Re: [PATCH net-next 1/3] net: macb: Make sure clk_init_data is fully initialized

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] The TX error recovery task modifies TX ring pointers and state without holding `queue->tx_ptr_lock`, racing with `macb_start_xmit`.
--

--- Patch [1]: [PATCH net-next 1/3] net: macb: Make sure clk_init_data is fully initialized ---
commit 72f05b5fbec22828f860396d626f5a6bbaacd336
Author: Geert Uytterhoeven <geert+renesas@glider.be>

net: macb: Make sure clk_init_data is fully initialized

This commit initializes the clk_init_data structure to prevent issues with
uninitialized mutually-exclusive members when stack initialization patterns
are enabled.

> diff --git a/drivers/net/ethernet/cadence/macb_main.c b/drivers/net/ethernet/cadence/macb_main.c
> index 1476bce77f34d..319e43098f57e 100644
> --- a/drivers/net/ethernet/cadence/macb_main.c
> +++ b/drivers/net/ethernet/cadence/macb_main.c

[ ... ]

[Severity: High]
This isn't a bug introduced by this patch, but I noticed a pre-existing data
race between macb_tx_error_task() and macb_start_xmit() in this file.

Does macb_tx_error_task() race with macb_start_xmit() by modifying TX ring
pointers without holding queue->tx_ptr_lock?

In macb_tx_error_task(), the ring pointers are reset without a lock:

    queue->tx_head = 0;
    queue->tx_tail = 0;

Concurrently, macb_start_xmit() acquires the lock and updates tx_head:

    spin_lock_irqsave(&queue->tx_ptr_lock, flags);

Since netif_tx_stop_all_queues() only marks the queue as stopped but does not
wait for an already-executing ndo_start_xmit() to finish, could this race
corrupt the software TX ring pointers? 

This might lead to out-of-bounds DMA or use-after-free when the
desynchronized ring is subsequently processed after a transmission error.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/cover.1787240118.git.geert+renesas@glider.be?part=1