2016-03-09 07:27:40

by Hui Wang

[permalink] [raw]
Subject: [PATCH] brcmfmac: Remove waitqueue_active check

We met a problem of pm_suspend when repeated closing/opening the lid
on a Lenovo laptop (1/20 reproduce rate), below is the log:

[ 199.735876] PM: Entering mem sleep
[ 199.750516] e1000e: EEE TX LPI TIMER: 00000011
[ 199.856638] Trying to free nonexistent resource <000000000000d000-000000000000d0ff>
[ 201.753566] brcmfmac: brcmf_pcie_suspend: Timeout on response for entering D3 substate
[ 201.753581] pci_legacy_suspend(): brcmf_pcie_suspend+0x0/0x1f0 [brcmfmac] returns -5
[ 201.753585] dpm_run_callback(): pci_pm_suspend+0x0/0x160 returns -5
[ 201.753589] PM: Device 0000:04:00.0 failed to suspend async: error -5

Through debugging, we found when problem happens, it is not the device
fails to enter D3, but the signal D3_ACK comes too early to pass the
waitqueue_active() check.

Just like this:
brcmf_pcie_send_mb_data(devinfo, BRCMF_H2D_HOST_D3_INFORM);
// signal is triggered here
wait_event_timeout(devinfo->mbdata_resp_wait, devinfo->mbdata_completed,
BRCMF_PCIE_MBDATA_TIMEOUT);

So far I think it is safe to remove waitqueue_active check since there
is only one place to trigger this signal (sending
BRCMF_H2D_HOST_D3_INFORM). And it is not a problem calling wake_up
event earlier than calling wait_event.

Cc: Brett Rudley <[email protected]>
Cc: Hante Meuleman <[email protected]>
Cc: Franky (Zhenhui) Lin <[email protected]>
Cc: Pieter-Paul Giesberts <[email protected]>
Cc: Arend van Spriel <[email protected]>
Signed-off-by: Hui Wang <[email protected]>
---
drivers/net/wireless/broadcom/brcm80211/brcmfmac/pcie.c | 6 ++----
1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/pcie.c b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/pcie.c
index 0480b70..5f12ff3 100644
--- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/pcie.c
+++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/pcie.c
@@ -675,10 +675,8 @@ static void brcmf_pcie_handle_mb_data(struct brcmf_pciedev_info *devinfo)
brcmf_dbg(PCIE, "D2H_MB_DATA: DEEP SLEEP EXIT\n");
if (dtoh_mb_data & BRCMF_D2H_DEV_D3_ACK) {
brcmf_dbg(PCIE, "D2H_MB_DATA: D3 ACK\n");
- if (waitqueue_active(&devinfo->mbdata_resp_wait)) {
- devinfo->mbdata_completed = true;
- wake_up(&devinfo->mbdata_resp_wait);
- }
+ devinfo->mbdata_completed = true;
+ wake_up(&devinfo->mbdata_resp_wait);
}
}

--
1.9.1



2016-03-10 12:58:23

by Kalle Valo

[permalink] [raw]
Subject: Re: brcmfmac: Remove waitqueue_active check


> We met a problem of pm_suspend when repeated closing/opening the lid
> on a Lenovo laptop (1/20 reproduce rate), below is the log:
>
> [ 199.735876] PM: Entering mem sleep
> [ 199.750516] e1000e: EEE TX LPI TIMER: 00000011
> [ 199.856638] Trying to free nonexistent resource <000000000000d000-000000000000d0ff>
> [ 201.753566] brcmfmac: brcmf_pcie_suspend: Timeout on response for entering D3 substate
> [ 201.753581] pci_legacy_suspend(): brcmf_pcie_suspend+0x0/0x1f0 [brcmfmac] returns -5
> [ 201.753585] dpm_run_callback(): pci_pm_suspend+0x0/0x160 returns -5
> [ 201.753589] PM: Device 0000:04:00.0 failed to suspend async: error -5
>
> Through debugging, we found when problem happens, it is not the device
> fails to enter D3, but the signal D3_ACK comes too early to pass the
> waitqueue_active() check.
>
> Just like this:
> brcmf_pcie_send_mb_data(devinfo, BRCMF_H2D_HOST_D3_INFORM);
> // signal is triggered here
> wait_event_timeout(devinfo->mbdata_resp_wait, devinfo->mbdata_completed,
> BRCMF_PCIE_MBDATA_TIMEOUT);
>
> So far I think it is safe to remove waitqueue_active check since there
> is only one place to trigger this signal (sending
> BRCMF_H2D_HOST_D3_INFORM). And it is not a problem calling wake_up
> event earlier than calling wait_event.
>
> Cc: Brett Rudley <[email protected]>
> Cc: Hante Meuleman <[email protected]>
> Cc: Franky (Zhenhui) Lin <[email protected]>
> Cc: Pieter-Paul Giesberts <[email protected]>
> Cc: Arend van Spriel <[email protected]>
> Signed-off-by: Hui Wang <[email protected]>

Thanks, applied to wireless-drivers-next.git.

Kalle Valo

2016-03-09 09:14:52

by Hante Meuleman

[permalink] [raw]
Subject: RE: [PATCH] brcmfmac: Remove waitqueue_active check

Hi Hui,

Excellent find. Looks like a perfect solution to me.

Regards,
Hante

-----Original Message-----
From: Hui Wang [mailto:[email protected]]
Sent: Wednesday, March 09, 2016 8:25 AM
To: [email protected]; [email protected]; [email protected];
[email protected]; [email protected]
Cc: [email protected]; [email protected];
[email protected]
Subject: [PATCH] brcmfmac: Remove waitqueue_active check

We met a problem of pm_suspend when repeated closing/opening the lid
on a Lenovo laptop (1/20 reproduce rate), below is the log:

[ 199.735876] PM: Entering mem sleep
[ 199.750516] e1000e: EEE TX LPI TIMER: 00000011
[ 199.856638] Trying to free nonexistent resource
<000000000000d000-000000000000d0ff>
[ 201.753566] brcmfmac: brcmf_pcie_suspend: Timeout on response for
entering D3 substate
[ 201.753581] pci_legacy_suspend(): brcmf_pcie_suspend+0x0/0x1f0
[brcmfmac] returns -5
[ 201.753585] dpm_run_callback(): pci_pm_suspend+0x0/0x160 returns -5
[ 201.753589] PM: Device 0000:04:00.0 failed to suspend async: error -5

Through debugging, we found when problem happens, it is not the device
fails to enter D3, but the signal D3_ACK comes too early to pass the
waitqueue_active() check.

Just like this:
brcmf_pcie_send_mb_data(devinfo, BRCMF_H2D_HOST_D3_INFORM);
// signal is triggered here
wait_event_timeout(devinfo->mbdata_resp_wait, devinfo->mbdata_completed,
BRCMF_PCIE_MBDATA_TIMEOUT);

So far I think it is safe to remove waitqueue_active check since there
is only one place to trigger this signal (sending
BRCMF_H2D_HOST_D3_INFORM). And it is not a problem calling wake_up
event earlier than calling wait_event.

Cc: Brett Rudley <[email protected]>
Cc: Hante Meuleman <[email protected]>
Cc: Franky (Zhenhui) Lin <[email protected]>
Cc: Pieter-Paul Giesberts <[email protected]>
Cc: Arend van Spriel <[email protected]>
Signed-off-by: Hui Wang <[email protected]>
---
drivers/net/wireless/broadcom/brcm80211/brcmfmac/pcie.c | 6 ++----
1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/pcie.c
b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/pcie.c
index 0480b70..5f12ff3 100644
--- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/pcie.c
+++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/pcie.c
@@ -675,10 +675,8 @@ static void brcmf_pcie_handle_mb_data(struct
brcmf_pciedev_info *devinfo)
brcmf_dbg(PCIE, "D2H_MB_DATA: DEEP SLEEP EXIT\n");
if (dtoh_mb_data & BRCMF_D2H_DEV_D3_ACK) {
brcmf_dbg(PCIE, "D2H_MB_DATA: D3 ACK\n");
- if (waitqueue_active(&devinfo->mbdata_resp_wait)) {
- devinfo->mbdata_completed = true;
- wake_up(&devinfo->mbdata_resp_wait);
- }
+ devinfo->mbdata_completed = true;
+ wake_up(&devinfo->mbdata_resp_wait);
}
}

--
1.9.1