@@ -1909,21 +1909,21 @@ static int rtl_set_coalesce(struct net_device *dev, struct ethtool_coalesce *ec)
* - then user does `ethtool -C eth0 rx-usecs 100`
*
* since ethtool sends to kernel whole ethtool_coalesce
- * settings, if we do not handle rx_usecs=!0, rx_frames=1
- * we'll reject it below in `frames % 4 != 0`.
+ * settings, if we want to ignore rx_frames then it has
+ * to be set to 0.
*/
if (p->frames == 1) {
p->frames = 0;
}
- units = p->usecs * 1000 / scale;
- if (p->frames > RTL_COALESCE_FRAME_MAX || p->frames % 4)
- return -EINVAL;
+ units = DIV_ROUND_UP(p->usecs * 1000, scale);
+ if (p->frames > RTL_COALESCE_FRAME_MAX)
+ return -ERANGE;
w <<= RTL_COALESCE_SHIFT;
w |= units;
w <<= RTL_COALESCE_SHIFT;
- w |= p->frames >> 2;
+ w |= DIV_ROUND_UP(p->frames, 4);
}
rtl_lock_work(tp);
The chip supports only frame limits 0, 4, 8, .. 60 internally. Returning EINVAL for all val % 4 != 0 seems to be a little bit too unfriendly to the user. Therefore round up the frame limit to the next supported value. In addition round up the time limit, else a very low limit could be rounded down to 0, and interpreted as "ignore value" by the chip. Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com> --- drivers/net/ethernet/realtek/r8169_main.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-)