2017-02-24 13:10:45

by Rafał Miłecki

Subject: [PATCH 4.12] brcmfmac: put chip into passive mode (when attaching) just once

From: Rafał Miłecki <[email protected]>

It avoids some unnecessary work. Most likely there wasn't much sense in
doing this before bus reset anyway, as reset is supposed to put chip
into default state. In PCIe code (only bus implementing reset) we e.g.
use watchdog to perform a chip "reboot".

Signed-off-by: Rafał Miłecki <[email protected]>
---
drivers/net/wireless/broadcom/brcm80211/brcmfmac/chip.c | 10 ++++------
1 file changed, 4 insertions(+), 6 deletions(-)

diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/chip.c b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/chip.c
index 05f22ff81d60..670f2f50f9e5 100644
--- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/chip.c
+++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/chip.c
@@ -967,16 +967,14 @@ static int brcmf_chip_recognition(struct brcmf_chip_priv *ci)
if (ret)
return ret;

- /* assure chip is passive for core access */
- brcmf_chip_set_passive(&ci->pub);
-
/* Call bus specific reset function now. Cores have been determined
* but further access may require a chip specific reset at this point.
*/
- if (ci->ops->reset) {
+ if (ci->ops->reset)
ci->ops->reset(ci->ctx, &ci->pub);
- brcmf_chip_set_passive(&ci->pub);
- }
+
+ /* assure chip is passive for core access */
+ brcmf_chip_set_passive(&ci->pub);

return brcmf_chip_get_raminfo(ci);
}
--
2.11.0

2017-02-24 13:31:49

by Arend Van Spriel

Subject: Re: [PATCH 4.12] brcmfmac: put chip into passive mode (when attaching) just once

On 24-2-2017 14:10, Rafał Miłecki wrote:
> From: Rafał Miłecki <[email protected]>
>
> It avoids some unnecessary work. Most likely there wasn't much sense in
> doing this before bus reset anyway, as reset is supposed to put chip
> into default state. In PCIe code (only bus implementing reset) we e.g.
> use watchdog to perform a chip "reboot".

Sorry, but removing the fisrt passive call will cause chip lockups for
sure on certain systems. We learned the hard way :-p

Regards,
Arend

> Signed-off-by: Rafał Miłecki <[email protected]>
> ---
> drivers/net/wireless/broadcom/brcm80211/brcmfmac/chip.c | 10 ++++------
> 1 file changed, 4 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/chip.c b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/chip.c
> index 05f22ff81d60..670f2f50f9e5 100644
> --- a/drivers/net/wireless/broadcom/brcm80211/brcmfmac/chip.c
> +++ b/drivers/net/wireless/broadcom/brcm80211/brcmfmac/chip.c
> @@ -967,16 +967,14 @@ static int brcmf_chip_recognition(struct brcmf_chip_priv *ci)
> if (ret)
> return ret;
>
> - /* assure chip is passive for core access */
> - brcmf_chip_set_passive(&ci->pub);
> -
> /* Call bus specific reset function now. Cores have been determined
> * but further access may require a chip specific reset at this point.
> */
> - if (ci->ops->reset) {
> + if (ci->ops->reset)
> ci->ops->reset(ci->ctx, &ci->pub);
> - brcmf_chip_set_passive(&ci->pub);
> - }
> +
> + /* assure chip is passive for core access */
> + brcmf_chip_set_passive(&ci->pub);
>
> return brcmf_chip_get_raminfo(ci);
> }
>

2017-02-24 13:48:34

by Rafał Miłecki

Subject: Re: [PATCH 4.12] brcmfmac: put chip into passive mode (when attaching) just once

On 2017-02-24 14:31, Arend Van Spriel wrote:
> On 24-2-2017 14:10, Rafał Miłecki wrote:
>> From: Rafał Miłecki <[email protected]>
>>
>> It avoids some unnecessary work. Most likely there wasn't much sense
>> in
>> doing this before bus reset anyway, as reset is supposed to put chip
>> into default state. In PCIe code (only bus implementing reset) we e.g.
>> use watchdog to perform a chip "reboot".
>
> Sorry, but removing the fisrt passive call will cause chip lockups for
> sure on certain systems. We learned the hard way :-p

OK, acknowledged! ;) I just tested it on my BCM43602 doing warm reboots
(they
were causing problems earlier) and it worked fine.

It also seems I don't use SDK recent enough as my reference :)

Kalle please drop this patch.