemplate ()
{
I2cSerialBusV2 (0x0069, ControllerInitiated, 0x00061A80,
AddressingMode7Bit, "\\_SB.I2CC",
0x00, ResourceConsumer, , Exclusive,
)
})
Return (RBUF) /* \_SB_.I2CC.BMA2._CRS.RBUF */
}
# omit some noise
}
}
At least for panel-bridges, the atomic_enable call is defined as being
called right after the preceding element in the display pipe is enabled.
It is also stated that "The bridge can assume that the display pipe (i.e.
clocks and timing signals) feeding it is running when this callback is
called"
This means the DSI controller driving this display would have already
switched over to video-mode from command mode and thus dcs functions
should not be called anymore at this point.
This caused a non-working display for me, when trying to enable
the rk3576 dsi controller using a display using this controller.
Therefore move the display_on/off calls the more appropriate
prepare/unprepare callbacks.
The prepare function turns on the power-supply regulator first, when
preparing the display. But in an error case, just returns the error
code, but does not power off the regulator again, fix that.
There are two registers filled in when reading data from
pcode besides the mailbox itself. Currently, we allow a NULL
value for the second of these two (data1) and assume the first
is defined. However, many of the routines that are calling
this function assume that pcode will ignore the value being
passed in and so leave that first value (data0) defined but
uninitialized. To be safe, make sure this value is always
initialized to something (0 generally) in the event pcode
behavior changes and starts using this value.
Due to a hardware errata, CN10k and earlier Octeon silicon series,
the hardware may incorrectly assert XOFF on certain channels during
reset. As a workaround, a write operation to the NIX_AF_RX_CHANX_CFG
register can be performed to broadcast XON signals on the affected
channels
Just add fixed struct size validations for UAC2 and UAC3 effect
units. The descriptor has a variable-length array, so it should be
validated with a proper function later once when the unit is really
parsed and used by the driver (currently only referred partially for
the input terminal parsing).
The dev_err_probe() doesn't do anything when error is '-ENOMEM'.
Make the following two changes:
(1) Replace -ENOMEM with -ENOSPC in max3100_probe().
(2) Just return -ENOMEM instead in max310x_probe().
The error handling path in pci_p2pdma_add_resource() contains a bug in its
`pgmap_free` label.
Memory is allocated for the `p2p_pgmap` struct, and the pointer is stored
in `p2p_pgmap`. However, the error path calls devm_kfree() with `pgmap`,
which is a pointer to a member field within the `p2p_pgmap` struct, not the
base pointer of the allocation.
Correct the bug by passing the correct base pointer, `p2p_pgmap`, to
devm_kfree().
As reported, on-disk footer.ino and footer.nid is the same and
out-of-range, let's add sanity check on f2fs_alloc_nid() to detect
any potential corruption in free_nid_list.
Initially, trace_sock_exceed_buf_limit() was invoked when
__sk_mem_raise_allocated() failed due to the memcg limit or the
global limit.
However, commit d6f19938eb031 ("net: expose sk wmem in
sock_exceed_buf_limit tracepoint") somehow suppressed the event
only when memcg failed to charge for SK_MEM_RECV, although the
memcg failure for SK_MEM_SEND still triggers the event.
The stmmac_rx function would previously set skb->ip_summed to
CHECKSUM_UNNECESSARY if hardware checksum offload (CoE) was enabled
and the packet was of a known IP ethertype.
However, this logic failed to check if the hardware had actually
reported a checksum error. The hardware status, indicating a header or
payload checksum failure, was being ignored at this stage. This could
cause corrupt packets to be passed up the network stack as valid.
This patch corrects the logic by checking the `csum_none` status flag,
which is set when the hardware reports a checksum error. If this flag
is set, skb->ip_summed is now correctly set to CHECKSUM_NONE,
ensuring the kernel's network stack will perform its own validation and
properly handle the corrupt packet.
When removing a nexthop, commit 90f33bffa382 ("nexthops: don't modify published nexthop groups") added a
call to synchronize_rcu() (later changed to _net()) to make sure
everyone sees the new nexthop-group before the rtnl-lock is released.
When one wants to delete a large number of groups and nexthops, it is
fastest to first flush the groups (ip nexthop flush groups) and then
flush the nexthops themselves (ip -6 nexthop flush). As that way the
groups don't need to be rebalanced.
However, `ip -6 nexthop flush` will still take a long time if there is
a very large number of nexthops because of the call to
synchronize_net(). Now, if there are no more groups, there is no point
in calling synchronize_net(). So, let's skip that entirely by checking
if nh->grp_list is empty.
This gives us a nice speedup:
BEFORE:
=======
$ time sudo ip -6 nexthop flush
Dump was interrupted and may be inconsistent.
Flushed 2097152 nexthops
real 1m45.345s
user 0m0.001s
sys 0m0.005s
$ time sudo ip -6 nexthop flush
Dump was interrupted and may be inconsistent.
Flushed 4194304 nexthops
real 3m10.430s
user 0m0.002s
sys 0m0.004s
AFTER:
======
$ time sudo ip -6 nexthop flush
Dump was interrupted and may be inconsistent.
Flushed 2097152 nexthops
real 0m17.545s
user 0m0.003s
sys 0m0.003s
$ time sudo ip -6 nexthop flush
Dump was interrupted and may be inconsistent.
Flushed 4194304 nexthops
real 0m35.823s
user 0m0.002s
sys 0m0.004s
Signed-off-by: Christoph Paasch <cpaasch@openai.com> Reviewed-by: Ido Schimmel <idosch@nvidia.com> Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org> Reviewed-by: Eric Dumazet <edumazet@google.com> Reviewed-by: David Ahern <dsahern@kernel.org> Link: https://patch.msgid.link/20250816-nexthop_dump-v2-2-491da3462118@openai.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
misc_open() may request module for miscdevice with dynamic minor, which
is meaningless since:
- The dynamic minor allocated is unknown in advance without registering
miscdevice firstly.
- Macro MODULE_ALIAS_MISCDEV() is not applicable for dynamic minor.
Fix by only requesting module for miscdevice with fixed minor.
For miscdevice who wants dynamic minor, it may fail to be registered again
without reinitialization after being de-registered, which is illustrated
by kunit test case miscdev_test_dynamic_reentry() newly added.
There is a real case found by cascardo when a part of minor range were
contained by range [0, 255):
1) wmi/dell-smbios registered minor 122, and acpi_thermal_rel registered
minor 123
2) unbind "int3400 thermal" driver from its device, this will de-register
acpi_thermal_rel
3) rmmod then insmod dell_smbios again, now wmi/dell-smbios is using minor
123
4) bind the device to "int3400 thermal" driver again, acpi_thermal_rel
fails to register.
Some drivers may reuse the miscdevice structure after they are deregistered
If the intention is to allocate a dynamic minor, if the minor number is not
reset to MISC_DYNAMIC_MINOR before calling misc_register(), it will try to
register a previously dynamically allocated minor number, which may have
been registered by a different driver.
One such case is the acpi_thermal_rel misc device, registered by the
int3400 thermal driver. If the device is unbound from the driver and later
bound, if there was another dynamic misc device registered in between, it
would fail to register the acpi_thermal_rel misc device. Other drivers
behave similarly.
Actually, this kind of issue is prone to happen if APIs
misc_register()/misc_deregister() are invoked by driver's
probe()/remove() separately.
Instead of fixing all the drivers, just reset the minor member to
MISC_DYNAMIC_MINOR in misc_deregister() in case it was a dynamically
allocated minor number, as error handling of misc_register() does.
Mark dm error as DM_TARGET_PASSES_INTEGRITY so that it can be stacked on
top of PI capable devices. The claim is strictly speaking as lie as dm
error fails all I/O and doesn't pass anything on, but doing the same for
integrity I/O work just fine :)
This helps to make about two dozen xfstests test cases pass on PI capable
devices.
In WoWLAN net-detect mode, the firmware periodically performs scans
and sends scan reports via C2H, which driver does not need. These
unnecessary C2H events cause firmware watchdog timeout, leading
to unexpected wakeups and SER 0x2599 on 8922AE.
On some Qualcomm platforms such as QCS6490-RB3Gen2, the multiple
WSA8830/WSA8835 speaker amplifiers share a common reset (shutdown) GPIO.
To handle such scenario, use the reset controller framework and its
"reset-gpio" driver to handle such case. This allows proper handling
of all WSA883x speaker amplifiers on QCS6490-RB3Gen2 board.
Signed-off-by: Mohammad Rafi Shaik <quic_mohs@quicinc.com> Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org> Reviewed-by: Srinivas Kandagatla <srinivas.kandagatla@oss.qualcomm.com> Link: https://patch.msgid.link/20250815172353.2430981-3-mohammad.rafi.shaik@oss.qualcomm.com Signed-off-by: Mark Brown <broonie@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
By sharing 'name' directly, tearing down components may lead to
use-after-free errors. Duplicate the name to avoid that.
At the same time, update the order of operations - since commit cee28113db17 ("ASoC: dmaengine_pcm: Allow passing component name via
config") the framework does not override component->name if set before
invoking the initializer.
Reviewed-by: Amadeusz Sławiński <amadeuszx.slawinski@linux.intel.com> Signed-off-by: Cezary Rojewski <cezary.rojewski@intel.com> Link: https://patch.msgid.link/20250818104126.526442-4-cezary.rojewski@intel.com Signed-off-by: Mark Brown <broonie@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
This fix is already present in f_ecm.c and was never
propagated to f_ncm.c
When creating multiple NCM ethernet devices
on a composite usb gadget device
each MAC address on the HOST side will be identical.
Having the same MAC on different network interfaces is bad.
This fix updates the MAC address inside the
ncm_strings_defs global during the ncm_bind call.
This ensures each device has a unique MAC.
In f_ecm.c ecm_string_defs is updated in the same way.
The defunct MAC assignment in ncm_alloc has been removed.
ADC calibration might fail because of the noise on reference voltage.
To avoid calibration fail, need to meet the following requirement:
ADC reference voltage Noise < 1.8V * 1/2^ENOB
For the case which the ADC reference voltage on board do not meet
the requirement, still load the calibrated values, so ADC can also
work but maybe not that accurate.
Signed-off-by: Haibo Chen <haibo.chen@nxp.com> Reviewed-by: Frank Li <Frank.Li@nxp.com> Reviewed-by: Primoz Fiser <primoz.fiser@norik.com> Link: https://patch.msgid.link/20250812-adc-v2-2-0260833f13b8@nxp.com Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
When a poison is consumed on the guest before the guest receives the host's poison creation msg, a corner case may occur to have poison_handler complete processing earlier than it should to cause the guest to hang waiting for the req_bad_pages reply during a VF FLR, resulting in the VM becoming inaccessible in stress tests.
To fix this issue, this patch refactored the mailbox sequence by seperating the bad_page_work into two parts req_bad_pages_work and handle_bad_pages_work.
Old sequence:
1.Stop data exchange work
2.Guest sends MB_REQ_RAS_BAD_PAGES to host and keep polling for IDH_RAS_BAD_PAGES_READY
3.If the IDH_RAS_BAD_PAGES_READY arrives within timeout limit, re-init the data exchange region for updated bad page info
else timeout with error message
New sequence:
req_bad_pages_work:
1.Stop data exhange work
2.Guest sends MB_REQ_RAS_BAD_PAGES to host
Once Guest receives IDH_RAS_BAD_PAGES_READY event
handle_bad_pages_work:
3.re-init the data exchange region for updated bad page info
HMM assumes that pages have READ permissions by default. Inside
svm_range_validate_and_map, we add READ permissions then add WRITE
permissions if the VMA isn't read-only. This will conflict with regions
that only have PROT_WRITE or have PROT_NONE. When that happens,
svm_range_restore_work will continue to retry, silently, giving the
impression of a hang if pr_debug isn't enabled to show the retries..
If pages don't have READ permissions, simply unmap them and continue. If
they weren't mapped in the first place, this would be a no-op. Since x86
doesn't support write-only, and PROT_NONE doesn't allow reads or writes
anyways, this will allow the svm range validation to continue without
getting stuck in a loop forever on mappings we can't use with HMM.
Signed-off-by: Kent Russell <kent.russell@amd.com> Reviewed-by: Felix Kuehling <felix.kuehling@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
If a amdgpu_bo_va is fpriv->prt_va, the bo of this one is always NULL.
So, such kind of amdgpu_bo_va should be updated separately before
amdgpu_vm_handle_moved.
Detecting the monitor for DisplayPort targets is more complicated than
just reading the HPD pin level: it requires reading the DPCD in order to
check what kind of device is attached to the port and whether there is
an actual display attached.
In order to let DRM framework handle such configurations, disable
DRM_BRIDGE_OP_DETECT for dp-connector devices, letting the actual DP
driver perform detection. This still keeps DRM_BRIDGE_OP_HPD enabled, so
it is valid for the bridge to report HPD events.
Currently inside the kernel there are only two targets which list
hpd-gpios for dp-connector devices: arm64/qcom/qcs6490-rb3gen2 and
arm64/qcom/sa8295p-adp. Both should be fine with this change.
The already fixed bug in SDL only affected conditional effects. This
should fix FFB in Forza Horizion 4/5 on Moza Devices as Forza Horizon
flips the constant force direction instead of using negative magnitude
values.
Changing the direction in the effect directly in pidff_upload_effect()
would affect it's value in further operations like comparing to the old
effect and/or just reading the effect values in the user application.
This, in turn, would lead to constant PID_SET_EFFECT spam as the effect
direction would constantly not match the value that's set by the
application.
This way, it's still transparent to any software/API.
Only affects conditional effects now so it's better for it to explicitly
state that in the name. If any HW ever needs fixed direction for other
effects, we'll add more quirks.
In certain scenarios, it is possible for multiple cache flushes to be
requested before the previous one completes. This patch introduces the
cache_flush_lock mutex to serialize these operations and ensure that
any requested cache flushes are completed instead of dropped.
Reviewed-by: Liviu Dudau <liviu.dudau@arm.com> Co-developed-by: Dennis Tsiang <dennis.tsiang@arm.com> Signed-off-by: Dennis Tsiang <dennis.tsiang@arm.com> Signed-off-by: Karunika Choo <karunika.choo@arm.com> Reviewed-by: Steven Price <steven.price@arm.com> Signed-off-by: Steven Price <steven.price@arm.com> Link: https://lore.kernel.org/r/20250807162633.3666310-6-karunika.choo@arm.com Signed-off-by: Sasha Levin <sashal@kernel.org>
syzbot is reporting that imon has three problems which result in
hung tasks due to forever holding device lock [1].
First problem is that when usb_rx_callback_intf0() once got -EPROTO error
after ictx->dev_present_intf0 became true, usb_rx_callback_intf0()
resubmits urb after printk(), and resubmitted urb causes
usb_rx_callback_intf0() to again get -EPROTO error. This results in
printk() flooding (RCU stalls).
Alan Stern commented [2] that
In theory it's okay to resubmit _if_ the driver has a robust
error-recovery scheme (such as giving up after some fixed limit on the
number of errors or after some fixed time has elapsed, perhaps with a
time delay to prevent a flood of errors). Most drivers don't bother to
do this; they simply give up right away. This makes them more
vulnerable to short-term noise interference during USB transfers, but in
reality such interference is quite rare. There's nothing really wrong
with giving up right away.
but imon has a poor error-recovery scheme which just retries forever;
this behavior should be fixed.
Since I'm not sure whether it is safe for imon users to give up upon any
error code, this patch takes care of only union of error codes chosen from
modules in drivers/media/rc/ directory which handle -EPROTO error (i.e.
ir_toy, mceusb and igorplugusb).
Second problem is that when usb_rx_callback_intf0() once got -EPROTO error
before ictx->dev_present_intf0 becomes true, usb_rx_callback_intf0() always
resubmits urb due to commit 8791d63af0cf ("[media] imon: don't wedge
hardware after early callbacks"). Move the ictx->dev_present_intf0 test
introduced by commit 6f6b90c9231a ("[media] imon: don't parse scancodes
until intf configured") to immediately before imon_incoming_packet(), or
the first problem explained above happens without printk() flooding (i.e.
hung task).
Third problem is that when usb_rx_callback_intf0() is not called for some
reason (e.g. flaky hardware; the reproducer for this problem sometimes
prevents usb_rx_callback_intf0() from being called),
wait_for_completion_interruptible() in send_packet() never returns (i.e.
hung task). As a workaround for such situation, change send_packet() to
wait for completion with timeout of 10 seconds.
Fix field-spanning memcpy warnings in ah6_output() and
ah6_output_done() where extension headers are copied to/from IPv6
address fields, triggering fortify-string warnings about writes beyond
the 16-byte address fields.
memcpy: detected field-spanning write (size 40) of single field "&top_iph->saddr" at net/ipv6/ah6.c:439 (size 16)
WARNING: CPU: 0 PID: 8838 at net/ipv6/ah6.c:439 ah6_output+0xe7e/0x14e0 net/ipv6/ah6.c:439
The warnings are false positives as the extension headers are
intentionally placed after the IPv6 header in memory. Fix by properly
copying addresses and extension headers separately, and introduce
helper functions to avoid code duplication.
`offset` is a common field name, yet using it triggers a build error due
to the conflict between the uppercased field constant (which becomes
`OFFSET` in this case) containing the bitrange of the field, and the
`OFFSET` constant constaining the offset of the register.
Fix this by adding `_RANGE` the field's range constant to avoid the
name collision.
[acourbot@nvidia.com: fix merge conflict due to switch from `as u32` to
`u32::from`.]
Reported-by: Timur Tabi <ttabi@nvidia.com> Reviewed-by: Daniel Almeida <daniel.almeida@collabora.com> Reviewed-by: Lyude Paul <lyude@redhat.com> Link: https://lore.kernel.org/r/20250718-nova-regs-v2-3-7b6a762aa1cd@nvidia.com Signed-off-by: Alexandre Courbot <acourbot@nvidia.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Add a NULL check before accessing the 'vccqx' pointer to prevent invalid
memory access. This ensures that the function safely handles cases where
'vccq' and 'vccq2' are not initialized, improving the robustness of the
power management code.
Signed-off-by: Alice Chao <alice.chao@mediatek.com> Reviewed-by: Peter Wang <peter.wang@mediatek.com> Signed-off-by: Peter Wang <peter.wang@mediatek.com> Link: https://lore.kernel.org/r/20250811131423.3444014-11-peter.wang@mediatek.com Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Modify the reset sequence to ensure that the device reset pin is set low
before the host is disabled. This change enhances the stability of the
reset process by ensuring the correct order of operations.
Assign power mode userdata settings before transitioning to FASTAUTO
power mode. This ensures that default timeout values are set for various
parameters, enhancing the reliability and performance of the power mode
change process.
Signed-off-by: Alice Chao <alice.chao@mediatek.com> Reviewed-by: Peter Wang <peter.wang@mediatek.com> Signed-off-by: Peter Wang <peter.wang@mediatek.com> Link: https://lore.kernel.org/r/20250811131423.3444014-7-peter.wang@mediatek.com Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Address a failure in switching to PWM mode by ensuring proper
configuration of power modes and adaptation settings. The changes
include checks for SLOW_MODE and adjustments to the desired working mode
and adaptation configuration based on the device's power mode and
hardware version.
Move the configuration of the Auto-Hibern8 (AHIT) timer from the
post-link stage to the 'fixup_dev_quirks' function. This change allows
setting the AHIT based on the vendor requirements:
(a) Samsung: 3.5 ms
(b) Micron: 2 ms
(c) Others: 1 ms
Additionally, the clock gating timer is adjusted based on the AHIT
scale, with a maximum setting of 10 ms. This ensures that the clock
gating delay is appropriately configured to match the AHIT settings.
The following Vitesse/Microsemi/Microchip PHYs, among those supported by
this driver, have the host interface configurable as SGMII or QSGMII:
- VSC8504
- VSC8514
- VSC8552
- VSC8562
- VSC8572
- VSC8574
- VSC8575
- VSC8582
- VSC8584
All these PHYs are documented to have bit 7 of "MAC SerDes PCS Control"
as "MAC SerDes ANEG enable".
Out of these, I could test the VSC8514 quad PHY in QSGMII. This works
both with the in-band autoneg on and off, on the NXP LS1028A-RDB and
T1040-RDB boards.
Notably, the bit is sticky (survives soft resets), so giving Linux the
tools to read and modify this settings makes it robust to changes made
to it by previous boot layers (U-Boot).
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com> Reviewed-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk> Link: https://patch.msgid.link/20250813074454.63224-1-vladimir.oltean@nxp.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
If a backup port is configured for a bridge port, the bridge will
redirect known unicast traffic towards the backup port when the primary
port is administratively up but without a carrier. This is useful, for
example, in MLAG configurations where a system is connected to two
switches and there is a peer link between both switches. The peer link
serves as the backup port in case one of the switches loses its
connection to the multi-homed system.
In order to avoid flooding when the primary port loses its carrier, the
bridge does not flush dynamic FDB entries pointing to the port upon STP
disablement, if the port has a backup port.
The above means that known unicast traffic destined to the primary port
will be blackholed when the port is put administratively down, until the
FDB entries pointing to it are aged-out.
Given that the current behavior is quite weird and unlikely to be
depended on by anyone, amend the bridge to redirect to the backup port
also when the primary port is administratively down and not only when it
does not have a carrier.
The change is motivated by a report from a user who expected traffic to
be redirected to the backup port when the primary port was put
administratively down while debugging a network issue.
Reviewed-by: Petr Machata <petrm@nvidia.com> Signed-off-by: Ido Schimmel <idosch@nvidia.com> Acked-by: Nikolay Aleksandrov <razor@blackwall.org> Link: https://patch.msgid.link/20250812080213.325298-2-idosch@nvidia.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Ever since uevent support was added for AER and EEH with commit 856e1eb9bdd4 ("PCI/AER: Add uevents in AER and EEH error/resume"), it
reported PCI_ERS_RESULT_NONE as uevent when recovery begins.
Commit 7b42d97e99d3 ("PCI/ERR: Always report current recovery status for
udev") subsequently amended AER to report the actual return value of
error_detected().
Make the same change to EEH to align it with AER and s390.
After commit 84a9582fd203 ("serial: core: Start managing serial controllers
to enable runtime PM") serial drivers need to provide a device in
struct uart_port.dev otherwise an oops happens. To fix this issue
for ip22zilog driver switch driver to a platform driver and setup
the serial device in sgi-ip22 code.
The thunderbolt driver sets up device link dependencies from hotplug ports
to the Host Router (aka Native Host Interface, NHI). When resuming from
system sleep, this allows the Host Router to re-establish tunnels to
attached Thunderbolt devices before the hotplug ports resume.
To identify the hotplug ports, the driver utilizes the is_hotplug_bridge
flag which also encompasses ACPI slots handled by the ACPI hotplug driver.
Thunderbolt hotplug ports are always Hot-Plug Capable PCIe ports, so it is
more apt to identify them with the is_pciehp flag.
Similarly, hotplug ports on older Thunderbolt controllers have broken MSI
support and are quirked to use legacy INTx interrupts instead. The quirk
identifies them with is_hotplug_bridge, even though all affected ports are
also matched by is_pciehp. So use is_pciehp here as well.
In the past %pK was preferable to %p as it would not leak raw pointer
values into the kernel log.
Since commit ad67b74d2469 ("printk: hash addresses printed with %p")
the regular %p has been improved to avoid this issue.
Furthermore, restricted pointers ("%pK") were never meant to be used
through printk(). They can still unintentionally leak raw pointers or
acquire sleeping locks in atomic contexts.
Switch to the regular pointer formatting which is safer and
easier to reason about.
There are still a few users of %pK left, but these use it through seq_file,
for which its usage is safe.
Signed-off-by: Thomas Weißschuh <thomas.weissschuh@linutronix.de> Acked-by: Przemek Kitszel <przemyslaw.kitszel@intel.com> Reviewed-by: Aleksandr Loktionov <aleksandr.loktionov@intel.com> Reviewed-by: Simon Horman <horms@kernel.org> Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de> Reviewed-by: Jacob Keller <jacob.e.keller@intel.com> Link: https://patch.msgid.link/20250811-restricted-pointers-net-v5-1-2e2fdc7d3f2c@linutronix.de Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
stmmac_hw_setup() may return 0 on success and an appropriate negative
integer as defined in errno.h file on failure, just check it and then
return early if failed in stmmac_resume().
Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn> Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com> Reviewed-by: Huacai Chen <chenhuacai@loongson.cn> Link: https://patch.msgid.link/20250811073506.27513-2-yangtiezhu@loongson.cn Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
emulate_vsyscall() expects to see X86_PF_INSTR in PFEC on a vsyscall
page fault, but the CPU does not report X86_PF_INSTR if neither
X86_FEATURE_NX nor X86_FEATURE_SMEP are enabled.
X86_FEATURE_NX should be enabled on nearly all 64-bit CPUs, except for
early P4 processors that did not support this feature.
Instead of explicitly checking for X86_PF_INSTR, compare the fault
address to RIP.
On machines with X86_FEATURE_NX enabled, issue a warning if RIP is equal
to fault address but X86_PF_INSTR is absent.
After a Fatal Error has been reported by a device and has been recovered
through a Secondary Bus Reset, AER updates the device's error_state to
pci_channel_io_normal before invoking its driver's ->resume() callback.
By contrast, EEH updates the error_state earlier, namely after resetting
the device and before invoking its driver's ->slot_reset() callback.
Commit c58dc575f3c8 ("powerpc/pseries: Set error_state to
pci_channel_io_normal in eeh_report_reset()") explains in great detail
that the earlier invocation is necessitated by various drivers checking
accessibility of the device with pci_channel_offline() and avoiding
accesses if it returns true. It returns true for any other error_state
than pci_channel_io_normal.
The device should be accessible already after reset, hence the reasoning
is that it's safe to update the error_state immediately afterwards.
This deviation between AER and EEH seems problematic because drivers
behave differently depending on which error recovery mechanism the
platform uses. Three drivers have gone so far as to update the
error_state themselves, presumably to work around AER's behavior.
For consistency, amend AER to update the error_state at the same recovery
steps as EEH. Drop the now unnecessary workaround from the three drivers.
Keep updating the error_state before ->resume() in case ->error_detected()
or ->mmio_enabled() return PCI_ERS_RESULT_RECOVERED, which causes
->slot_reset() to be skipped. There are drivers doing this even for Fatal
Errors, e.g. mhi_pci_error_detected().
In rare servicing scenarios, the hardware may delete the
state of the GSI QP, preventing it from generating CQEs
for pending send WRs. Since WRs submitted to the GSI QP
hold CM resources, the device cannot be removed until
those WRs are completed. This patch marks all pending
send WRs as failed, allowing the GSI QP to release the CM
resources and enabling safe device removal.
TIDSS uses crtc_* fields to propagate its registers and set the
clock rates. So set the CRTC modesetting timing parameters with
the adjusted mode when needed, to set correct values.
Cc: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com> Signed-off-by: Jayesh Choudhary <j-choudhary@ti.com> Link: https://lore.kernel.org/r/20250624080402.302526-1-j-choudhary@ti.com Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
While the cdns-dsi does not support DSI burst mode, the burst mode is
essentially DSI event mode with more versatile clocking and timings.
Thus cdns-dsi doesn't need to fail if the DSI peripheral driver requests
MIPI_DSI_MODE_VIDEO_BURST.
In my particular use case, this allows the use of ti-sn65dsi83 driver.
The driver tries to calculate the value for REG_WAKEUP_TIME. However,
the calculation itself is not correct, and to add on it, the resulting
value is almost always larger than the field's size, so the actual
result is more or less random.
According to the docs, figuring out the value for REG_WAKEUP_TIME
requires HW characterization and there's no way to have a generic
algorithm to come up with the value. That doesn't help at all...
However, we know that the value must be smaller than the line time, and,
at least in my understanding, the proper value for it is quite small.
Testing shows that setting it to 1/10 of the line time seems to work
well. All video modes from my HDMI monitor work with this algorithm.
Hopefully we'll get more information on how to calculate the value, and
we can then update this.
Use the crtc_* fields from drm_display_mode, instead of the "logical"
fields. This shouldn't change anything in practice, but afaiu the crtc_*
fields are the correct ones to use here.
The v4l2_fh initialized and added in vpu_v4l2_open() is delete and
cleaned up when the last reference to the vpu_inst is released. This may
happen later than at vpu_v4l2_close() time.
Not deleting and cleaning up the v4l2_fh when closing the file handle to
the video device is not ideal, as the v4l2_fh will still be present in
the video device's fh_list, and will store a copy of events queued to
the video device. There may also be other side effects of keeping alive
an object that represents an open file handle after the file handle is
closed.
The v4l2_fh instance is embedded in the vpu_inst structure, and is
accessed in two different ways:
- in vpu_notify_eos() and vpu_notify_source_change(), to queue V4L2
events to the file handle ; and
- through the driver to access the v4l2_fh.m2m_ctx pointer.
The v4l2_fh.m2m_ctx pointer is not touched by v4l2_fh_del() and
v4l2_fh_exit(). It is set to NULL by the driver when closing the file
handle, in vpu_v4l2_close().
The vpu_notify_eos() and vpu_notify_source_change() functions are called
in vpu_set_last_buffer_dequeued() and vdec_handle_resolution_change()
respectively, only if the v4l2_fh.m2m_ctx pointer is not NULL. There is
therefore a guarantee that no new event will be queued to the v4l2_fh
after vpu_v4l2_close() destroys the m2m_ctx.
The vpu_notify_eos() function is also called from vpu_vb2_buf_finish(),
which is guaranteed to be called for all queued buffers when
vpu_v4l2_close() calls v4l2_m2m_ctx_release(), and will not be called
later.
It is therefore safe to assume that the driver will not touch the
v4l2_fh, except to check the m2m_ctx pointer, after vpu_v4l2_close()
destroys the m2m_ctx. We can safely delete and cleanup the v4l2_fh
synchronously in vpu_v4l2_close().
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com> Reviewed-by: Ming Qian <ming.qian@oss.nxp.com> Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
The ivtv driver has a structure named ivtv_open_id that models an open
file handle for the device. It embeds a v4l2_fh instance for file
handles that correspond to a V4L2 video device, and stores a pointer to
that v4l2_fh in struct ivtv_stream to identify which open file handle
owns a particular stream.
In addition to video devices, streams can be owned by ALSA PCM devices.
Those devices do not make use of the v4l2_fh instance for obvious
reasons, but the snd_ivtv_pcm_capture_open() function still initializes
a "fake" v4l2_fh for the sole purpose of using it as an open file handle
identifier. The v4l2_fh is not properly destroyed when the ALSA PCM
device is closed, leading to possible resource leaks.
Fortunately, the v4l2_fh instance pointed to by ivtv_stream is not
accessed, only the pointer value is used for comparison. Replace it with
a pointer to the ivtv_open_id structure that embeds the v4l2_fh, and
don't initialize the v4l2_fh for ALSA PCM devices.
projects / thirdparty / kernel / stable.git / log
