On Tue, 22 Jan 2019 04:41:16 +0100, Sameer Pujar wrote:
On 1/22/2019 2:58 AM, Takashi Iwai wrote:
On Mon, 21 Jan 2019 18:41:36 +0100, Sameer Pujar wrote:
Kernel panic is seen during device boot. It appears that before the probe completes, runtime callbacks happen and the device setup is not done yet. This could be initiated from framework through exposed callbacks. This issue can be fixed by having a flag to indicate completion of device probe. Hence 'probed' flag is introduced to notify completion of probe and runtime callbacks can check this flag before doing any device access.
Such a fix should be rather folded into the previous, especially if you know it's already broken :)
Ok, I will meld this with previous commit.
And, IMO, it's better to put such a check into runtime_idle callback instead of doing in each (runtime_)suspend/resume callback.
Check in runtime_idle would take care of suspend path. But how the check in runtime_resume can be avoided? Kernel panic happened during runtime_resume().
Hm, right, we need the check in both places, since runtime PM is called from the system PM, too.
BTW, I guess we can check chip->running instead of adding a new flag.
thanks,
Takashi