From: Raja Mani <[email protected]>
Below two scenarios are taken care in this patch which helped
to fix the firmware crash during wow suspend/resume.
* TX operation (ctrl tx and data tx) has to be controlled based
on suspend state. i.e, with respect to WOW mode, control packets
are allowed to send from the host until the suspend state goes
ATH6KL_STATE_WOW and the data packets are allowed until WOW
suspend operation starts.
* Similarly, wow resume is NOT allowed if WOW suspend is in progress.
Signed-off-by: Raja Mani <[email protected]>
---
V2 changes:
* Separate wow sub state introduced in V1 is removed and common
states (ATH6KL_STATE_SUSPENDING and ATH6KL_STATE_RESUMING)
are added in enum ath6kl_state.
* netdev queue is stopped while dropping skb in ath6kl_data_tx()
and resumed back in ath6kl_wow_resume().
drivers/net/wireless/ath/ath6kl/cfg80211.c | 29 ++++++++++++++++++++++++++-
drivers/net/wireless/ath/ath6kl/core.h | 2 +
drivers/net/wireless/ath/ath6kl/sdio.c | 4 +++
drivers/net/wireless/ath/ath6kl/txrx.c | 10 +++++++++
4 files changed, 43 insertions(+), 2 deletions(-)
diff --git a/drivers/net/wireless/ath/ath6kl/cfg80211.c b/drivers/net/wireless/ath/ath6kl/cfg80211.c
index d1922d8..dfd3ba4 100644
--- a/drivers/net/wireless/ath/ath6kl/cfg80211.c
+++ b/drivers/net/wireless/ath/ath6kl/cfg80211.c
@@ -1941,6 +1941,8 @@ static int ath6kl_wow_suspend(struct ath6kl *ar, struct cfg80211_wowlan *wow)
if (ret)
return ret;
+ ar->state = ATH6KL_STATE_SUSPENDING;
+
/* Setup own IP addr for ARP agent. */
in_dev = __in_dev_get_rtnl(vif->ndev);
if (!in_dev)
@@ -2019,15 +2021,33 @@ static int ath6kl_wow_resume(struct ath6kl *ar)
if (!vif)
return -EIO;
+ ar->state = ATH6KL_STATE_RESUMING;
+
ret = ath6kl_wmi_set_host_sleep_mode_cmd(ar->wmi, vif->fw_vif_idx,
ATH6KL_HOST_MODE_AWAKE);
- return ret;
+ if (ret) {
+ ath6kl_warn("Failed to configure host sleep mode for "
+ "wow resume: %d\n", ret);
+ ar->state = ATH6KL_STATE_WOW;
+ return ret;
+ }
+
+ ar->state = ATH6KL_STATE_ON;
+
+ /* Wake up netdev queue if it's stopped in ath6kl_data_tx() */
+ if (test_bit(NETQ_STOPPED, &vif->flags)) {
+ clear_bit(NETQ_STOPPED, &vif->flags);
+ netif_wake_queue(vif->ndev);
+ }
+
+ return 0;
}
int ath6kl_cfg80211_suspend(struct ath6kl *ar,
enum ath6kl_cfg_suspend_mode mode,
struct cfg80211_wowlan *wow)
{
+ enum ath6kl_state prev_state;
int ret;
switch (mode) {
@@ -2038,9 +2058,12 @@ int ath6kl_cfg80211_suspend(struct ath6kl *ar,
/* Flush all non control pkts in TX path */
ath6kl_tx_data_cleanup(ar);
+ prev_state = ar->state;
+
ret = ath6kl_wow_suspend(ar, wow);
if (ret) {
ath6kl_err("wow suspend failed: %d\n", ret);
+ ar->state = prev_state;
return ret;
}
ar->state = ATH6KL_STATE_WOW;
@@ -2114,7 +2137,6 @@ int ath6kl_cfg80211_resume(struct ath6kl *ar)
return ret;
}
- ar->state = ATH6KL_STATE_ON;
break;
case ATH6KL_STATE_DEEPSLEEP:
@@ -2188,6 +2210,9 @@ static int __ath6kl_cfg80211_resume(struct wiphy *wiphy)
*/
void ath6kl_check_wow_status(struct ath6kl *ar)
{
+ if (ar->state == ATH6KL_STATE_SUSPENDING)
+ return;
+
if (ar->state == ATH6KL_STATE_WOW)
ath6kl_cfg80211_resume(ar);
}
diff --git a/drivers/net/wireless/ath/ath6kl/core.h b/drivers/net/wireless/ath/ath6kl/core.h
index c4d66e0..04dfdda 100644
--- a/drivers/net/wireless/ath/ath6kl/core.h
+++ b/drivers/net/wireless/ath/ath6kl/core.h
@@ -521,6 +521,8 @@ enum ath6kl_dev_state {
enum ath6kl_state {
ATH6KL_STATE_OFF,
ATH6KL_STATE_ON,
+ ATH6KL_STATE_SUSPENDING,
+ ATH6KL_STATE_RESUMING,
ATH6KL_STATE_DEEPSLEEP,
ATH6KL_STATE_CUTPOWER,
ATH6KL_STATE_WOW,
diff --git a/drivers/net/wireless/ath/ath6kl/sdio.c b/drivers/net/wireless/ath/ath6kl/sdio.c
index 4febee7..402087e 100644
--- a/drivers/net/wireless/ath/ath6kl/sdio.c
+++ b/drivers/net/wireless/ath/ath6kl/sdio.c
@@ -901,8 +901,12 @@ static int ath6kl_sdio_resume(struct ath6kl *ar)
case ATH6KL_STATE_WOW:
break;
+
case ATH6KL_STATE_SCHED_SCAN:
break;
+
+ default:
+ break;
}
ath6kl_cfg80211_resume(ar);
diff --git a/drivers/net/wireless/ath/ath6kl/txrx.c b/drivers/net/wireless/ath/ath6kl/txrx.c
index a3dc694..1336204 100644
--- a/drivers/net/wireless/ath/ath6kl/txrx.c
+++ b/drivers/net/wireless/ath/ath6kl/txrx.c
@@ -284,6 +284,9 @@ int ath6kl_control_tx(void *devt, struct sk_buff *skb,
int status = 0;
struct ath6kl_cookie *cookie = NULL;
+ if (ar->state == ATH6KL_STATE_WOW)
+ return -EACCES;
+
spin_lock_bh(&ar->lock);
ath6kl_dbg(ATH6KL_DBG_WLAN_TX,
@@ -359,6 +362,13 @@ int ath6kl_data_tx(struct sk_buff *skb, struct net_device *dev)
return 0;
}
+ if (WARN_ON_ONCE(ar->state != ATH6KL_STATE_ON)) {
+ set_bit(NETQ_STOPPED, &vif->flags);
+ netif_stop_queue(dev);
+ dev_kfree_skb(skb);
+ return 0;
+ }
+
if (!test_bit(WMI_READY, &ar->flag))
goto fail_tx;
--
1.7.1
On 02/09/2012 11:31 AM, [email protected] wrote:
> From: Raja Mani <[email protected]>
>
> Below two scenarios are taken care in this patch which helped
> to fix the firmware crash during wow suspend/resume.
>
> * TX operation (ctrl tx and data tx) has to be controlled based
> on suspend state. i.e, with respect to WOW mode, control packets
> are allowed to send from the host until the suspend state goes
> ATH6KL_STATE_WOW and the data packets are allowed until WOW
> suspend operation starts.
>
> * Similarly, wow resume is NOT allowed if WOW suspend is in progress.
>
> Signed-off-by: Raja Mani <[email protected]>
> ---
> V2 changes:
>
> * Separate wow sub state introduced in V1 is removed and common
> states (ATH6KL_STATE_SUSPENDING and ATH6KL_STATE_RESUMING)
> are added in enum ath6kl_state.
>
> * netdev queue is stopped while dropping skb in ath6kl_data_tx()
> and resumed back in ath6kl_wow_resume().
[...]
> diff --git a/drivers/net/wireless/ath/ath6kl/sdio.c b/drivers/net/wireless/ath/ath6kl/sdio.c
> index 4febee7..402087e 100644
> --- a/drivers/net/wireless/ath/ath6kl/sdio.c
> +++ b/drivers/net/wireless/ath/ath6kl/sdio.c
> @@ -901,8 +901,12 @@ static int ath6kl_sdio_resume(struct ath6kl *ar)
>
> case ATH6KL_STATE_WOW:
> break;
> +
> case ATH6KL_STATE_SCHED_SCAN:
> break;
> +
> + default:
> + break;
> }
Don't add the default case, instead add all states explicitly. That way
it's easier to track state changes.
> --- a/drivers/net/wireless/ath/ath6kl/txrx.c
> +++ b/drivers/net/wireless/ath/ath6kl/txrx.c
> @@ -284,6 +284,9 @@ int ath6kl_control_tx(void *devt, struct sk_buff *skb,
> int status = 0;
> struct ath6kl_cookie *cookie = NULL;
>
> + if (ar->state == ATH6KL_STATE_WOW)
> + return -EACCES;
Should this have WARN_ON_ONCE()?
> @@ -359,6 +362,13 @@ int ath6kl_data_tx(struct sk_buff *skb, struct net_device *dev)
> return 0;
> }
>
> + if (WARN_ON_ONCE(ar->state != ATH6KL_STATE_ON)) {
> + set_bit(NETQ_STOPPED, &vif->flags);
> + netif_stop_queue(dev);
> + dev_kfree_skb(skb);
> + return 0;
> + }
Don't stop the queue here, dropping the packet is enough.
Kalle
On Monday 27 February 2012 08:09 PM, Kalle Valo wrote:
> On 02/09/2012 11:31 AM, [email protected] wrote:
>> From: Raja Mani<[email protected]>
>>
>> Below two scenarios are taken care in this patch which helped
>> to fix the firmware crash during wow suspend/resume.
>>
>> * TX operation (ctrl tx and data tx) has to be controlled based
>> on suspend state. i.e, with respect to WOW mode, control packets
>> are allowed to send from the host until the suspend state goes
>> ATH6KL_STATE_WOW and the data packets are allowed until WOW
>> suspend operation starts.
>>
>> * Similarly, wow resume is NOT allowed if WOW suspend is in progress.
>>
>> Signed-off-by: Raja Mani<[email protected]>
>> ---
>> V2 changes:
>>
>> * Separate wow sub state introduced in V1 is removed and common
>> states (ATH6KL_STATE_SUSPENDING and ATH6KL_STATE_RESUMING)
>> are added in enum ath6kl_state.
>>
>> * netdev queue is stopped while dropping skb in ath6kl_data_tx()
>> and resumed back in ath6kl_wow_resume().
>
> [...]
>
>> diff --git a/drivers/net/wireless/ath/ath6kl/sdio.c b/drivers/net/wireless/ath/ath6kl/sdio.c
>> index 4febee7..402087e 100644
>> --- a/drivers/net/wireless/ath/ath6kl/sdio.c
>> +++ b/drivers/net/wireless/ath/ath6kl/sdio.c
>> @@ -901,8 +901,12 @@ static int ath6kl_sdio_resume(struct ath6kl *ar)
>>
>> case ATH6KL_STATE_WOW:
>> break;
>> +
>> case ATH6KL_STATE_SCHED_SCAN:
>> break;
>> +
>> + default:
>> + break;
>> }
>
> Don't add the default case, instead add all states explicitly. That way
> it's easier to track state changes.
I don't see any value addition having CASE statement for each state
here. We are not doing any specific operation other than the state
ATH6KL_STATE_OFF and ATH6KL_STATE_CUTPOWER.
IMHO, cases can be added later if we want to do anything specific to the
state in future.
To fix sparse errors, i added default case here.
>
>> --- a/drivers/net/wireless/ath/ath6kl/txrx.c
>> +++ b/drivers/net/wireless/ath/ath6kl/txrx.c
>> @@ -284,6 +284,9 @@ int ath6kl_control_tx(void *devt, struct sk_buff *skb,
>> int status = 0;
>> struct ath6kl_cookie *cookie = NULL;
>>
>> + if (ar->state == ATH6KL_STATE_WOW)
>> + return -EACCES;
>
> Should this have WARN_ON_ONCE()?
I'll correct this in V3.
>
>> @@ -359,6 +362,13 @@ int ath6kl_data_tx(struct sk_buff *skb, struct net_device *dev)
>> return 0;
>> }
>>
>> + if (WARN_ON_ONCE(ar->state != ATH6KL_STATE_ON)) {
>> + set_bit(NETQ_STOPPED,&vif->flags);
>> + netif_stop_queue(dev);
>> + dev_kfree_skb(skb);
>> + return 0;
>> + }
>
> Don't stop the queue here, dropping the packet is enough.
As per your comments in V1, i added this change.
Do you want me to change it again ?
This is what i received from you.
"Don't think about dropping packets, instead try to make sure that the
packet flow is _stopped_. Dropping packets should be the last resort,
instead we need to go to the source of the packets. For data packets
this means that we need to stop netdev queues. For control packets we
need to make sure that cfg80211 ops in cfg80211.c don't issue any new
wmi commands. And maybe we should also consider debugfs and wmi events
as they can also issue new wmi commands."
>
> Kalle
On 02/28/2012 07:47 AM, Raja Mani wrote:
> On Monday 27 February 2012 08:09 PM, Kalle Valo wrote:
>> On 02/09/2012 11:31 AM, [email protected] wrote:
>>
>>> --- a/drivers/net/wireless/ath/ath6kl/sdio.c
>>> +++ b/drivers/net/wireless/ath/ath6kl/sdio.c
>>> @@ -901,8 +901,12 @@ static int ath6kl_sdio_resume(struct ath6kl *ar)
>>>
>>> case ATH6KL_STATE_WOW:
>>> break;
>>> +
>>> case ATH6KL_STATE_SCHED_SCAN:
>>> break;
>>> +
>>> + default:
>>> + break;
>>> }
>>
>> Don't add the default case, instead add all states explicitly. That way
>> it's easier to track state changes.
>
> I don't see any value addition having CASE statement for each state
> here. We are not doing any specific operation other than the state
> ATH6KL_STATE_OFF and ATH6KL_STATE_CUTPOWER.
>
> IMHO, cases can be added later if we want to do anything specific to the
> state in future.
>
> To fix sparse errors, i added default case here.
The idea is to force the person adding a new state to think if something
special needs to handled for the state. If we have a default case it's
easy to miss it as compiler doesn't warn anything.
>>> @@ -359,6 +362,13 @@ int ath6kl_data_tx(struct sk_buff *skb, struct net_device *dev)
>>> return 0;
>>> }
>>>
>>> + if (WARN_ON_ONCE(ar->state != ATH6KL_STATE_ON)) {
>>> + set_bit(NETQ_STOPPED,&vif->flags);
>>> + netif_stop_queue(dev);
>>> + dev_kfree_skb(skb);
>>> + return 0;
>>> + }
>>
>> Don't stop the queue here, dropping the packet is enough.
>
> As per your comments in V1, i added this change.
> Do you want me to change it again ?
>
> This is what i received from you.
>
> "Don't think about dropping packets, instead try to make sure that the
> packet flow is _stopped_. Dropping packets should be the last resort,
> instead we need to go to the source of the packets. For data packets
> this means that we need to stop netdev queues. For control packets we
> need to make sure that cfg80211 ops in cfg80211.c don't issue any new
> wmi commands. And maybe we should also consider debugfs and wmi events
> as they can also issue new wmi commands."
Sorry for writing in a confusing way, but I didn't mean this. I meant
stopping the queue from suspend path, not in the data path. The
WARN_ON_ONCE() should not happen so dropping packets should be safe.
It's better to stop and stop the queue from one place (ie suspend/resume
handlers) than doing it also here. It doesn't gain anything.
Kalle