Re: [PATCH net v4 2/4] net: ethernet: oa_tc6: No return on rx buffer overflow error

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 4 potential issue(s) to consider:

New issues:
- [High] The newly added check for `-EAGAIN` in the threaded IRQ loop is dead code for its intended purpose (RX overflow recovery) and introduces a dangerous tight busy-loop if the underlying SPI driver fails.

Pre-existing issues:
- [Critical] Missing tailroom check before calling `skb_put()` on the receive path allows a malformed or malicious MAC-PHY stream to cause an `skb_over_panic()`.
- [Critical] Out-of-bounds memory access due to unvalidated hardware inputs for TX credits and RX chunks.
- [High] The device IRQ is not disabled before freeing SKB pointers in `oa_tc6_exit()`, leading to a race condition and use-after-free.
--

commit bd14c34a7b1c3dad290ba34e28faf15014377d8e
Author: Selvamani Rajagopal <Selvamani.Rajagopal@onsemi.com>

net: ethernet: oa_tc6: No return on rx buffer overflow error

This patch modifies the threaded IRQ handler to retry the SPI transfer if
oa_tc6_try_spi_transfer() returns -EAGAIN. This is intended to gracefully
recover from RX buffer overflow errors by discarding bad data chunks.

> diff --git a/drivers/net/ethernet/oa_tc6.c b/drivers/net/ethernet/oa_tc6.c
> --- a/drivers/net/ethernet/oa_tc6.c
> +++ b/drivers/net/ethernet/oa_tc6.c
> @@ -1111,6 +1111,8 @@ static irqreturn_t oa_tc6_macphy_threaded_irq(int irq, void *data)
>  
>  	while (tc6->int_flag || (tc6->waiting_tx_skb && tc6->tx_credits)) {
>  		ret = oa_tc6_try_spi_transfer(tc6);
> +		if (ret == -EAGAIN)
> +			continue;

[Severity: High]
Can oa_tc6_try_spi_transfer() actually return -EAGAIN for an RX overflow
condition here?

Looking at oa_tc6_try_spi_transfer(), it catches the -EAGAIN internally:

static int oa_tc6_try_spi_transfer(struct oa_tc6 *tc6)
{
        ...
        ret = oa_tc6_process_spi_data_rx_buf(tc6, spi_len);
        if (ret) {
                if (ret == -EAGAIN)
                        continue;
        ...
}

Since it loops internally, it seems it won't return -EAGAIN to the IRQ
handler for an RX overflow.

The only way oa_tc6_try_spi_transfer() returns -EAGAIN to the caller
is if the underlying SPI transfer fails with -EAGAIN.

If that happens, wouldn't this new continue statement create a tight infinite
busy-loop in the threaded IRQ handler context without sleeping?

>  		if (ret)
>  			break;
>  	}

[Severity: Critical]
This is a pre-existing issue, but does oa_tc6_update_rx_skb() verify that
the skb has sufficient tailroom before calling skb_put()?

In the receive path, data is copied into the skb:

static void oa_tc6_update_rx_skb(struct oa_tc6 *tc6, u8 *payload, u16 length)
{
        ...
        memcpy(skb_put(tc6->rx_skb, length), payload, length);
}

This is called repeatedly for ongoing frames by oa_tc6_prcs_ongoing_rx_frame()
which appends 64-byte payload chunks to the skb.

If a faulty MAC-PHY or bit-flip streams chunks without setting the end_valid
bit, could the unconditional calls to skb_put() overrun the SKB's tailroom
and trigger a kernel panic via skb_over_panic()?

[Severity: High]
This is a pre-existing issue, but is the device IRQ properly disabled
before oa_tc6_exit() frees the SKB pointers?

oa_tc6_exit() does the following:

static void oa_tc6_exit(struct oa_tc6 *tc6)
{
        ...
        dev_kfree_skb_any(tc6->ongoing_tx_skb);
        dev_kfree_skb_any(tc6->waiting_tx_skb);
        dev_kfree_skb_any(tc6->rx_skb);
        ...
}

Since the driver uses devm_request_threaded_irq(), the IRQ remains active
until the driver unbinds, which happens after oa_tc6_exit() returns.

If an interrupt fires concurrently while oa_tc6_exit() calls
dev_kfree_skb_any(), could the still-active oa_tc6_macphy_threaded_irq()
handler attempt to access these freed pointers?

[Severity: Critical]
This is a pre-existing issue, but do we validate the chunk counts read
from the hardware?

The driver updates the chunk counts from hardware registers:

static void oa_tc6_update_buffer_status_from_register(struct oa_tc6 *tc6,
                                                      u32 value)
{
        tc6->rx_chunks_available = FIELD_GET(BUFFER_STATUS_RX_CHUNKS_AVAILABLE,
                                             value);
        ...
}

These bitfields are 8 bits wide, allowing values up to 255. However, the
SPI data buffers are statically sized to handle a maximum of
OA_TC6_MAX_TX_CHUNKS (48).

If the hardware reports a value larger than 48, could loops like
oa_tc6_add_empty_chunks_to_spi_buf() write past the bounds of
spi_data_tx_buf and cause memory corruption during the SPI transfer?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260609-level-trigger-v4-0-6f389abdd192@onsemi.com?part=2