pdsc_check_pci_health() queues pci_reset_work when it sees a broken PCI
connection, and nothing cancels it. When the PCI core starts AER
recovery, pdsc_pci_error_detected() runs pdsc_reset_prepare() and
recovers the device, but a pci_reset_work queued just before is left
pending. If it runs after recovery released the device lock, it resets a
device the driver now considers healthy, bouncing the link for no reason.
Cancel pci_reset_work in pdsc_pci_error_detected() after
pdsc_reset_prepare(), which has already stopped the health thread so it
cannot requeue the work. cancel_work_sync() is safe under the device
lock here because pdsc_pci_reset_thread() uses pci_try_reset_function(),
which returns instead of blocking on the lock. Only PFs initialize
pci_reset_work, so guard the cancel with !is_virtfn.
Fixes: 81665adf25 ("pds_core: Fix pdsc_check_pci_health function to use work thread")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Closes: https://sashiko.dev/#/patchset/20260714180223.1642792-2-nikhil.rao%40amd.com?part=1
Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://patch.msgid.link/20260727170030.361116-1-nikhil.rao@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Commit d9407ff118 ("pds_core: Prevent health thread from running
during reset/remove") stops the health thread with cancel_work_sync()
before a reset, but a devcmd timeout during pdsc_fw_down() re-queues
health_work, so pdsc_health_thread() runs again mid-reset and double
allocates the core DMA queues via pdsc_fw_up().
Only the reset path is affected: on remove PDSC_S_STOPPING_DRIVER gates
the health thread and the workqueue is destroyed.
Use disable_work_sync() to cancel health_work and block further
queue_work() on it, and enable_work() in pdsc_restart_health_thread() to
re-allow it after the reset.
disable_work_sync() keeps a disable depth, so every disable must be
matched by one enable. pdsc_reset_prepare() stops the health thread and
pdsc_reset_done() restarts it. On the AER path pdsc_pci_error_detected()
calls pdsc_reset_prepare(), then pdsc_pci_error_resume() re-inits via
pci_reset_function_locked() (pds_core has no .slot_reset handler), which
runs the pair again - stopping the thread twice but restarting it once.
Gate the disable and enable on a health_stopped flag so each fires at
most once per stopped/running transition.
Fixes: d9407ff118 ("pds_core: Prevent health thread from running during reset/remove")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Closes: https://sashiko.dev/#/patchset/20260629200358.2626129-1-nikhil.rao%40amd.com?part=2
Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://patch.msgid.link/20260727164548.359562-1-nikhil.rao@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
pdsc_init_pf() does not check whether create_singlethread_workqueue()
succeeded.
Fail probe on failure. The workqueue is set up before the timer and
mutexes, so its failure path must unwind only the earlier setup.
Fixes: c2dbb09043 ("pds_core: health timer and workqueue")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Closes: https://sashiko.dev/#/patchset/20260629200358.2626129-1-nikhil.rao%40amd.com?part=2
Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Link: https://patch.msgid.link/20260714212713.1788438-1-nikhil.rao@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Two paths add or delete the same slot (pf->vfs[vf_id].padev): a VF's
pdsc_reset_done() and the PF's devlink enable_vnet/disable_vnet handler.
They serialize on config_lock, but neither guards the slot under it
correctly.
add() registers and stores a new auxiliary device without first checking
the slot, so a second add of an already-populated slot leaks the first
device. del() makes that check outside config_lock, so two concurrent
dels can both pass it; the first clears the slot, and the second
dereferences a NULL pointer.
Check and update the slot under config_lock in both paths.
Fixes: b699bdc720 ("pds_core: specify auxiliary_device to be created")
Reported-by: sashiko-bot@kernel.org # Running on a local machine
Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Reviewed-by: Brett Creeley <brett.creeley@amd.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Link: https://patch.msgid.link/20260714210745.1785625-1-nikhil.rao@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
pdsc_process_adminq() and pdsc_process_notifyq() decide a completion is
valid from its ownership field - the color bit for the adminq, the event
id for the notifyq - then read the rest of the descriptor, with no
barrier in between.
On a weakly ordered architecture the CPU may read the payload first. Add
dma_rmb() between the ownership read and the payload reads.
Fixes: 7e82a8745b ("pds_core: Prevent race issues involving the adminq")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Closes: https://sashiko.dev/#/patchset/20260629200358.2626129-1-nikhil.rao%40amd.com?part=2
Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Reviewed-by: Eric Joyner <eric.joyner@amd.com>
Link: https://patch.msgid.link/20260714204145.1782390-1-nikhil.rao@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
pdsc_adminq_wait_and_dec_once_unused() busy-waits for adminq_refcnt to
drop to one:
while (!refcount_dec_if_one(&pdsc->adminq_refcnt))
cpu_relax();
The refcount is held by pdsc_adminq_post() for the duration of an
in-flight command, which can wait up to devcmd_timeout seconds
(PDS_CORE_DEVCMD_TIMEOUT is 5) for the hardware to complete. cpu_relax()
is not a reschedule point, so on a non-preemptible kernel this loop can
spin on the CPU for several seconds, starving other tasks on that core.
Add cond_resched() to the loop so the waiter yields to other runnable
tasks while it polls, keeping cpu_relax() as the busy-wait hint between
checks.
Fixes: 7e82a8745b ("pds_core: Prevent race issues involving the adminq")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Closes: https://sashiko.dev/#/patchset/20260629200358.2626129-1-nikhil.rao%40amd.com?part=2
Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Reviewed-by: Eric Joyner <eric.joyner@amd.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Link: https://patch.msgid.link/20260714201456.1776153-1-nikhil.rao@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
In pdsc_remove(), the workqueue is destroyed before pdsc_teardown()
is called. This ordering allows two paths to queue work on the
destroyed workqueue:
1. If pdsc_teardown() -> pdsc_devcmd_reset() times out, the error
path in pdsc_devcmd_locked() queues health_work.
2. A NotifyQ event can trigger the ISR and queue work before free_irq()
is called in pdsc_teardown().
Fix by moving destroy_workqueue() after pdsc_teardown() so the
workqueue outlives every queuer; destroy_workqueue() then flushes any
work still pending.
Draining the queued work also requires ordering the teardown so the
resources that work touches are freed last:
- In pdsc_qcq_free(), after freeing the interrupt, cancel_work_sync()
the queue's work and only then clear qcq->intx, so
pdsc_process_adminq()'s read of qcq->intx for interrupt-credit
return cannot race with the clear.
- Free adminqcq before notifyqcq: the shared adminq ISR is released
when adminqcq is freed, and the adminq work accesses notifyqcq, so
both must be stopped before notifyqcq is freed.
Fixes: 01ba61b55b ("pds_core: Add adminq processing and commands")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Closes: https://patchwork.kernel.org/comment/27002369/
Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Link: https://patch.msgid.link/20260714180223.1642792-3-nikhil.rao@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
pci_reset_function() acquires device_lock before performing the reset.
pdsc_remove() is called by the PCI core with device_lock already held.
If pdsc_pci_reset_thread() is running when pdsc_remove() is called,
destroy_workqueue() will block waiting for the work to complete, while
the work is blocked waiting for device_lock - deadlock.
Use pci_try_reset_function() which uses pci_dev_trylock() internally.
This acquires both the device lock and the PCI config access lock
without blocking - if either lock is contended, it returns -EAGAIN
immediately. This avoids the deadlock while also ensuring proper
config space access serialization during the reset.
The pci_dev_get/put calls are also removed as they were unnecessary -
the driver-owned workqueue is destroyed in pdsc_remove(), guaranteeing
the work completes before remove returns. The PCI core holds its
reference to pci_dev throughout the entire unbind sequence.
Fixes: 81665adf25 ("pds_core: Fix pdsc_check_pci_health function to use work thread")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Closes: https://patchwork.kernel.org/comment/27002369/
Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Reviewed-by: Harshitha Ramamurthy <hramamurthy@google.com>
Link: https://patch.msgid.link/20260714180223.1642792-2-nikhil.rao@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
MAC_AUTO_SW (VR_MII_DIG_CTRL1 bit 9) enables automatic XPCS speed
mode switching after CL37 auto-negotiation and is only meaningful in
SGMII MAC mode. The original code unconditionally set this bit on
every call to xgbe_an37_set(), including when called from
xgbe_an37_disable() with enable=false. This left MAC_AUTO_SW=1 after
AN was disabled, causing the XPCS to autonomously switch speed from
stale AN state during subsequent mode changes, breaking SGMII speed
negotiation on 1G copper SFP modules.
Patrick: This was breaking negotiation for all 1G SFP modules,
not just copper modules.
Fixes: 42fd432fe6 ("amd-xgbe: align CL37 AN sequence as per databook")
Reported-by: Patrick Oppenlander <patrick.oppenlander@gmail.com>
Link: https://lore.kernel.org/netdev/CAEg67GmFS0Q4oSZkz8zWdOzckSth9_vBPiOy6a7-d697C2w2Xg@mail.gmail.com
Signed-off-by: Prashanth Kumar KR <PrashanthKumar.K.R@amd.com>
Tested-by: Patrick Oppenlander <patrick.oppenlander@gmail.com>
Link: https://patch.msgid.link/20260709095006.3683940-1-prashanthkumar.k.r@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The legacy firmware update path does not support per-component updates.
If a user specifies a component parameter with devlink flash, reject
the request with -EOPNOTSUPP rather than silently ignoring the component
parameter and flashing the entire firmware image.
Fixes: 49ce92fbee ("pds_core: add FW update feature to devlink")
Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260708163649.128620-1-nikhil.rao@amd.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
au1000_close() calls free_irq() while aup->lock is still held with
spin_lock_irqsave(). free_irq() can sleep because it takes the IRQ
descriptor request mutex, so it does not belong inside the close-time
spinlocked section.
This was found by our static analysis tool and then confirmed by manual
review of the in-tree au1000_close() .ndo_stop path. The reviewed path
keeps aup->lock held across the MAC reset, queue stop and
free_irq(dev->irq, dev).
A directed runtime validation kept that ndo_stop carrier and the same
free_irq(dev->irq, dev) operation under the driver lock. Lockdep reported
"BUG: sleeping function called from invalid context" and "Invalid wait
context" while free_irq() was taking desc->request_mutex, with
au1000_close() and free_irq() on the stack.
Drop aup->lock before freeing the IRQ. The protected close-time work still
stops the device and queue before IRQ teardown, but the sleepable IRQ core
path now runs outside the spinlocked section.
Signed-off-by: Runyu Xiao <runyu.xiao@seu.edu.cn>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260619151816.1144289-1-runyu.xiao@seu.edu.cn
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
pdsc_teardown() frees DMA buffers but does not disable bus mastering,
leaving the device able to perform DMA after the buffers are freed.
This can lead to use-after-free if the device writes to freed memory.
Add pci_clear_master() to pdsc_teardown() to disable bus mastering
before freeing resources, ensuring all DMA is quiesced.
Add pci_set_master() to pdsc_setup() to re-enable bus mastering,
which is needed for the firmware recovery path since pdsc_teardown()
now disables it.
Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Link: https://patch.msgid.link/20260604213637.3844317-1-nikhil.rao@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
napi_complete_done may call gro_flush_normal (though not currently, as GRO
is unsupported at the moment), which may result in packet TX. This will
eventually result in calling pcnet32_start_xmit - resulting in a deadlock
while trying to re-acquire the already locked spin lock.
It is safe to split the spinlock block into two, because the hardware
registers are still protected from concurrent access, and the two blocks
perform unrelated operations that don't need to happen atomically.
Fixes: 5b2ec6f2be ("pcnet32: use napi_complete_done()")
Reviewed-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: Oscar Maes <oscmaes92@gmail.com>
Reviewed-by: Alexander Lobakin <aleksander.lobakin@intel.com>
Link: https://patch.msgid.link/20260528140320.5556-1-oscmaes92@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Using named initializers is more explicit and thus easier to parse for a
human.
While touching these arrays, drop explicit zeros from the list terminator.
This change doesn't introduce changes to the compiled zorro_device_id
arrays.
Signed-off-by: Uwe Kleine-König (The Capable Hub) <u.kleine-koenig@baylibre.com>
Reviewed-by: Geert Uytterhoeven <geert@linux-m68k.org>
Link: https://patch.msgid.link/2243204c0f57f79750de00c072914354d4f65707.1779803053.git.u.kleine-koenig@baylibre.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
union devlink_param_value grows substantially once U64 array
parameters are added to devlink (from 32 bytes to over 264 bytes).
devlink_nl_param_value_fill_one() and devlink_nl_param_value_put()
copy the union by value in several places. Passing two instances as
value arguments alone consumes over 528 bytes of stack; combined with
deeper call chains the parameter stack can approach 800 bytes and trip
CONFIG_FRAME_WARN more easily.
Switch internal helpers and exported driver APIs to pass pointers to
union devlink_param_value rather than passing the union by value.
Reviewed-by: Petr Machata <petrm@nvidia.com> # for mlxsw
Acked-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>
Reviewed-by: Arthur Kiyanovski <akiyano@amazon.com> #for ena
Signed-off-by: Ratheesh Kannoth <rkannoth@marvell.com>
Link: https://patch.msgid.link/20260521095303.2395584-4-rkannoth@marvell.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The driver passes fw_version directly to devlink_info_version_stored_put()
without ensuring null-termination. While current firmware null-terminates
these strings, the driver should not rely on this behavior. Add explicit
null-termination to prevent potential issues if firmware behavior changes.
Fixes: 45d76f4929 ("pds_core: set up device and adminq")
Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Link: https://patch.msgid.link/20260520205842.1486718-1-nikhil.rao@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
debugfs_lookup() returns a dentry with an elevated reference count that
must be released with dput(). The current code discards the returned
dentry without calling dput(), causing a reference leak on every
firmware reset recovery.
Additionally, when CONFIG_DEBUG_FS is disabled, debugfs_lookup()
returns ERR_PTR(-ENODEV), not NULL. The current check passes for error
pointers and would call dput() on an invalid pointer, causing a crash.
Fixes: bc90fbe0c3 ("pds_core: Rework teardown/setup flow to be more common")
Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Link: https://patch.msgid.link/20260515212907.998028-3-nikhil.rao@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Fix two cases where pdsc_devcmd_wait() returns stale success from
the completion register instead of an error:
1. FW crash: If firmware stops running, the wait loop breaks early with
running=false. The condition "if ((!done || timeout) && running)" is
false, so error handling is bypassed and stale status is returned.
Check !running first and return -ENXIO.
2. Timeout: If a command times out, err is set to -ETIMEDOUT but then
overwritten by pdsc_err_to_errno(status) which reads stale status.
Return -ETIMEDOUT immediately after cleaning up.
Both errors now propagate to pdsc_devcmd_locked() which queues
health_work for recovery.
Fixes: 45d76f4929 ("pds_core: set up device and adminq")
Signed-off-by: Nikhil P. Rao <nikhil.rao@amd.com>
Link: https://patch.msgid.link/20260515212907.998028-1-nikhil.rao@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
XGBE_PTP_ACT_CLK_FREQ and XGBE_V2_PTP_ACT_CLK_FREQ were 10x too
large (500MHz/1GHz instead of 50MHz/100MHz), causing the computed
addend to overflow the 32-bit tstamp_addend. In the general case
this would result in the clock advancing at the wrong rate. For v2
(PCI), ptpclk_rate is hardcoded to 125MHz, so the addend formula
(ACT_CLK_FREQ << 32) / ptpclk_rate yields exactly 8 * 2^32, and
when stored to the 32-bit tstamp_addend the value is zero. With
addend = 0 the hardware accumulator never overflows and the PTP
clock is fully stopped. For v1 (platform), ptpclk_rate is read from
ACPI/DT so the exact overflow behavior depends on the
firmware-reported frequency.
Define the constants as NSEC_PER_SEC / SSINC so the relationship is
explicit and cannot drift out of sync.
Fixes: fbd47be098 ("amd-xgbe: add hardware PTP timestamping support")
Tested-by: Gregory Fuchedgi <gfuchedgi@gmail.com>
Signed-off-by: Gregory Fuchedgi <gfuchedgi@gmail.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Link: https://patch.msgid.link/20260429-fix-xgbe-ptp-addend-v1-1-fca5b0ca5e62@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The address latched in the I/O ASIC LANCE DMA Pointer Register uses the
TURBOchannel bus address encoding and therefore bits 33:29 of location
referred occupy bits 4:0, bits 28:2 are left-shifted by 3, and bits 1:0
are hardwired to zero. In reality no TURBOchannel system exceeds 1GiB
of RAM though, so the address reported will always fit in 8 hex digits.
Signed-off-by: Maciej W. Rozycki <macro@orcam.me.uk>
Link: https://patch.msgid.link/alpine.DEB.2.21.2603291839220.60268@angie.orcam.me.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add intelligent TX descriptor cleanup mechanism to reclaim abandoned
descriptors when the physical link goes down.
When the link goes down while TX packets are in-flight, the hardware
stops processing descriptors with the OWN bit still set. The current
driver waits indefinitely for these descriptors to complete, which
never happens. This causes:
- TX ring exhaustion (no descriptors available for new packets)
- Memory leaks (skbs never freed)
- DMA mapping leaks (mappings never unmapped)
- Network stack backpressure buildup
Add force-cleanup mechanism in xgbe_tx_poll() that detects link-down
state and reclaims abandoned descriptors. The helper functions and DMA
optimizations support efficient TX shutdown:
- xgbe_wait_for_dma_tx_complete(): Wait for DMA completion with
link-down optimization
- Restructure xgbe_disable_tx() for proper shutdown sequence
Implementation:
1. Check link state at the start of tx_poll
2. If link is down, set force_cleanup flag
3. For descriptors that hardware hasn't completed (!tx_complete):
- If force_cleanup: treat as completed and reclaim resources
- If link up: break and wait for hardware (normal behavior)
The cleanup process:
- Frees skbs that will never be transmitted
- Unmaps DMA mappings
- Resets descriptors for reuse
- Does NOT count as successful transmission (correct statistics)
Benefits:
- Prevents TX ring starvation
- Eliminates memory and DMA mapping leaks
- Enables fast link recovery when link comes back up
- Critical for link aggregation failover scenarios
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
Link: https://patch.msgid.link/20260319163251.1808611-4-Raju.Rangoju@amd.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Optimize the TX shutdown sequence when link goes down by skipping
futile hardware wait operations and immediately stopping TX queues.
Current behavior creates delays and resource issues during link-down:
1. xgbe_txq_prepare_tx_stop() waits up to XGBE_DMA_STOP_TIMEOUT for
TX queues to drain, but when link is down, hardware will never
complete the pending descriptors. This causes unnecessary delays
during interface shutdown.
2. TX queues remain active after link-down, allowing the network stack
to continue queuing packets that cannot be transmitted. This leads
to resource buildup and complicates recovery.
This patch adds two optimizations:
Optimization 1: Skip TX queue drain when link is down
In xgbe_txq_prepare_tx_stop(), detect link-down state and return
immediately instead of waiting for hardware. Abandoned descriptors
will be cleaned up by the force-cleanup mechanism (next patch).
Optimization 2: Immediate TX queue stop on link-down
In xgbe_phy_adjust_link(), call netif_tx_stop_all_queues() as soon
as link-down is detected. Also wake TX queues on link-up to resume
transmission.
Benefits:
- Faster interface shutdown (no pointless timeout waits)
- Prevents packet queue buildup in network stack
- Cleaner state management during link transitions
- Enables orderly descriptor cleanup by NAPI poll
Note: We do not call netdev_tx_reset_queue() on link-down because
NAPI poll may still be running, which would trigger BQL assertions.
BQL state is cleaned up naturally during descriptor reclamation.
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
Link: https://patch.msgid.link/20260319163251.1808611-3-Raju.Rangoju@amd.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Implement adaptive link status polling to enable fast link-down detection
while conserving CPU resources during link-down periods.
Currently, the driver polls link status at a fixed 1-second interval
regardless of link state. This creates a trade-off:
- Slow polling (1s): Misses rapid link state changes, causing delays
- Fast polling: Wastes CPU when link is stable or down
This enhancement introduces state-aware polling:
When carrier is UP:
Poll every 100ms to enable rapid link-down detection. This provides
~100-200ms response time to link failures, minimizing packet loss and
enabling fast failover in link aggregation configurations.
When carrier is DOWN:
Poll every 1s to conserve CPU resources. Link-up detection is less
time-critical since no traffic is flowing.
Performance impact:
- Link-down detection: 1000ms → 100-200ms (10x improvement)
- CPU overhead when link up: 0.1% → 1% (acceptable for active links)
- CPU overhead when link down: unchanged at 0.1%
This is particularly valuable for:
- Link aggregation deployments requiring sub-second failover
- Environments with flaky links or cable issues
- Applications sensitive to connection recovery time
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
Link: https://patch.msgid.link/20260319163251.1808611-2-Raju.Rangoju@amd.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
device_get_mac_addr is basically device_property_read_u8_array with an
is_valid_ether_addr call. Allows just checking for ret.
Remove XGBE_MAC_ADDR_PROPERTY. device_get_mac_addr supports more
properties than just "mac-address".
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Reviewed-by: Sai Krishna <saikrishnag@marvell.com>
Link: https://patch.msgid.link/20260310194647.3794-1-rosenp@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The current suspend/resume implementation does not correctly handle PCI
device power state transitions, which prevents AMD platforms from
reaching the deepest suspend state (S0i3) when the amd-xgbe driver is
enabled.
In particular, the amd_pmc driver reports:
"Last suspend didn't reach deepest state"
when this device is present.
Implement proper PCI power management operations following the standard
PCI PM model so that the device can be cleanly powered down and resumed.
Suspend path:
- Power down the network interface
- Put the PHY into low-power mode
- Disable bus mastering to prevent DMA activity
- Save PCI configuration space
- Disable the PCI device
- Disable wake from D3 (S0i3 does not require Wake-on-LAN)
- Set the device to D3hot
Resume path:
- Restore the PCI power state to D0
- Restore PCI configuration space
- Enable the PCI device
- Re-enable bus mastering
- Re-enable device interrupts
- Clear the PHY low-power mode
- Power up the network interface
This allows systems using amd-xgbe to reach the deepest suspend state
when entering modern standby (S0i3).
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
Link: https://patch.msgid.link/20260308092851.1510214-3-Raju.Rangoju@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
The caller parameter in xgbe_powerdown() and xgbe_powerup() was intended
to differentiate between driver and ioctl contexts, but the only
remaining usage is from the driver suspend/resume path.
Simplify this by:
- Removing the unused XGMAC_DRIVER_CONTEXT and XGMAC_IOCTL_CONTEXT
macros
- Dropping the now-unused caller parameter
- Reordering operations in xgbe_powerdown() to disable NAPI before
stopping TX/RX, matching the order used in xgbe_stop()
This makes the powerdown/powerup paths easier to follow and keeps the
ordering consistent with the rest of the driver.
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
Link: https://patch.msgid.link/20260308092851.1510214-2-Raju.Rangoju@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
commit f93505f357 ("amd-xgbe: let the MAC manage PHY PM") moved
xgbe_phy_reset() from xgbe_open() to xgbe_start(), placing it after
phy_start(). As a result, the PHY settings were being reset after the
PHY had already started.
Reorder the calls so that the PHY settings are reset before
phy_start() is invoked.
Fixes: f93505f357 ("amd-xgbe: let the MAC manage PHY PM")
Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
Link: https://patch.msgid.link/20260306111629.1515676-4-Raju.Rangoju@amd.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
When operating in 10GBASE-KR mode with auto-negotiation disabled and RX
adaptation enabled, CRC errors can occur during the RX adaptation
process. This happens because the driver continues transmitting and
receiving packets while adaptation is in progress.
Fix this by stopping TX/RX immediately when the link goes down and RX
adaptation needs to be re-triggered, and only re-enabling TX/RX after
adaptation completes and the link is confirmed up. Introduce a flag to
track whether TX/RX was disabled for adaptation so it can be restored
correctly.
This prevents packets from being transmitted or received during the RX
adaptation window and avoids CRC errors from corrupted frames.
The flag tracking the data path state is synchronized with hardware
state in xgbe_start() to prevent stale state after device restarts.
This ensures that after a restart cycle (where xgbe_stop disables
TX/RX and xgbe_start re-enables them), the flag correctly reflects
that the data path is active.
Fixes: 4f3b20bfbb ("amd-xgbe: add support for rx-adaptation")
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
Link: https://patch.msgid.link/20260306111629.1515676-3-Raju.Rangoju@amd.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The link status bit is latched low to allow detection of momentary
link drops. If the status indicates that the link is already down,
read it again to obtain the current state.
Fixes: 4f3b20bfbb ("amd-xgbe: add support for rx-adaptation")
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
Link: https://patch.msgid.link/20260306111629.1515676-2-Raju.Rangoju@amd.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Cross-merge networking fixes after downstream PR (net-7.0-rc3).
No conflicts.
Adjacent changes:
net/netfilter/nft_set_rbtree.c
fb7fb40163 ("netfilter: nf_tables: clone set on flush only")
3aea466a43 ("netfilter: nft_set_rbtree: don't disable bh when acquiring tree lock")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Add hardware support for the AMD P100a platform featuring the ethernet
controller PCI device ID 0x1122.
Platform-specific changes include:
1. PCI device ID and register configuration:
- Add XGBE_P100a_PCI_DEVICE_ID (0x1122) for recognition
- Configure platform-specific XPCS window registers
- Disable CDR workaround and RRC for this platform
2. XPCS window offset calculation fix:
The P100a platform uses a different memory mapping scheme for XPCS
register access. The offset calculation differs between platforms:
- Older platforms (YC): offset = base + (addr & mask)
The address is masked first, then added to the window base.
- P100a: offset = (base + addr) & mask
The full address is added to base first, then masked.
This is critical because using the wrong calculation causes register
reads/writes to access incorrect addresses, leading to incorrect
behaviour.
3. 2.5G speed mode handling:
P100a uses XGMII mode (ss=0x06) for 2.5G instead of GMII mode
(ss=0x02) used by older platforms. The MAC version check determines
which mode to use.
4. Port speed bits extended:
Extend XP_PROP_0_PORT_SPEEDS from 5 bits to 6 bits to support the
additional 5G speed capability.
5. Rx adaptation disabled:
Rx adaptation is disabled for P100a (MAC version 0x33) as this
feature requires further development for this platform.
6. Rate change command for 2.5G:
Use XGBE_MB_SUBCMD_2_5G_KX subcommand for 2.5G mode on P100a
instead of XGBE_MB_SUBCMD_NONE used on older platforms.
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
Link: https://patch.msgid.link/20260302044634.1388661-2-Raju.Rangoju@amd.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
The xgbe_powerdown() and xgbe_powerup() functions use spinlocks
(spin_lock_irqsave) while calling functions that may sleep:
- napi_disable() can sleep waiting for NAPI polling to complete
- flush_workqueue() can sleep waiting for pending work items
This causes a "BUG: scheduling while atomic" error during suspend/resume
cycles on systems using the AMD XGBE Ethernet controller.
The spinlock protection in these functions is unnecessary as these
functions are called from suspend/resume paths which are already serialized
by the PM core
Fix this by removing the spinlock. Since only code that takes this lock
is xgbe_powerdown() and xgbe_powerup(), remove it completely.
Fixes: c5aa9e3b81 ("amd-xgbe: Initial AMD 10GbE platform driver")
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
Link: https://patch.msgid.link/20260302042124.1386445-1-Raju.Rangoju@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Extend the MAC_TCR_SS (Speed Select) register field width from 2 bits
to 3 bits to properly support all speed settings.
The MAC_TCR register's SS field encoding requires 3 bits to represent
all supported speeds:
- 0x00: 10Gbps (XGMII)
- 0x02: 2.5Gbps (GMII) / 100Mbps
- 0x03: 1Gbps / 10Mbps
- 0x06: 2.5Gbps (XGMII) - P100a only
With only 2 bits, values 0x04-0x07 cannot be represented, which breaks
2.5G XGMII mode on newer platforms and causes incorrect speed select
values to be programmed.
Fixes: 07445f3c7c ("amd-xgbe: Add support for 10 Mbps speed")
Co-developed-by: Guruvendra Punugupati <Guruvendra.Punugupati@amd.com>
Signed-off-by: Guruvendra Punugupati <Guruvendra.Punugupati@amd.com>
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
Link: https://patch.msgid.link/20260226170753.250312-1-Raju.Rangoju@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
This converts some of the visually simpler cases that have been split
over multiple lines. I only did the ones that are easy to verify the
resulting diff by having just that final GFP_KERNEL argument on the next
line.
Somebody should probably do a proper coccinelle script for this, but for
me the trivial script actually resulted in an assertion failure in the
middle of the script. I probably had made it a bit _too_ trivial.
So after fighting that far a while I decided to just do some of the
syntactically simpler cases with variations of the previous 'sed'
scripts.
The more syntactically complex multi-line cases would mostly really want
whitespace cleanup anyway.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This was done entirely with mindless brute force, using
git grep -l '\<k[vmz]*alloc_objs*(.*, GFP_KERNEL)' |
xargs sed -i 's/\(alloc_objs*(.*\), GFP_KERNEL)/\1)/'
to convert the new alloc_obj() users that had a simple GFP_KERNEL
argument to just drop that argument.
Note that due to the extreme simplicity of the scripting, any slightly
more complex cases spread over multiple lines would not be triggered:
they definitely exist, but this covers the vast bulk of the cases, and
the resulting diff is also then easier to check automatically.
For the same reason the 'flex' versions will be done as a separate
conversion.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This is the result of running the Coccinelle script from
scripts/coccinelle/api/kmalloc_objs.cocci. The script is designed to
avoid scalar types (which need careful case-by-case checking), and
instead replace kmalloc-family calls that allocate struct or union
object instances:
Single allocations: kmalloc(sizeof(TYPE), ...)
are replaced with: kmalloc_obj(TYPE, ...)
Array allocations: kmalloc_array(COUNT, sizeof(TYPE), ...)
are replaced with: kmalloc_objs(TYPE, COUNT, ...)
Flex array allocations: kmalloc(struct_size(PTR, FAM, COUNT), ...)
are replaced with: kmalloc_flex(*PTR, FAM, COUNT, ...)
(where TYPE may also be *VAR)
The resulting allocations no longer return "void *", instead returning
"TYPE *".
Signed-off-by: Kees Cook <kees@kernel.org>
Add the support to read the rx alignment errors and update
them in the standard rtnl_link_stats64 structure.
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Link: https://patch.msgid.link/20260129111520.1567097-1-Raju.Rangoju@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Passing IRQF_ONESHOT ensures that the interrupt source is masked until
the secondary (threaded) handler is done. If only a primary handler is
used then the flag makes no sense because the interrupt can not fire
(again) while its handler is running.
The flag also disallows force-threading of the primary handler and the
irq-core will warn about this as of commit aef30c8d56 ("genirq: Warn
about using IRQF_ONESHOT without a threaded handler").
The IRQF_ONESHOT flag was added in commit 0fabe1021f ("MIPS:
DECstation I/O ASIC DMA interrupt classes"). It moved
clear_ioasic_dma_irq() from the driver into the irq-chip.
For EOI interrupts the clear_ioasic_dma_irq() callback is now invoked as
->irq_eoi() which is invoked after the IRQ was handled while the
interrupt is masked due to IRQF_ONESHOT. Without IRQF_ONESHOT it would
be invoked while interrupt is unmasked (but interrupts are disabled).
If it is *required* to invoke EOI-ack while the interrupt is masked (and
not a misunderstanding) due to irq-chip cascading/ hierarchical reasons
then using handle_fasteoi_mask_irq() as flow-handler would be the right
way to do so.
Remove IRQF_ONESHOT to irqflags.
Cc: Andrew Lunn <andrew+netdev@lunn.ch>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Acked-by: Maciej W. Rozycki <macro@orcam.me.uk>
Link: https://patch.msgid.link/20260127135334.qUEaYP9G@linutronix.de
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Use the newly introduced .get_rx_ring_count ethtool ops callback instead
of handling ETHTOOL_GRXRINGS directly in .get_rxnfc().
Since ETHTOOL_GRXRINGS was the only command handled by xgbe_get_rxnfc(),
remove the function entirely.
Signed-off-by: Breno Leitao <leitao@debian.org>
Link: https://patch.msgid.link/20260115-grxring_big_v2-v1-6-b3e1b58bced5@debian.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
On the receive path, packet can be damaged because of buffer
overflow in Rx FIFO. Avoid misleading per-packet error log when
packet->errors is set, this can flood the log. Instead, rely on the
standard rtnl_link_stats64 stats.
Fixes: c5aa9e3b81 ("amd-xgbe: Initial AMD 10GbE platform driver")
Signed-off-by: Raju Rangoju <Raju.Rangoju@amd.com>
Link: https://patch.msgid.link/20260114163037.2062606-1-Raju.Rangoju@amd.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
