On 25/02/2010 18:22, Gertjan van Wingerde wrote:
On 02/25/10 05:19, Alban Browaeys wrote:
This is an implementation that support WCID being the key_index coming
from benoit without the change in the meaning of the tx fallback flag.
Ivo: in previous patch I forgot about your comment months ago against the
fallback meaning change. This version thus avoid this change.
Signed-off-by: Benoit Papillault<benoit.papillault@xxxxxxx>
Signed-off-by: Alban Browaeys<prahal@xxxxxxxxx>
---
Your commit message does not seem to explain what this change is for, and
why this change is being made (and the original patch submission doesn't
explain this as well)
What problem with the original code is being fixed with your patch?
Please include a commit message that explains these things.
Doing? Sending it asap. The previous version was mine and relied on WCID
behing the entry idx.
But it got broken by another commit that restore WCID behing the key idx.
Thus
index = rt2x00_get_field32(reg, TX_STA_FIFO_WCID) - 1;
cannot work. As it will grab the key idx instead of the entry idx.
This implementation is Benoit Papillault one that support hw crypt and WCID behing key idx.
drivers/net/wireless/rt2x00/rt2800pci.c | 86
+++++++++++++------------------
1 files changed, 36 insertions(+), 50 deletions(-)
diff --git a/drivers/net/wireless/rt2x00/rt2800pci.c
b/drivers/net/wireless/rt2x00/rt2800pci.c
index 7899789..46b06af 100644
--- a/drivers/net/wireless/rt2x00/rt2800pci.c
+++ b/drivers/net/wireless/rt2x00/rt2800pci.c
@@ -920,76 +920,61 @@ static void rt2800pci_txdone(struct rt2x00_dev
*rt2x00dev)
{
struct data_queue *queue;
struct queue_entry *entry;
- struct queue_entry *entry_done;
- struct queue_entry_priv_pci *entry_priv;
+ __le32 *txwi;
struct txdone_entry_desc txdesc;
u32 word;
u32 reg;
- u32 old_reg;
- unsigned int type;
- unsigned int index;
- u16 mcs, real_mcs;
-
+ int i;
+ int wcid, ack, pid, tx_wcid, tx_ack, tx_pid;
+ u16 mcs, tx_mcs;
+
/*
- * During each loop we will compare the freshly read
- * TX_STA_FIFO register value with the value read from
- * the previous loop. If the 2 values are equal then
- * we should stop processing because the chance it
- * quite big that the device has been unplugged and
- * we risk going into an endless loop.
+ * To avoid an endlees loop, we only read the TX_STA_FIFO register up
+ * to 256 times (this is enought to get all values from the FIFO). In
+ * normal situation, the loop is terminated when we reach a value with
+ * TX_STA_FIFO_VALID bit is 0.
*/
- old_reg = 0;
-
- while (1) {
+
+ for (i=0; i<256; i++) {
Why this magical change to only try 256 times?
see comment above "to 256 times (this is enought to get all values from
the FIFO)".
I seems to make more sense that the previous infinite loop that relied
on the comparison
of old_reg and reg to detect if we were to exit.
rt2800_register_read(rt2x00dev, TX_STA_FIFO,®);
if (!rt2x00_get_field32(reg, TX_STA_FIFO_VALID))
break;
- if (old_reg == reg)
- break;
- old_reg = reg;
+ wcid = rt2x00_get_field32(reg, TX_STA_FIFO_WCID);
+ ack = rt2x00_get_field32(reg, TX_STA_FIFO_TX_ACK_REQUIRED);
+ pid = rt2x00_get_field32(reg, TX_STA_FIFO_PID_TYPE);
/*
* Skip this entry when it contains an invalid
* queue identication number.
*/
- type = rt2x00_get_field32(reg, TX_STA_FIFO_PID_TYPE) - 1;
- if (type>= QID_RX)
+ if (pid< 1)
continue;
- queue = rt2x00queue_get_queue(rt2x00dev, type);
+ queue = rt2x00queue_get_queue(rt2x00dev, pid - 1);
if (unlikely(!queue))
continue;
/*
- * Skip this entry when it contains an invalid
- * index number.
+ * Inside each queue, we process each entry in a chronological
+ * order. We first check that the queue is not empty.
*/
- index = rt2x00_get_field32(reg, TX_STA_FIFO_WCID) - 1;
- if (unlikely(index>= queue->limit))
+ if (queue->length == 0)
continue;
+ entry = rt2x00queue_get_entry(queue, Q_INDEX_DONE);
- entry =&queue->entries[index];
- entry_priv = entry->priv_data;
- rt2x00_desc_read((__le32 *)entry->skb->data, 0,&word);
-
- entry_done = rt2x00queue_get_entry(queue, Q_INDEX_DONE);
- while (entry != entry_done) {
- /*
- * Catch up.
- * Just report any entries we missed as failed.
- */
- WARNING(rt2x00dev,
- "TX status report missed for entry %d\n",
- entry_done->entry_idx);
-
- txdesc.flags = 0;
- __set_bit(TXDONE_UNKNOWN,&txdesc.flags);
- txdesc.retry = 0;
-
- rt2x00lib_txdone(entry_done,&txdesc);
- entry_done = rt2x00queue_get_entry(queue, Q_INDEX_DONE);
- }
+ /* Check if we got a match by looking at WCID/ACK/PID
+ * fields */
+ txwi = (__le32 *)(entry->skb->data -
+ rt2x00dev->hw->extra_tx_headroom);
+
+ rt2x00_desc_read(txwi, 1,&word);
+ tx_wcid = rt2x00_get_field32(word, TXWI_W1_WIRELESS_CLI_ID);
+ tx_ack = rt2x00_get_field32(word, TXWI_W1_ACK);
+ tx_pid = rt2x00_get_field32(word, TXWI_W1_PACKETID);
+
+ if ((wcid != tx_wcid) || (ack != tx_ack) || (pid != tx_pid))
+ WARNING(rt2x00dev, "invalid TX_STA_FIFO content\n");
You're patch seems to remove the protection we had to detect transmitted frames
for which we didn't get a TX status confirmation. This had been added in the
past for the other rt2x00 drivers to handle real issues with the rt2x00 devices.
Why is this being removed?
This protection relied on the fact that we could get the current entry
idx (which could only be shipped
in WCID) to compare it to Q_INDEX_DONE and loop over the diff.
This cannot work with WCID behing the key idx as thus we have to rely on
the fact that the current entry
index is the current one (thus the diff is always void and looping over
this void diff is pointless).
Best regards and thanks for the comments.
Alban
--
To unsubscribe from this list: send the line "unsubscribe linux-wireless" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
