[PATCH 3/5] net: ks8851_mll: do not read past the receive buffer
Sascha Hauer
s.hauer at pengutronix.de
Mon May 27 00:29:13 PDT 2024
the hardware may report a packet longer than our receive buffer. Instead
of reading past the receive buffer, discard too long packets.
Signed-off-by: Sascha Hauer <s.hauer at pengutronix.de>
---
drivers/net/ks8851_mll.c | 28 +++++++++++++++++++++++++---
1 file changed, 25 insertions(+), 3 deletions(-)
diff --git a/drivers/net/ks8851_mll.c b/drivers/net/ks8851_mll.c
index 2120657bd9..5e41888b28 100644
--- a/drivers/net/ks8851_mll.c
+++ b/drivers/net/ks8851_mll.c
@@ -420,6 +420,23 @@ static inline void ks_inblk(struct ks_net *ks, u16 *wptr, u32 len)
*wptr++ = (u16)readw(ks->hw_addr);
}
+/**
+ * ks_discardblk - read a block of data from QMU discarding the data
+ * @ks: The chip state
+ * @len: length in byte to read
+ *
+ * This discards a block of data, used when a packet is longer than our receive
+ * buffer. I don't know if it is necessary to discard the data like this, but
+ * it is the easy way out to fix the behaviour with too large packets without
+ * risking regressions.
+ */
+static inline void ks_discardblk(struct ks_net *ks, u32 len)
+{
+ len >>= 1;
+ while (len--)
+ (void)readw(ks->hw_addr);
+}
+
/**
* ks_outblk - write data to QMU.
* @ks: The chip information
@@ -676,9 +693,14 @@ static int ks8851_rx_frame(struct ks_net *ks)
tmp_rxqcr = ks_rdreg16(ks, KS_RXQCR);
ks_wrreg16(ks, KS_RXQCR, tmp_rxqcr | RXQCR_SDA);
- /* read 2 bytes for dummy, 2 for status, 2 for len*/
- ks_inblk(ks, ks->rx_buf, 2 + 2 + 2);
- ks_inblk(ks, ks->rx_buf, ALIGN(RxLen, 4));
+ if (RxLen <= PKTSIZE) {
+ /* read 2 bytes for dummy, 2 for status, 2 for len*/
+ ks_inblk(ks, ks->rx_buf, 2 + 2 + 2);
+ ks_inblk(ks, ks->rx_buf, ALIGN(RxLen, 4));
+ } else {
+ ks_discardblk(ks, 2 + 2 + 2 + ALIGN(RxLen, 4));
+ }
+
ks_wrreg16(ks, KS_RXQCR, tmp_rxqcr);
if (RxStatus & RXFSHR_RXFV) {
--
2.39.2
More information about the barebox
mailing list