On 03/31/10 20:38, Gertjan van Wingerde wrote:
> On 03/31/10 19:41, Matthijs Kooijman wrote:
>> Hi Ondrej,
>>
>>> phy0 -> rt2500pci_set_device_state: Error - Device failed to enter state 1 (-16).
>> I've been running with your patches (applied to 2.6.32.7) for a day of normal
>> work, and the steady stream of these errors has stopped. During the entire
>> day, I've seen the error once, perhaps a corner case here or there?
>>
>> Overall, I have the feeling that connectivity is a bit more unstable with
>> these two patches, however. On a few occasions, my SSH connections would hang
>> for a bit. Sometimes pinging would be ok (and sometimes it wouldn't), so I
>> haven't quite figured out if these problems were actually caused by the
>> driver, or some other cause). Once I seemd to solve the problem by turning off
>> powersaving, but that might have been a coincidence. Other times the problems
>> went away by themselves.
>>
>> Is there anything else I can do at the moment I'm seeing connection problems?
>> Make a register dump or check some command?
>>
>> I'll do some more testing, tonight on another (probably more stable)
>> connection.
>
> Could you try the attached patch on top of Ondrej's?
>
> The legacy Ralink drivers do not go to sleep when there are still active entries
> in the TX queues. This patch mimics that behavior by refusing to go to sleep if
> there still are active entries in any of the TX queues.
Sorry, I got confused on which driver we were working on. Find attached the correct
patch, now for the rt2500pci driver instead of the rt2500usb driver.
---
Gertjan
diff --git a/drivers/net/wireless/rt2x00/rt2500pci.c b/drivers/net/wireless/rt2x00/rt2500pci.c
index b0c9502..a0441fa 100644
--- a/drivers/net/wireless/rt2x00/rt2500pci.c
+++ b/drivers/net/wireless/rt2x00/rt2500pci.c
@@ -558,9 +558,17 @@ static void rt2500pci_config_ps(struct rt2x00_dev *rt2x00dev,
enum dev_state state =
(libconf->conf->flags & IEEE80211_CONF_PS) ?
STATE_SLEEP : STATE_AWAKE;
+ struct data_queue *queue;
u32 reg;
if (state == STATE_SLEEP) {
+ /*
+ * Don't go to sleep when the TX queues are not empty.
+ */
+ tx_queue_for_each(rt2x00dev, queue)
+ if (!rt2x00queue_empty(queue))
+ return;
+
rt2x00pci_register_read(rt2x00dev, CSR20, ®);
rt2x00_set_field32(®, CSR20_DELAY_AFTER_TBCN,
(rt2x00dev->beacon_int - 20) * 16);
