drivers: ethernet: nxp_enet_qos: reclaim TX resources on link drop - #117399
Open
bperseghetti wants to merge 2 commits into
Open
drivers: ethernet: nxp_enet_qos: reclaim TX resources on link drop#117399bperseghetti wants to merge 2 commits into
bperseghetti wants to merge 2 commits into
Conversation
Every RX descriptor writeback clobbers RDES0, which in read format is the buffer address: the writeback formats alias it with the VLAN tag or the timestamp low word. The success path restores it before handing the descriptor back to the DMA, but the four drop paths (errored frame, packet alloc failure, bad length, fragment alloc failure) re-arm the descriptor with whatever the writeback left there, so the next use of that slot makes the DMA write into flash or reserved address space and receive stops permanently. Restore the reserved buffer's address on every drop path, as the success path does. Signed-off-by: Benjamin Perseghetti <bperseghetti@rudislabs.com>
A frame handed to the ENET QoS transmit DMA when the link goes down is never written back, so the completion interrupt that returns the packet and frees the single transmit slot never arrives. The slot stays taken across the link coming back, every later send on the interface fails with -EBUSY, and the packet and its fragments are gone from their pools for good, so a handful of link transitions exhaust the buffer pool and the interface stops transmitting. Give the driver a claim and release pair so that completion and abandonment race for the packet and exactly one of them returns it, and arm a transmit watchdog, CONFIG_ETH_NXP_ENET_QOS_TX_TIMEOUT_MS, that takes the descriptors back, flushes the transmit queue and reinitializes the ring when the DMA does not finish. Abandon an outstanding frame from the PHY callback on carrier loss, and program speed and duplex with the transmitter and receiver disabled before the carrier is announced, so the MAC configuration write cannot land on a frame the stack has already handed to the DMA. Publish the packet, arm the watchdog and start the DMA as one step under irq_lock so an abort can never observe a half-built submission, and flush the transmit queue with a bounded wait. Signed-off-by: Benjamin Perseghetti <bperseghetti@rudislabs.com>
zephyrbot
requested review from
ClaCodes,
Holt-Sun,
butok,
dbaluta,
iuliana-prodan,
lmajewski,
maass-hamburg,
mmahadevan108,
pdgendt,
tpambor and
zejiang0jason
August 26, 2026 01:51
8 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Two RX/TX reliability fixes for the enet_qos driver.
Restore the RX buffer address on every drop path. The DMA writeback clobbers RDES0, which in read format is the buffer address, so any dropped frame handed the descriptor back to the DMA with a corrupt address and permanently poisoned that ring slot.
Reclaim TX resources when the link drops. A frame handed to the transmit DMA when carrier is lost is never written back, so the completion interrupt that returns the packet and the single transmit slot never arrives, and every later send fails with -EBUSY until reset. A bounded TX watchdog abandons the transmission, the abort and completion paths are serialized under irq_lock so exactly one of them releases the packet, and speed and duplex are programmed only with the transmitter and receiver disabled before carrier is announced.
Validated on NXP MCXN947 hardware across repeated physical link flaps. RX and TX both recover on every cycle and the TX buffer pool returns to its baseline with no leaked buffers.