On 2017-06-07 23:57, Johannes Berg wrote:
On Sun, 2017-06-04 at 15:11 +0200, Erik Stromdahl wrote:
The softirq count is not always incremented during driver
operation. This is the case for usb and sdio network
drivers.
I'm pretty sure the warning is correct, and we do rely on having
local_bh_disable(), otherwise we may end up taking a soft-IRQ and I
believe there are some things that could get messed up in that case.
Ok, I will make sure to increment the softirq counter before calling
ieee80211_rx then.
So - I think the warning is there for a reason, and drivers should just
local_bh_disable() before calling into that. What's wrong with that?
I guess there is nothing wrong with that, it's just that ath10k does not
call local_bh_disable anywhere in the code.
I guess it is relying on lower layers (pcie?) to do that.
When introducing sdio and usb support these calls will have to be added
explicitly in ath10k.
johannes
