Skip to content

Commit

Permalink
sfc: Use correct fields of struct ethtool_coalesce
Browse files Browse the repository at this point in the history
An earlier developer misunderstood the meaning of the 'irq' fields and
the driver did not support the standard fields.  To avoid invalidating
existing user documentation, we report and accept changes through
either the standard or 'irq' fields.  If both are changed at the same
time, we prefer the standard field.

Also explain why we don't currently use the 'max_frames' fields.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
  • Loading branch information
Ben Hutchings authored and David S. Miller committed Sep 16, 2011
1 parent 9e393b3 commit 1322597
Showing 1 changed file with 27 additions and 9 deletions.
36 changes: 27 additions & 9 deletions drivers/net/ethernet/sfc/ethtool.c
Original file line number Diff line number Diff line change
Expand Up @@ -595,6 +595,20 @@ static int efx_ethtool_nway_reset(struct net_device *net_dev)
* automatically changed too, but otherwise we fail if the two values
* are requested to be different.
*
* The hardware does not support a limit on the number of completions
* before an IRQ, so we do not use the max_frames fields. We should
* report and require that max_frames == (usecs != 0), but this would
* invalidate existing user documentation.
*
* The hardware does not have distinct settings for interrupt
* moderation while the previous IRQ is being handled, so we should
* not use the 'irq' fields. However, an earlier developer
* misunderstood the meaning of the 'irq' fields and the driver did
* not support the standard fields. To avoid invalidating existing
* user documentation, we report and accept changes through either the
* standard or 'irq' fields. If both are changed at the same time, we
* prefer the standard field.
*
* We implement adaptive IRQ moderation, but use a different algorithm
* from that assumed in the definition of struct ethtool_coalesce.
* Therefore we do not use any of the adaptive moderation parameters
Expand All @@ -610,7 +624,9 @@ static int efx_ethtool_get_coalesce(struct net_device *net_dev,

efx_get_irq_moderation(efx, &tx_usecs, &rx_usecs, &rx_adaptive);

coalesce->tx_coalesce_usecs = tx_usecs;
coalesce->tx_coalesce_usecs_irq = tx_usecs;
coalesce->rx_coalesce_usecs = rx_usecs;
coalesce->rx_coalesce_usecs_irq = rx_usecs;
coalesce->use_adaptive_rx_coalesce = rx_adaptive;

Expand All @@ -629,22 +645,24 @@ static int efx_ethtool_set_coalesce(struct net_device *net_dev,
if (coalesce->use_adaptive_tx_coalesce)
return -EINVAL;

if (coalesce->rx_coalesce_usecs || coalesce->tx_coalesce_usecs) {
netif_err(efx, drv, efx->net_dev, "invalid coalescing setting. "
"Only rx/tx_coalesce_usecs_irq are supported\n");
return -EINVAL;
}

efx_get_irq_moderation(efx, &tx_usecs, &rx_usecs, &adaptive);

rx_usecs = coalesce->rx_coalesce_usecs_irq;
if (coalesce->rx_coalesce_usecs != rx_usecs)
rx_usecs = coalesce->rx_coalesce_usecs;
else
rx_usecs = coalesce->rx_coalesce_usecs_irq;

adaptive = coalesce->use_adaptive_rx_coalesce;

/* If channels are shared, TX IRQ moderation can be quietly
* overridden unless it is changed from its old value.
*/
rx_may_override_tx = coalesce->tx_coalesce_usecs_irq == tx_usecs;
tx_usecs = coalesce->tx_coalesce_usecs_irq;
rx_may_override_tx = (coalesce->tx_coalesce_usecs == tx_usecs &&
coalesce->tx_coalesce_usecs_irq == tx_usecs);
if (coalesce->tx_coalesce_usecs != tx_usecs)
tx_usecs = coalesce->tx_coalesce_usecs;
else
tx_usecs = coalesce->tx_coalesce_usecs_irq;

rc = efx_init_irq_moderation(efx, tx_usecs, rx_usecs, adaptive,
rx_may_override_tx);
Expand Down

0 comments on commit 1322597

Please sign in to comment.