Restock the RX queue when there are a lot of unused frames so that the
RX ring buffer doesn't overrun, causing a ucode assertion. Backport of
patch "iwlwifi: fix ucode assertion for RX queue overrun".
Signed-off-by: Dan Williams <[email protected]>
-------------------------------
Sebastian, can you test and ensure that this works for you? Also, can
you do some throughput testing to ensure that this patch doesn't cause a
throughput regression? Thanks!
diff --git a/drivers/net/wireless/ipw2200.c b/drivers/net/wireless/ipw2200.c
index 3e6ad7b..a56d9fc 100644
--- a/drivers/net/wireless/ipw2200.c
+++ b/drivers/net/wireless/ipw2200.c
@@ -3365,7 +3365,6 @@ static void ipw_rx_queue_reset(struct ipw_priv *priv,
/* Set us so that we have processed and used all buffers, but have
* not restocked the Rx queue with fresh buffers */
rxq->read = rxq->write = 0;
- rxq->processed = RX_QUEUE_SIZE - 1;
rxq->free_count = 0;
spin_unlock_irqrestore(&rxq->lock, flags);
}
@@ -3607,7 +3606,22 @@ static int ipw_load(struct ipw_priv *priv)
* Driver allocates buffers of this size for Rx
*/
-static inline int ipw_queue_space(const struct clx2_queue *q)
+/**
+ * ipw_rx_queue_space - Return number of free slots available in queue.
+ */
+static int ipw_rx_queue_space(const struct ipw_rx_queue *q)
+{
+ int s = q->read - q->write;
+ if (s <= 0)
+ s += RX_QUEUE_SIZE;
+ /* keep some buffer to not confuse full and empty queue */
+ s -= 2;
+ if (s < 0)
+ s = 0;
+ return s;
+}
+
+static inline int ipw_tx_queue_space(const struct clx2_queue *q)
{
int s = q->last_used - q->first_empty;
if (s <= 0)
@@ -4947,7 +4961,7 @@ static int ipw_queue_tx_reclaim(struct ipw_priv *priv,
priv->tx_packets++;
}
done:
- if ((ipw_queue_space(q) > q->low_mark) &&
+ if ((ipw_tx_queue_space(q) > q->low_mark) &&
(qindex >= 0) &&
(priv->status & STATUS_ASSOCIATED) && netif_running(priv->net_dev))
netif_wake_queue(priv->net_dev);
@@ -4965,7 +4979,7 @@ static int ipw_queue_tx_hcmd(struct ipw_priv *priv, int hcmd, void *buf,
struct clx2_queue *q = &txq->q;
struct tfd_frame *tfd;
- if (ipw_queue_space(q) < (sync ? 1 : 2)) {
+ if (ipw_tx_queue_space(q) < (sync ? 1 : 2)) {
IPW_ERROR("No space for Tx\n");
return -EBUSY;
}
@@ -5070,7 +5084,7 @@ static void ipw_rx_queue_restock(struct ipw_priv *priv)
spin_lock_irqsave(&rxq->lock, flags);
write = rxq->write;
- while ((rxq->write != rxq->processed) && (rxq->free_count)) {
+ while ((ipw_rx_queue_space(rxq) > 0) && (rxq->free_count)) {
element = rxq->rx_free.next;
rxb = list_entry(element, struct ipw_rx_mem_buffer, list);
list_del(element);
@@ -5187,7 +5201,6 @@ static struct ipw_rx_queue *ipw_rx_queue_alloc(struct ipw_priv *priv)
/* Set us so that we have processed and used all buffers, but have
* not restocked the Rx queue with fresh buffers */
rxq->read = rxq->write = 0;
- rxq->processed = RX_QUEUE_SIZE - 1;
rxq->free_count = 0;
return rxq;
@@ -8223,13 +8236,17 @@ static void ipw_rx(struct ipw_priv *priv)
struct ieee80211_hdr_4addr *header;
u32 r, w, i;
u8 network_packet;
+ u8 fill_rx = 0;
DECLARE_MAC_BUF(mac);
DECLARE_MAC_BUF(mac2);
DECLARE_MAC_BUF(mac3);
r = ipw_read32(priv, IPW_RX_READ_INDEX);
w = ipw_read32(priv, IPW_RX_WRITE_INDEX);
- i = (priv->rxq->processed + 1) % RX_QUEUE_SIZE;
+ i = priv->rxq->read;
+
+ if (ipw_rx_queue_space (priv->rxq) > (RX_QUEUE_SIZE / 2))
+ fill_rx = 1;
while (i != r) {
rxb = priv->rxq->queue[i];
@@ -8404,11 +8421,17 @@ static void ipw_rx(struct ipw_priv *priv)
list_add_tail(&rxb->list, &priv->rxq->rx_used);
i = (i + 1) % RX_QUEUE_SIZE;
+
+ /* If there are a lot of unsued frames, restock the Rx queue
+ * so the ucode won't assert */
+ if (fill_rx) {
+ priv->rxq->read = i;
+ ipw_rx_queue_replenish(priv);
+ }
}
/* Backtrack one entry */
- priv->rxq->processed = (i ? i : RX_QUEUE_SIZE) - 1;
-
+ priv->rxq->read = i;
ipw_rx_queue_restock(priv);
}
@@ -10336,7 +10359,7 @@ static int ipw_tx_skb(struct ipw_priv *priv, struct ieee80211_txb *txb,
q->first_empty = ipw_queue_inc_wrap(q->first_empty, q->n_bd);
ipw_write32(priv, q->reg_w, q->first_empty);
- if (ipw_queue_space(q) < q->high_mark)
+ if (ipw_tx_queue_space(q) < q->high_mark)
netif_stop_queue(priv->net_dev);
return NETDEV_TX_OK;
@@ -10357,7 +10380,7 @@ static int ipw_net_is_queue_full(struct net_device *dev, int pri)
struct clx2_tx_queue *txq = &priv->txq[0];
#endif /* CONFIG_IPW2200_QOS */
- if (ipw_queue_space(&txq->q) < txq->q.high_mark)
+ if (ipw_tx_queue_space(&txq->q) < txq->q.high_mark)
return 1;
return 0;
Sebastian Siewior wrote:
> * Dan Williams | 2008-02-11 09:27:50 [-0500]:
>
>
>> Restock the RX queue when there are a lot of unused frames so that the
>> RX ring buffer doesn't overrun, causing a ucode assertion. Backport of
>> patch "iwlwifi: fix ucode assertion for RX queue overrun".
>>
>> Signed-off-by: Dan Williams <[email protected]>
>>
> Tested-by: Sebastian Siewior <[email protected]>
>
>
>> Sebastian, can you test and ensure that this works for you? Also, can
>>
> Yep, it does.
>
>
>> you do some throughput testing to ensure that this patch doesn't cause a
>> throughput regression? Thanks!
>>
> 00:49:41 (666.64 KB/s) - `/dev/null' saved [122563640/122563640]
>
> this was a wget from the inet and it hits almost my cap. Currently I don't
> have another local machine where I can test a "local" transfer.
>
> Is this okey for you?
>
> Sebastian
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by: Microsoft
> Defy all challenges. Microsoft(R) Visual Studio 2008.
> http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
> _______________________________________________
> ipw2100-devel mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/ipw2100-devel
>
>
The patch seem stable. Throughput are ok.
sftp> get 1001_20071125080100.mpg
Fetching /myth/tv/1001_20071125080100.mpg to 1001_20071125080100.mpg
/myth/tv/1001_20071125080100.mpg 100% 1120MB 2.6MB/s
07:16
sftp>
/Christian
On , Christian Fredriksson wrote:
> assertion for RX queue overrun
>
> Sebastian Siewior wrote:
>> * Dan Williams | 2008-02-11 09:27:50 [-0500]:
>>
>>
>>> Restock the RX queue when there are a lot of unused frames so that
>>> the RX ring buffer doesn't overrun, causing a ucode assertion.
>>> Backport of patch "iwlwifi: fix ucode assertion for RX queue
>>> overrun".
>>>
>>> Signed-off-by: Dan Williams <[email protected]>
>>>
>> Tested-by: Sebastian Siewior <[email protected]>
>>
>>
>>> Sebastian, can you test and ensure that this works for you? Also,
>>> can
>>>
>> Yep, it does.
>>
>>
>>> you do some throughput testing to ensure that this patch doesn't
>>> cause a throughput regression? Thanks!
>>>
>> 00:49:41 (666.64 KB/s) - `/dev/null' saved [122563640/122563640]
>>
>> this was a wget from the inet and it hits almost my cap. Currently I
>> don't have another local machine where I can test a "local" transfer.
>>
>> Is this okey for you?
>>
>> Sebastian
>>
>>
>>
> The patch seem stable. Throughput are ok.
>
> sftp> get 1001_20071125080100.mpg
> Fetching /myth/tv/1001_20071125080100.mpg to 1001_20071125080100.mpg
> /myth/tv/1001_20071125080100.mpg 100% 1120MB 2.6MB/s
> 07:16 sftp>
>
> /Christian
Sounds great!
Dan, could you please submit a final version - I'll ack it.
Thank you very much
Reinette
* Dan Williams | 2008-02-11 09:27:50 [-0500]:
>Restock the RX queue when there are a lot of unused frames so that the
>RX ring buffer doesn't overrun, causing a ucode assertion. Backport of
>patch "iwlwifi: fix ucode assertion for RX queue overrun".
>
>Signed-off-by: Dan Williams <[email protected]>
Tested-by: Sebastian Siewior <[email protected]>
>Sebastian, can you test and ensure that this works for you? Also, can
Yep, it does.
>you do some throughput testing to ensure that this patch doesn't cause a
>throughput regression? Thanks!
00:49:41 (666.64 KB/s) - `/dev/null' saved [122563640/122563640]
this was a wget from the inet and it hits almost my cap. Currently I don't
have another local machine where I can test a "local" transfer.
Is this okey for you?
Sebastian